2010

Lippincott Williams & Wilkins

Philadelphia
530 Walnut Street, Philadelphia, PA 19106 USA
978-1-60831-115-6

Acquisitions Editor: Charles W. Mitchell

Product Manager: Ryan Shaw
Vendor Manager: Bridgett Dougherty
Senior Manufacturing Manager: Benjamin Rivera
Senior Marketing Manager: Angela Panetta
Design Coordinator: Doug Smock
Production Service: Macmillan
© 2010 by LIPPINCOTT WILLIAMS & WILKINS, a WOLTERS KLUWER
business
530 Walnut Street
Philadelphia, PA 19106 USA
LWW.com
All rights reserved. This book is protected by copyright. No part of this
book may be reproduced in any form by any means, including
photocopying, or utilized by any information storage and retrieval
system without written permission from the copyright owner, except for
brief quotations embodied in critical articles and reviews. Materials
appearing in this book prepared by individuals as part of their official
duties as U.S. government employees are not covered by the above-
mentioned copyright.
Printed in China
Library of Congress Cataloging-in-Publication Data
Hashemi, Ray H.
MRI: the basics/Ray Hashman Hashemi, William G. Bradley Jr.,
Christopher J. Lisanti. — 3rd ed.
p.; cm.
Includes bibliographical references and index.
ISBN 978-1-60831-115-6
1. Magnetic resonance imaging. I. Bradley, William G. II. Lisanti,
Christopher J. III. Title. [DNLM: 1. Magnetic Resonance Imaging—
Examination Questions. 2. Magnetic Resonance Imaging— Problems and
Exercises. 3. Mathematics—Examination Questions. 4. Mathematics—
Problems and Exercises. 5. Physics—Examination Questions. 6. Physics—
Problems and Exercises. WN 18.2 H348m 2010]
RC78.7.N83H44 2010
616.07'548—dc22
2010001237
Care has been taken to confirm the accuracy of the information
presented and to describe generally accepted practices. However, the
authors, editors, and publisher are not responsible for errors or
omissions or for any consequences from application of the information
in this book and make no warranty, expressed or implied, with respect
to the currency, completeness, or accuracy of the contents of the
publication. Application of the information in a particular situation
remains the professional responsibility of the practitioner.
The authors, editors, and publisher have exerted every effort to ensure
that drug selection and dosage set forth in this text are in accordance
with current recommendations and practice at the time of publication.
However, in view of ongoing research, changes in government
regulations, and the constant flow of information relating to drug
therapy and drug reactions, the reader is urged to check the package
insert for each drug for any change in indications and dosage and for
added warnings and precautions. This is particularly important when
the recommended agent is a new or infrequently employed drug.
Some drugs and medical devices presented in the publication have Food
and Drug Administration (FDA) clearance for limited use in restricted
research settings. It is the responsibility of the health care providers to
ascertain the FDA status of each drug or device planned for use in their
clinical practice.
To purchase additional copies of this book, call our customer service
department at (800) 638-3030 or fax orders to (301) 223-2320.
International customers should call (301) 223-2300.
Visit Lippincott Williams & Wilkins on the Internet: at LWW.com.
Lippincott Williams & Wilkins customer service representatives are
available from 8:30 am to 6 pm, EST.
10 9 8 7 6 5 4 3 2 1
Authors
Ray Hashman Hashemi M.D., Ph.D.
Director and CEO
Advanced Imaging Center, Inc.
Lancaster, California

William G. Bradley Jr M.D., Ph.D., F.A.C.R.

Professor and Chairman
Department of Radiology
University of California, San Diego
San Diego, California

Christopher J. Lisanti M.D., Col. (ret) USAF, MC, SFS

Chief, Body MRI Department of Radiology Brooke Army Medical Center
Fort Sam Houston, Texas
Dedication

To my wonderful wife, Heidi Hame, DDS, MS, for her relentless

support and love over the past fifteen years; our son Tristen for
bringing more joy to our lives; and our parents for all their caring
and generosity.
R H H

To my MRI fellows over the last quarter century who raised many of
the questions answered in this book. At UCSD I now go through “The
Basics” with the MRI fellows from cover to cover over the course of
the year.
W G B

Soli Deo Gloria.

C J L
Preface to the First Edition

“Things should be made as simple as possible—

but no simpler.”
—ALBERT EINSTEIN

MRI has been called “the most important development in medical

diagnosis since the discovery of the x-ray” 100 years ago. It has become
one of the major new tools of radiology, now being applied to virtually
every part of the body. So, one might ask, if this MRI is so wonderful,
why are so many radiologists reluctant to “get into” it? In a word:
“physics.” The physics of MRI can be truly terrifying, particularly for
those attempting to enter the explanation halfway through without a
proper foundation in the basics. And without a proper understanding of
the basics, any MR clinician is merely “faking it,” not truly
understanding the physical basis for the signal changes in the image.
MRI: The Basics attempts to rectify that situation.
In this book we have attempted to present this complicated topic in a
readable, understandable, and even entertaining fashion without
sacrificing the fundamental concepts. The reader will find a
comprehensive coverage of MRI physics, from basic principles to more
advanced topics such as MR angiography and fast scanning techniques.
Some of the latest MR techniques made possible with high performance
gradients, e.g., echo planar imaging, are also discussed. Because most
of the chapters arose from lectures given by the first author to
radiology residents, the language used through most of the text is of an
informal or conversational nature that makes it easy to follow.
While attempting to be thorough, this book does not get bogged down
in the minor details. An introductory math chapter is designed to
introduce the reader to the most fundamental mathematics used in MRI
(no knowledge of calculus is required!). Considerable attention is paid
to the process of image creation, including the concepts of gradients,
signal/image processing, and k-space. One of the distinguishing
features of this book is its introduction and treatment of signal/image
processing. There are two chapters dedicated to image creation, two to
k-space, one to Fourier transform, and one to signal processing, as well
as several chapters on fast scanning (fast spin echo, gradient echo, fast
gradient echo, and echo planar imaging). There are also chapters that
deal in depth with flow and MR angiography as well as MRI artifacts.
More than 400 illustrations provide the reader with a clear visual tool to
follow the text. The key points of each chapter are summarized at the
end of that chapter. In addition, there is a set of problem solving and
multiple choice questions at the end of each chapter (with answers at
the end of the book) to test the reader's knowledge of that chapter.
The material presented in smaller print may be skipped because it is
intended for the mathematically oriented reader.
This book is intended primarily for radiologists and radiology residents
and fellows as well as radiologic technologists. However, other
physicians, medical students, scientists, and professionals dealing with
MRI could also benefit from it. It is intended to provide the shortest
learning path from the basics through the applications, avoiding the
extraneous. This book can be used by radiology residents preparing for

the physics portion of the American Board of Radiology exam and by MR

technologists preparing for their MR certification exam.
In short, you can find in this book almost everything you always wanted
to know about MR physics but were afraid to ask. In addition, this book
may not only be read cover to cover as a textbook to learn about all
aspects of MR basics, but also may be used as a reference for the
fundamentals and advanced technological breakthroughs in MRI. We
hope you'll enjoy reading it as much as we did writing it.
R H H

W G B
Preface

The first two editions of MRI: The Basics were well received, and we
trust enhanced MRI knowledge and practical application. This current
edition of the The Basics spans the gamut from basic physics to
multiuse MR options to specific applications. Although MRI is a maturing
technology, it, nevertheless, continues to challenge radiologists,
residents, and technologists. More improvements and new features are
constantly being developed and applied producing diagnostic images
faster, with higher image quality and reproducibility. This power,
however, invariably gives the MR user more options, and these various
options must be sorted and thought through in order to give optimal
results.
In response to advancements in MRI, we have updated the book and
added completely new chapters addressing parallel imaging, cardiac
MRI, and MR Spectroscopy. These areas are used with increasing
regularity in MR imaging, and we hope that you will glean new insight
into these features and applications in order to enhance your practice.
Again, we are pleased to present to you the third edition of MRI: The
Basics, and hope that it will be a trusty companion during your MRI
journey.
Ray Hashman Hashemi MD, PhD

William G. Bradley Jr MD, PhD, FACR

Col. (ret) Christopher J. Lisanti MD

Acknowledgments

We would like to thank Robert Mulkern (Brigham and Women's Hospital)

for helpful comments on the equations in Chapter 17. This generated an
e-mail debate on the formula for signal-to-noise, which went on for a
month and included MR giants such as Mark Haacke (Wayne State), John
Mugler (UVa), Felix Wehrli (Penn), Gary Fullerton (UColo), and Mark
Bydder (UCSD—Graeme's son and future giant).
FRONT OF BOOK

[+]
Authors
-
Dedication
-
Preface to the First Edition
-
Preface
-
Acknowledgments
↑
TABLE OF CONTENTS

[-]
Part I - Basic Concepts
[+]
1 - Introductory Math
[+]
2 - Basic Principles of MRI
[+]
3 - Radio Frequency Pulse
[+]
4 - T1, T2, and T2*
[+]
5 - TR, TE, and Tissue Contrast
[+]
6 - Tissue Contrast: Some Clinical Applications
[+]
7 - Pulse Sequences: Part I (Saturation, Partial Saturation, Inversion
Recovery)
[+]
8 - Pulse Sequences Part II (Spin Echo)
[+]
9 - Fourier Transform
[+]
10 - Image Construction: Part I (Slice Selection)
[+]
11 - Image Construction: Part II (Spatial Encoding)
[+]
12 - Signal Processing
[+]
13 - Data Space
[+]
14 - Pulse Sequence Diagram
[+]
15 - Field of View
[+]
16 - k-Space: The Final Frontier!
[+]
17 - Scan Parameters and Image Optimization
[+]
 18 - Artifacts in MRI
[-]
Part II - Fast Scanning
[+]
19 - Fast Spin Echo
[+]
20 - Gradient Echo: Part I (Basic Principles)
[+]
21 - Gradient Echo: Part II (Fast Scanning Techniques)
[+]
22 - Echo Planar Imaging
[+]
23 - New Scanning Features
[+]
24 - Parallel Imaging
[+]
25 - Tissue Suppression Techniques
[+]
26 - Flow Phenomena
[+]
27 - MR Angiography
[+]
28 - Cardiac MRI
[+]
29 - MR Spectroscopy in the Brain
[+]
30 - High Performance Gradients
[+]
31 - The Many Combinations of MRI
↑
BACK OF BOOK

-
Appendix A: Suggested Readings
[+]
Appendix B - Answers
[+]
Appendix C
[+]
Index
1
Introductory Math

Introduction
In this chapter, we review some of the basic mathematical concepts
that are used in MR imaging. We don't want to scare you away so we
keep things as simple as possible. An understanding of these basic
concepts will help the reader a great deal to comprehend the subtleties
of MR imaging and obtain the necessary tools for manipulating the scan
parameters to improve the quality of the images.
It is not that important to memorize these mathematical formulas;
what's crucial is the understanding of the concepts behind these
formulas. In this chapter, we hope to emphasize the most important
mathematical concepts of MRI physics.

Sinusoidals
Consider a right triangle (having a right angle) with sides a and b and
hypotenuse c and angle x formed by a and c (Fig. 1-1). We can define
sin x (read sine of x), cos x (read cosine of x), tan x (read tangent of x),
cotan x (read co-tangent of x), and arctan x (read arc-tangent of x) in
terms of a, b, and c:
Figure 1-1. A rigt triangle with sides a and b and hypotenuse c
angle x formed by a and c.

Table 1-1

0 π/6 π/4 π/3 π/2 π

x 0° 30° 45° 60° 90° 180°

sin x 0 1/2 ( )/2 ( )/2 1 0

cos x 1 ( ( )/2 1/2 0 -1

)/2
 tan 0 1/( 1 ( ) ∞ 0
x )

sin x = b/c
cos x = a/c
tan x = sin x/cos x = a/b
arctan b/a = arctan(tan x) = x (Eqn. 1-1)
The variable x can be represented in degrees, that is, 45°, 90°, and
180°, or it can be represented in radians, that is, π/4, π/2, and π,
where π = 180°. Table 1-1 shows x versus. sin x, cos x, and tan x, where
≅ 1.4 so /2 ≅ 0.7 and ≅ 1.7 so /2 ≅ 0.85.
Let's plot x versus sin x (Fig. 1-2). This is called a sinusoidal function.
What about cos x? (Fig. 1-3). Let's now draw cos x and sin x on a single
graph (Fig. 1-4). We can appreciate the symmetry between sin x and
cos x. The difference between the two functions is that sin x is shifted
to the right of cos x by 90°. Later, when we talk about phase and phase
shifts, this mathematical concept will become more important. We can
think of sin x as being cos x with a phase difference of 90°.

Figure 1-2. Graph of sin (x).

 Figure 1-3. Graph of cos (x).

Figure 1-4. Sin (x) and cos (x) plotted on the same graph.

Now, let's go back to Figure 1-1. What is c in terms of a, b? According to

the Pythagorean theorem:
By Equation 1-1

So (sin x)2 + (cos x)2 = 1 If we go back to our graph of sin x and cos x
(Fig. 1-4), we can see graphically that because of the phase difference
between the cos x and sin x, the sum of their squares will always equal
1. Another way of looking at sine and cosine is to consider a circle with
a radius of 1 (Fig. 1-5). To understand this concept, it is necessary to
bring up the concepts of vectors, imaginary numbers, and exponentials.
Vector. We'll designate a vector by using a letter such as v with an
arrow above it ( ). This concept will become important later on in the
understanding of resonance of spins and dephasing. A vector is a
mathematical entity that has both a magnitude and a direction. For
example, speed is not a vector—it only has magnitude.

Velocity, however, is a vector—it has both magnitude and direction.

Another example of a vector is force that describes a magnitude
(weight) and a direction (direction where the force is applied).
 Figure 1-5. A vector v with magnitude 1 and angle x to the
horizontal. cos x and sin x are the horizontal and vertical
components of this vector, respectively.

The vector that we've drawn in the circle (Fig. 1-5) has a magnitude of
1. The angle between the vector and the horizontal axis is denoted x. If
we draw perpendiculars from the vector both horizontally and
vertically, we'll get two components of the vector:
a. The horizontal component of the vector would correspond to cos x.
 (Remember that the ratio a/c in Figure 1-1 is cos x.)
b. The vertical component of the vector would correspond to sin x.
(Remember that the ratio b/c in Figure 1-1 is sin x.)
Imaginary Numbers. A positive number n2 has two square roots, + n and
− n. For example,

It is impossible to square a real number and have a negative product.

Therefore, we shall make up an entity and call the number

an imaginary number. Any of the following would be imaginary
numbers:

Imaginary numbers can be manipulated in the following way:

Any imaginary number can be written as a positive number times ( ).

The expression ( ) is designated by the letter “i”. (Mathematicians
use the symbol i to denote an imaginary number, whereas engineers use
the symbol “j” in lieu of i because i is reserved to symbolize electric
current!) The symbol i is then known as the imaginary unit. In other
words, i × i = −1.
Example

So i is an imaginary number. It doesn't exist. When you take a square

root of a number, it has to be a positive number. However, in this case,
if we multiply i by i, we get (−1). Therefore, i is an imaginary number
that doesn't exist.
Complex Numbers. A complex number is a number that has both a real
and an imaginary component: Complex = real + imaginary
Example
Let's say a complex number has two components: 2 and 3. The
imaginary component is multiplied by (i), where i = is the imaginary
unit. Then, c = (2) + i(3) If you draw this complex number on an x − y
plane (Fig. 1-6), the vector (2,3) illustrates the complex number 2 + 3i.
Usually you only care about the real part of a complex number, but it
makes life easier to deal with the complex number and carry all the
computations using complex numbers or vectors and then, at the end,
just keep the real part.
Magnitude and Angle. Sometimes, however, the imaginary part is also
helpful. For example, consider Figure 1-6. In this diagram, if we take

the tangent of the angle between the vector and the x-axis, we get tan
θ = 3/2 = imaginary/real In other words, the ratio of the imaginary part
to the real part gives us the tangent of the angle. The magnitude of
the vector (sometimes called the modulus) is given by the Pythagorean
theorem

Thus, vector magnitude

When you're dealing with a complex number, if you take the ratio of the
imaginary part to the real part, you get some sort of measure of the
angle of the vector. If you sum the squares of the imaginary and real
part, you get the magnitude of the vector (squared).

Figure 1-6. Representation of a complex number 2 + i 3 as a point

in a two-dimensional (Cartesian) coordinate system and the
relationship between a complex number, its angle θ, and its real
and imaginary components.

Function. A mathematical function, designated as f (x), is an entity that

varies with respect to a variable, x. For example, sin x is a function
that varies with respect to x in a sinusoidal manner, as we saw earlier in
the chapter.
Signal. A signal is a time-varying function, that is, something that varies
over time, usually millivolts versus time. If the x-axis is time, and the
y-axis is magnitude, then a signal is a waveform that varies in
magnitude with time.
In an electrical system, a signal is a timevarying current or voltage that
can be measured. In MRI, the signal is just a current or voltage that is
induced by an oscillating magnetic field. Some signals are periodic—
they repeat themselves—as the sine wave or cosine wave repeats itself.
Frequency, Period, and Cycle. Let's now introduce the concepts of
frequency and period. Every periodic function has a frequency, which
we will call f. If we measure the time interval between two peaks (or
where the signal crosses 0), this interval is called a period, and we'll
denote this by T. Now, frequency = 1/period = 1/T: f = 1/T A cycle in a
periodic function is any part of the function over one period. For
example, let's say that we have three complete cycles occurring in 1
sec (Fig. 1-7). In this case, 3 periods take 1 sec 3T = 1 sec T = 1/3 sec
Frequency = f = 1/T = 3 cycles per sec = 3 Hz The unit we use to
describe frequency is Hertz or Hz (for cycles per sec). There are 2π
radians in one cycle. That is to say,
f = frequency when we refer to linear frequency in cycles per sec
ω = frequency when we refer to angular frequency in radians per sec;
angular frequency (in radians/sec) = ω = (2π) × linear frequency (in Hz)
where π = 3.1415927 ≅ 3.14. In short,
 Figure 1-7. An example of a periodic signal spanning three cycles
in 1 sec. The period is thus one third of 1 sec (1/3 sec).

Example
If the vector in the circle makes 3 revolutions/sec, then the vector has
a frequency of 3 Hz. Therefore, ω = (2π)(3) = 6π = 18.85 radians/sec.
Sometimes a signal (such as a sine wave) is represented in the following
way: S(t) = sin (ωt) = sin (2πft) Here, we can say that the signal is a
sine wave with a frequency of ω. So, frequency = f = ω/2π.
Example
Draw the signal sin (ωt) versus time t assuming f = 1 Hz (i.e., T = 1 sec
or ω = 2π radians/sec). Thus, sin (ωt) 5 sin (2πt). This equation is
illustrated in Figure 1-8:
when t = 0, then ωt = 0, resulting in
sin ωt = sin 0 = 0;
when t = 1/4, then ωt = (2π) (1/4) = π/2,
resulting in sin ωt = sin π/2 = 1;
when t = 1/2, then ωt= (2π)(1/2) = π,
resulting in sin ωt = sin π = 0; etc.
Phase. Now, let's talk about phase. Consider two sine waves, with one
shifted slightly compared with the other (Fig. 1-9). The two sinusoids
have the same frequency—they oscillate at the same rate—but one of
them is shifted just a little from the other. Suppose they are shifted
apart from each other by a time interval = τ = 1 sec. Suppose further
that the period of 1 cycle = T = 4 sec (Fig. 1-9).

Figure 1-8. Graph of sin (2 πt).

T = 1 period = 360° in 4 sec

τ = 1 sec = 1/4 of the total time of 1 period
Thus, 1/4 of 360° = 90°. This measurement is the phase offset or
phase shift of the two sinusoidals.
Example
What is sin (x + 90°)?
Here we have the sine wave and a phase offset of 90°. Let's figure out
what this looks like on a circle diagram (Fig. 1-10). If x is the angle,
sine of vector x is the initial component perpendicular to vector. Take a
phase offset of 90°. This offset causes the original vector to be rotated
counterclockwise by 90°, as can be seen in Figure 1-10. Now, drop a
vertical perpendicular from this new vector. Because the vertical
component of a vector is equal to the sine of its angle, then the
vertical component of the new vector (with its new angle x + 90°) is
sine of (x + 90°). Now, by the law of congruent triangles, this new
vertical component (which is longer than the horizontal component in
Fig. 1-10) is equal to the longer horizontal component of the original
angle x, which is, in fact, cos x. In other words,

From this, we see that sine and cosine of any given vector have a phase
difference of 90°.

Figure 1-9. Two similar sinusoidal signals with a phase offset.

Exponentials
With exponential functions (ex), the letter e is the base of the natural
logarithm with a numerical value of e equals 2.7182818 ≅ 2.72 First,
let's consider the values of ex for various values of x:

for x = 0, e0 = 1(anything to the power of 0 = 1)

for x = 1, e1 = 2.72

for x = 2, e2 =(2.72)(2.72) = 7.4

 for x = ∞, e∞ = ∞

for x = −1, e−1 = 1/e = 1/2.72 = 0.37

for x = −2, e−2 = 1/e2 = 1/(2.72)(2.72) = 0.14

for x = −3, e−3 = 0.05

for x = −∞, e−∞ = 0

If we graph ex, we see that it is an exponentially growing function (Fig.

1-11). From this graph, we can see that the function ex grows
exponentially from −∞ to +∞.
 Figure 1-10. Two vectors with a phase difference of 90°. This
difference demonstrates the relationship between cos (x) and sin
(x + 90°).

Now let's consider a decaying function by drawing (e−x) (Fig. 1-12):

for x = 0, e0 = 1

for x = 1, e−1 = 0.37

for x = 2, e−2 = 0.14

 for x = 3, e−3 = 0.05

for x = ∞, e−∞ = 0

for x = −1, e−(−1) = e1 = 2.72

for x = −2, e−(−2) = e2 = 7.4

This represents an exponentially decaying function. Let's change x to t

and look at a time-varying function e−t (Fig. 1-13). This now becomes a
decaying function of time. Table 1-2 shows the values of e−t for various
values of t.
How about a graph of (1 − e−t)? for t = 0, (1 − e−t) = 1 −1 = 0 for t = 1, (1
− e−t) = 1 − 2.7−1 = 1 − 1/2.7 = 1 − 0.37 = 0.63 for t = 2, (1 − e−t) = 1 −
(2.7)−2 = 1 − 1/(2.7)(2.7) = 1 − 0.13 = 0.86 As t gets larger, e−t gets
much smaller, so that (1 − e−t) approaches 1. Therefore, a graph of 1 −
e−t (Fig. 1-14) is the reverse of the exponential curve for e−t (Fig. 1-13).
Time Constant or Decay Constant. In the graph of (1 − e−t), the value
gets close to 1 after about four to five time constants. Let's make the
exponential function a little more complicated. Consider e−t/τ

where τ is a time constant. So when t = τ, then e−t/τ = e−1 = 0.37 If we

draw a tangent at t = 0 (where the function equals 1 on the vertical
axis) along the exponential decay curve e−t/τ, the point of intersection
on the time line (i.e., the horizontal axis) is the time constant τ (Fig. 1-
15). At this point, the value of e−t/τ = e−1 = 0.37 ≅ 1/3. This means that
at the time of one time constant, we have about one third of the
original signal left. The other interesting thing about this decaying
function is that after two time constants (2τ), we are left with one
third of the signal that remained after one time constant, which is (e−2)
= e−1 × e−1 ≅ 1/3 × 1/3 = 1/9.
Figure 1-11. Graph of an exponentially growing function ex.

Figure 1-12. Graph of an exponentially decaying function e−x.

 Figure 1-13. Graph of an exponentially decaying function of time e
−t.

The nice thing about this exponentially decaying function is that you
can stop anywhere on the curve and the same equation will apply. No
matter how many time constants you extend out along the decay curve,
you still end up with a calculable percentage value of the original
signal.

Table 1-2

t 0 1 2 3 4 5 … ∞

e−t 1 .37 .14 .05 .02 .01 … 0

 1 − e−t 0 .63 .86 .95 .98 .99 … 1

Figure 1-14. Graph of an exponentially growing function of time 1

− e−t.
Figure 1-15. Graph of an exponentially decaying function e−t/τ,
where τ is the time constant (decay rate).

Figure 1-16. Graph of an exponentially growing function 1 − e−t/τ,

 where τ is the time constant (growth rate).

On the recovery curve (1 − e−t/τ), things are just the opposite (Fig. 1-
16). If you draw a tangent at the origin along the exponential recovery
curve (1 − e−t/τ), the point of intersection on the maximum line
corresponding to complete recovery occurs at one time constant (τ). At
this point, we have recovered 63% of the original signal: for t = τ, (1 − e
−t/τ) = 1 − e−1 ≅ 1 − 0.37 = 0.63 At two time constants (2τ), we will have

recovered 63% of the remaining signal, which comes to 86%.

Exponentially Decaying Sinusoidals

We've talked about a sine wave (sin ωt). We've also talked about an
exponential function (e−t/τ). What happens if we multiply these
functions together? (e−t/τ) · (sin ωt) The first function is a decaying
curve. The second function is a sinusoidal. If you multiply the two
together, you'll get a graph, as in Figure 1-17. The sinusoidal function
will have the same frequency but it's contained between the
“envelopes” of the decaying exponential curve and its mirror image,
causing the magnitude of the sine wave to decrease exponentially with
time.
 Figure 1-17. Graph of e−t/τ χ cos ωt. This is a sinusoidal with
angular frequency ω whose “envelope” is given by e−t/τ and −e−t/τ.

Sinc Function
There is another function that looks somewhat like Figure 1-17. It is
called a sinc function and is expressed as (sin t)/t:
Question: What is sinc t at time t = 0?
Answer: sinc (0) = sin (0)/0 = 0/0, which is indeterminate. However,
using the principles of differential equations and limits, it can be shown
that sinc (0) = 0/0 = 1 (refer to Question 1-3, at the end). From here
on, sinc (t) is an oscillating wave of t (Fig. 1-18). However, the
envelope of this wave is not an exponential function. This is what a
Radio Frequency (RF) pulse generally looks like. (The frequency or
Fourier transform [FT]

of this pulse is a rectangle. More about this later.)

Figure 1-18. Graph of sinc (t) = sin (t)/t.

Having been introduced to the concepts of vectors, imaginary numbers,

and exponentials, let's again go back to the vector in Figure 1-5 and
introduce a new equation called Euler's equation. (The importance of
this equation will become clear later when the concept of “spinning
protons” is addressed.)

Euler's Equation
This equation describes the vector that we saw in Figure 1-5 with an
angle = θ and a magnitude = 1. The symbol i is the imaginary unit .
Equation 1-5 expresses a complex exponential function of an imaginary
number (iθ) in terms of the sine and cosine functions of the angle θ.
Now consider the Euler's equation for the complex signal eiωt (using ωt
instead of θ):

This formula has a “real” part and an “imaginary” part. The real part
(cos ωt) is what we are interested in because it corresponds to the
measured signal. We use the imaginary part (sin ωt) because it makes
the mathematics simpler (believe it or not!). At the end, we ignore the
imaginary part and keep the real component corresponding to the
actual signal.
Each value of (eiωt) is a vector that spins around at an angular
frequency of ω (Fig. 1-19). It is important to understand the concept of
angular frequency because later, when we talk about proton precession,
we'll see that the principle of precessional frequency (which is an
angular frequency) is used frequently.

Logarithms
The logarithm (log) is sort of the inverse of an exponential, that is, log
(ex) = x

The base of log is usually 10. However, a logarithm can have any base.
Let's say that the log of a number y with base a is equal to x: logay = x
Then, taking the exponential of both sides, we obtain ax = y Thus, logay
= x ⇔ ax = y That is, if the log of y with base a equals x, this implies
that a to the power of x equals y. Example log2 8 = 3 ⇒ 23 = (2)(2)(2)
= 8 The log of a number x to base e is also denoted as the ln of that
number, or “the natural log” of x. logex = lnx (ln) is just the notation
for log to the base e (e is the base of the natural logarithm, e =
2.71…). ln e = loge e = 1 This means log of e to its own base is 1.

Figure 1-19. Representation of eiωt in terms of a rotating vector.

Properties of Exponentials
 Mathematical operations done with exponentials include (ex is e to the
power of x):

a. Multiplying exponentials (ex) · (ey) = ex+y

b. Negative exponentials (e−x) = 1/ex
c. Dividing exponentials ex/ey = ex−y
Derivation
Equation (c) is derived from the first two Equations (a) and (b) in the
following way: ex/ey = (ex)(e−y) derived from the statement about
negative exponentials; (ex) · (e−y) = ex−y derived from the statement
about multiplying exponentials. Therefore, ex/ey = (ex) (e−y) = ex−y.

Properties of Logarithms
a. The logarithm of the product of two numbers (A and B) is equal to
the sum of their logarithms. log (A · B) = log A + log B This holds for
any base. So ln (A · B) = ln A + ln B
b. The log of x to a power a is log xa = a log x This also holds for any
base, so logb xa = a logb x ln xa = a ln x

Example
Let's solve the equation: 1/2 = e−t/T2 for t. (This would be the
formula for finding the time t when a function [with a first-order
decay constant T2] decays to half of its initial value.)
1. Take ln of both sides: ln (1/2) = ln e −t/T2
2. Remember that 1/2 = 2−1, so ln (1/2) = ln (2−1) = − ln 2
3. Now from Equation (b) ln (e−t/T2) = −t/T2 ln (e) so −ln 2 = −t/T2 ln (e)
4. Remember that ln (e) = logee e = 1 (log of a number to its own base =
1), so −ln 2 = −t/T2 (1) or −ln 2 = −t/T2
5. Multiply both sides by −1: ln 2 = t/T2 or loge 2 = t/T2 or 0.693 = t/T2
or t = 0.693
This formula also calculates the half-time (or t1/2) in nuclear medicine.

Key Points
Understanding a few mathematical concepts will help
you a great deal in understanding MRI physics.
1 Four sinusoidal functions were discussed: sin x, cos
x, tan x, cotan x tan x = sin x/cos x cotan x = 1/tan
x
2 The angle (or arc) on which the above sinusoidal
wave is applied is defined as arcsin, arccos, arctan,
arccotan For example, arctan(tan x) = x, arcsin(sin
x) = x, etc.
3 A vector possesses magnitude and direction. Force,
for example, has magnitude (weight) and direction.
Another example of a vector is velocity, which has a
speed and a direction.
4 Imaginary numbers are represented by a real and
an imaginary component: c = a + ib where i is the
imaginary unit .
5 A function f of a variable x is designated f(x) and
represents variations in f as x is varied.
6 A signal is a function of time.
7 A periodic signal is a function of time that repeats
itself after a certain period T.
8 The (linear) frequency of a periodic signal is
defined as f = 1/T, where T is the period.
9 The angular frequency ω is defined as ω = 2πf.
10 A periodic signal over one period represents one
cycle.
11 An example of a periodic signal is cos ωt = cos
(2πft).
12 Phase represents the offset between two periodic
signals of the same frequency. For example, cos (ωt)
and cos (ωt + θ) have the same frequency (ω) but
are out of phase by θ.
13 The function et is an exponential function of time.
It is actually an exponentially growing function. It is
1 at t = 0 and ∞ at t = ∞. At t = 1, e1 = e =
2.7182818 ≅ 2.72.
14 The function e−t is also an exponential function of
time. It is an exponentially decaying function. It is 1
at t = 0 and 0 at t = ∞. At t = 1, e−1 = 0.37.
15 The function e−t/τ is an exponentially decaying
function with a time constant or decay constant τ.
The value of the signal at time constant τ is 37% (e−1
= 0.37) of its previous value. The signal is practically
zero after five time constants (e−5 ≅ 0).
16 The function eiωt is given by Euler's equation as
eiωt = cos ωt + i sin ωt which represents a vector of
radius 1 spinning at an angular frequency ω (in
radians/sec).
17 The sinc function is defined as sinc t = sin t/t
which is 1 at t = 0. An ideal RF pulse is a sinc wave
because its Fourier transform (as we will see later)
has a perfect rectangular shape.
18 A logarithm (base 10) of a variable y is
represented as log y and is related to an exponential
as if log y = x then 10x = y
19 The natural logarithm (base e) of y is designated
ln y. Therefore if ln y = x then ex = y Having
 understood the above mathematical concepts, the
reader can now read and understand the remainder
of this book with greater ease. As mentioned
previously, it is not that important to memorize the
formulas, but rather to understand the concepts
behind them.

Questions
1-1 Draw the following functions versus x (from − ∞ to + ∞)
(a) e−x cos x
(b) sin x/x ≡ sinc (x)

1-2 Prove the following equalities cos (x + y) = cos x · cos y + sin x ·

sin y and sin (x + y) = cos x · sin y + sin x · cos y Hint: Make use of
the Euler equation eix = cos x+i sin x and note that i × i = i2 = − 1

1-3 It can be shown that for certain functions

where f and g are functions of x and f' and g' are their derivatives,
that is, [ f' (x) = d/dx · f (x)]; “lim” denotes the “limit” as x
approaches 0.
Using the above fact, show that

Hint: d/dx (sin x) = cos x and d/dx (xn) = n · xn−1 (n = any integer)
 Figure 1-20. A-B

1-4
(a) What is the value of an exponential function e−t/T at one
time constant T?
 (b) How about at two time constants 2T?
(c) What is the ratio of (b) to (a)?

1-5 Take the exponentially decaying function f (t) = A · e−t/T as

shown in Figure 1-20A. Prove that if a tangent is drawn at point A, it
will cross the t-axis at t = T (T = decay constant).
Hint: The derivative d/dt (eat) = a = eat where a = a constant.

1-6 Solve the equation ex = 8 for x. Note: ln 2 = 0.693.

1-7 What is (a) sin (0°); (b) sin (30°); (c) sin (90°); (d) sin (180°); (e)
cos (0°); (f) cos (60°); (g) cos (90°); (h) cos (180°)?
2
Basic Principles of MRI

Introduction
In this chapter, we will discuss the basic principles behind the physics of
magnetic resonance imaging (MRI). Some of these principles are
explained using Newtonian physics, and some using quantum
mechanics, whichever can convey the message more clearly. Although
this might be confusing at times, it seems to be unavoidable. In any
case, we'll try to keep it straightforward.
Nuclear magnetic resonance (NMR) is a chemical analytical technique
that has been used for over 50 years. It is the basis for the imaging
technique we now call MRI. (The word nuclear had the false
connotation of the use of nuclear material; thus, it was discarded from
the MR lexicon, and “NMR tomography” was replaced by the phrase
magnetic resonance imaging [MRI].)

Table 2-1 The Electromagnetic Spectrum Illustrating the

Windows for Radio waves, Microwaves, Visible Light, and
X-Rays

Frequency Energy Wavelength

(Hz ) (eV) (m)

Gamma rays and

X-rays 1024 1010 10-16

1023 109 10-15

 1022 108 10-14

1021 107 10-13

106 (1 10-12 (1 pm
1020 MeV) )

1019 105 10-11

1018 104 10-10

103 (1
Ultraviolet 1017 keV) 10-9 (1 nm )

1016 102 10-8

Visible light 1015 101 10-7

100 (1
Infrared 1014 eV) 10-6 (1 µm)

1013 10-1 10-5

Microwaves 1012 10-2 10-4

1011 10-3 10-3 (1 mm )

1010 10-4 10-2 (1 cm)

109 (1 GHz) 10-5 10-1

 108 (100
MRI MHz ) 10-6 100 (1 m)

107 10-7 101

Radio waves 106 (1 MHz ) 10-8 102

105 10-9 103 (1 km)

104 10-10 104

103 (1 kHz ) 10-11 105

102 10-12 106

Electromagnetic Waves
To understand MRI, we first need to understand what an
electromagnetic wave is. Table 2-1 demonstrates the characteristics of
a variety of electromagnetic waves, including X-ray, visible light,
microwaves, and radio waves. All electromagnetic waves have certain
fundamental properties in common:

1. They all travel at the speed of light c = 3 × 108 m/sec in a vacuum.

2. By Maxwell's wave theory, they all have two components—an electric
field E and

a magnetic field B—that are perpendicular to each other (Fig. 2-1).

We will designate the sinusoidal wave, which is drawn in the plane of
the paper, the electrical field E. Perpendicular to it is another
sinusoidal wave, the magnetic field B. They are perpendicular to
each other and both are traveling at the speed of light (c). The
 electric and magnetic fields have the same frequency and are 90° out
of phase with each other. (This is because the change in the electric
field generates the magnetic field, and the change in the magnetic
field generates the electric field. For this reason, electromagnetic
waves are self-propagating once started and continue out to infinity.)

Figure 2-1. Two components of an electromagnetic wave, the

electric component E and the magnetic component B. These two
components are perpendicular to each other, are 90° out of
phase, and travel at the speed of light (c).

3. If we think in terms of vectors, the vectors B and E are perpendicular

to each other, and the propagation factor C is perpendicular to both
(Fig. 2-2). Both the electrical and magnetic components have the
same frequency ω. So what we get is a vector that is spinning
(oscillating) around a point at angular frequency ω. Remember, the
angular frequency ω is related to the linear frequency f:
ω = 2πf
4. We are interested in the magnetic field component—the electric field
component is undesirable because it generates heat.
Figure 2-2. The vector representation of B, E, and C.

Table 2-2

Wave
Frequency Energy Length
 1.7-3.6 × 1018 30-150
X-ray Hz keV 80-400 pm

Visible light
(violet) 7.5 × 1014 Hz 3.1 eV 400 nm

Visible light
(red) 4.3 × 1014 Hz 1.8 eV 700 nm

20-200
MRI 3-100 MHz MeV 3-100 m

Table 2-3

AM radio frequency 0.54-1.6 MHz (540-1600 kHz )

TV (Channel 2) Slightly over 64 MHz

FM radio frequency 88.8-108.8 MHz

RF used in MRI 3-100 MHz

Table 2-2 summarizes the important electromagnetic windows in

nature. In this table, the following notations are used:
keV = 103 eV = kilo-electron-volts
pm = 10-12 m = picometer
nm = 10-9 m = nanometer
MHz = 106 Hz = megahertz
MeV = 10-3 eV = million electron volts
In MRI, we deal with much lower energies than X-ray or even visible
light. We also deal with much lower frequencies. (The energy of an
electromagnetic wave is directly proportional to its frequency, E = hv.)
The wave lengths are also much longer in the radio frequency (RF)
window. Table 2-3 contains a few examples of frequency ranges in the
electromagnetic spectrum.

Figure 2-3. A spinning charged particle generates a magnetic field.

This is why the electromagnetic pulse used in MRI to get a signal is

called an RF pulse—it is in the RF range. It belongs to the RF window of
the electromagnetic spectrum.

Spins and Electromagnetic Field

One of the pioneers of NMR theory was Felix Bloch of Stanford
University, who won the Nobel Prize in 1946 for his theories. He
theorized that any spinning charged particle (like the hydrogen
nucleus) creates an electromagnetic field (Fig. 2-3). The magnetic
component of this field causes certain nuclei to act like a bar magnet,
that is, a magnetic field emanating from the south pole to the north
pole (Fig. 2-4).

In MRI, we are interested in charged nuclei, like the hydrogen

nucleus, which is a single, positively charged proton (Fig. 2-5).
Figure 2-4. A bar magnet with its associated magnetic field.
 Figure 2-5. A spinning charged hydrogen nucleus (i.e., a proton)
generating a magnetic field.

The other thing that we know from quantum theory is that atomic
nuclei each have specific energy levels related to a property called
spin quantum number S. For example, the hydrogen nucleus (a single
proton) has a spin quantum number S of 1/2:
S (1H) = 1/2
The number of energy states of a nucleus is determined by the formula:
Number of energy states = 2S + 1
For a proton with a spin S = 1/2, we have
Number of energy states = 2 (1/2) + 1 = 1 + 1 = 2 Therefore, a hydrogen
proton has two energy states denoted as − 1/2 and + 1/2. This means
that the hydrogen protons are spinning about their axis and creating a
magnetic field. Some hydrogen protons spin the opposite way and have
a magnetic field in just the opposite direction. The pictorial
representation of the direction of proton spins in Figure 2-6 represents
the two energy states of the hydrogen proton. Each one of these
directions of spin has a different energy state.

Figure 2-6. The direction of the generated magnetic field depends

on the direction of rotation of the spinning protons.

Other nuclei have different numbers of energy states. For example,

13Na has a spin S = 3/2

Number of energy states of 13Na = 2 (3/2) + 1 = 4

The four energy states of 13Na are denoted as (−3/2, −1/2, 1/2, 3/2).
The important fact about all this is that in the hydrogen proton, we
have one proton with two energy states that are aligned in opposite
directions, one pointing north (parallel), and the other pointing south
(antiparallel). (If there were an even number of protons in the nucleus,
then every proton would be paired: for every proton spin with magnetic
field pointing up, we'd have a paired proton spin with magnetic field
pointing down (Fig. 2-7). The magnetic fields of these paired protons
would then cancel each other out, and the net magnetic field would be
zero.) When there are an odd number of protons, then there always
exists one proton that is unpaired. That proton is pointing either north
or south and gives a net magnetic field (Fig. 2-8) or a magnetic dipole
moment (MDM) to the nucleus. Actually, an MDM is found in any nucleus
with an odd number of protons, neutrons, or both. Dipole-dipole
interactions refer to interactions between two protons or between a
proton and an electron.

Figure 2-7. The magnetic fields of paired protons (rotating in

opposite directions) cancel each other out, leaving no net magnetic
field.

The nuclei of certain elements, such as hydrogen (1H) and fluorine

(19F), have these properties (Table 2-4). Every one of these nuclei with
an odd number of protons or neutrons can be used for imaging in MR.
However, there is a reason why we stay with hydrogen. We use
hydrogen for imaging because of its abundance. Approximately 60% of
the body is water. We find hydrogen protons (1H), for example, in
water (H2O) and fat (—CH2—). Later on we'll find out how we use the
spin of the hydrogen proton and avoid the spins of all the other nuclei
with odd numbers of protons.

Figure 2-8. Unpaired protons yield a net magnetic field.

Table 2-4

Spin Quantum Gyromagnetic Ratio

Nucleus Number (S) (MHz /T )

1H 1/2 42.6

19F 1/2 40.0

 23Na 3/2 11.3

13C 1/2 10.7

17O 5/2 5.8

Magnetic Susceptibility
All substances get magnetized to a degree when placed in a magnetic
field. However, the degree of magnetization varies. The magnetic
susceptibility of a substance (denoted by the Greek symbol χ) is a
measure of how magnetized they get. In other words, χ is the measure
of magnetizability of a substance.
To develop a mathematical relationship between the applied and
induced magnetic fields, we first need to address the confusing issue
regarding the differences between the two symbols encountered when
dealing with magnetic fields: B and H. We caution the reader that the
following discussion is merely a simplification; an advanced physics
textbook will have details on the theory of electromagnetism. The field
B is referred to as the magnetic induction field or magnetic flux
density, which is the net magnetic field effect caused by an external
magnetic field. The field H is referred to as the magnetic field
intensity. These two magnetic fields are related by the following:
B = µH or µ = B/H
where µ represents the magnetic permeability, which is the ability of a
substance to concentrate magnetic fields. The magnetic susceptibility χ
is defined as the ratio of the induced magnetic field (M) to the applied
magnetic field H:
M = χH or χ = M/H
Furthermore, χ and µ are related by the following:
µ = 1 + χ
 making sure that the units used are consistent.

Three types of substances—each with a different magnetic

susceptibility—are commonly dealt with in MRI: paramagnetic,
diamagnetic, and ferromagnetic. These are described below.

Paramagnetism, Diamagnetism, and

Ferromagnetism.
1. Diamagnetic substances have no unpaired orbital electrons. When
such a substance is placed in an external magnetic field B0, a weak
magnetic field (M) is induced in the opposite direction to B0. As a
result, the effective magnetic field is reduced. Thus, diamagnetic
substances have a small, negative magnetic susceptibility χ (i.e., χ <
0 and µ < 1). They are basically nonmagnetic. The vast majority of
tissues in the body have this property. An example of diamagnetic
effect is the distortion that occurs at an air-tissue interface (such as
around paranasal sinuses).
2. Paramagnetic substances have unpaired orbital electrons. They
become magnetized while the external magnetic field B0 is on and
become demagnetized once the field has been turned off. Their
induced magnetic field (M) is in the same direction as the external
magnetic field. Consequently, their presence causes an increase in
the effective magnetic field. They, therefore, have a small positive χ
(i.e., χ > 0 and µ > 1) and are weakly attracted by the external
magnetic field. In such substances, dipole-dipole (i.e., proton-proton
and proton-electron) interactions cause T1 shortening (bright signal
on T1-weighted images). The element in the periodic table with the
greatest number of unpaired electrons is the rare earth element
gadolinium (Gd) with seven unpaired electrons, which is a strong
paramagnetic substance. Gd is a member of the lanthanide group in
the periodic table. The rare earth element dysprosium (Dy) is
another strong paramagnetic substance that belongs to this group.
Certain breakdown products of hemoglobin are paramagnetic:
deoxyhemoglobin has four unpaired electrons, and methemoglobin
has five. Hemosiderin, the end stage of hemorrhage, contains, in
 comparison, more than 10,000 unpaired electrons. Hemosiderin
belongs to a group of substances referred to as superparamagnetic,
which have magnetic susceptibilities 100 to 1000 times stronger than
paramagnetic substances.
3. Ferromagnetic substances are strongly attracted by a magnetic field.
They become permanently magnetized even after the magnetic field
has been turned off. They have a large positive χ, even larger than
that of superparamagnetic substances. Three types of ferromagnets
are known: iron (Fe), cobalt (Co), and nickel (Ni). Examples include
aneurysm clips and shrapnel.
As stated earlier, most tissues in the body are diamagnetic. For
example, bulk water is diamagnetic. This may be surprising to hear
because the protons that make up the water are the basis for NMR. It is
true that the individual protons in a water molecule exhibit a magnetic
moment (referred to as a nuclear magnetic moment or nuclear
paramagnetism), but bulk water is diamagnetic, and its net induced
magnetization is in the direction opposite to the main magnetic field.
This has to do with the fact that NMR depends on nuclei (protons and
neutrons), whereas bulk magnetism depends on electrons.

How Do We Actually Perform MR Imaging?

Let's review a few examples of different types of imaging.
a. In photography, there is an object and a light source that emits light.
The light is reflected off the object and is then received by a
photographic plate within a camera (Fig. 2-9). This, of course, is
utilization of the visible light window of the electromagnetic
spectrum. Visible light does not penetrate the object, but instead is
reflected off of it.
 Figure 2-9. In photography, light is reflected off the object and is
received by a photographic plate in a camera.

b. In X-ray imaging, we have an X-ray source that emits radiation that

penetrates the object. This penetrating radiation is then received by
a photographic (X-ray) plate (Fig. 2-10).
c. In MRI, low-frequency radio waves penetrate the tissue and reflect
back off magnetized spins within the object (Fig. 2-11).
RF and MR Signal. If spinning, unpaired protons are placed in an
external magnetic field, they will line up with that magnetic field. If an
RF wave of a very specific frequency is then sent into the patient, some
spins will change their alignment as a result of this new magnetic field.
After the RF pulse, they generate a signal as they return to their
original alignment. This is the MR signal that we measure (Fig 2-12).

Figure 2-10. In X-ray, radiation penetrates the object and reaches

a photographic plate behind the object.

Spatial Encoding. If we generate a signal from the entire body, how do

we differentiate whether the signal is coming from the head or the foot
of the patient? This is the process of spatial encoding, which is used to
create an image.

It requires the use of gradient coils, which will be discussed later in this
chapter.
 Figure 2-11. In MRI, an RF wave, or an RF pulse, is transmitted
into the patient, and a signal is received from magnetized spins
(protons) in the body.

B0 Field. The external magnetic field is denoted B0. In MRI, B0 is on the

order of 1 Tesla (1 T). One Tesla is equal to 10,000 Gauss. To appreciate
the strength of this field, the earth's magnetic field, in comparison, is
only about 0.5 Gauss (30,000 times weaker than a 1.5-T scanner!). This
field is not uniform in reality. These nonuniformities are usually caused
 by improper shimming or environmental distortions. The required
standard for magnetic uniformity is on the order of 6 to 7 ppm (parts
per million). Proper shim coils (see later text) can help minimize this
problem.

Figure 2-12. After the RF pulse, spins in the patient generate a

signal, which can be measured by a receiver.

Types of Magnets
First of all, magnets can be categorized in terms of their field strength;
five types exist:
1. Ultrahigh field (4.0 to 7.0 T); mainly used for research
2. High field (1.5 to 3.0 T)
3. Midfield (0.5 to 1.4 T)
4. Low field (0.2 to 0.4 T)
5. Ultralow field (<0.2 T)
Next, we can categorize magnets in terms of their design; three main
types exist:
1. Permanent magnets
2. Resistive magnets
3. Superconducting magnets
1. Permanent magnets (mainly seen with OPEN MRI scanners such as
Hitachi AIRIS) always stay on and cannot be turned off. They have
the advantage of lower cost and lower maintenance (they require
no cryogens for cooling).
2. Resistive magnets (such as the 0.23 T Philips Panorama and the
Fonar 0.6 T Standup) are based on the electromagnetic principle
that electric current running through a coil produces a magnetic
field. These magnets can be turned off and on.
3. Superconducting magnets are a form of electromagnets. These
magnets operate near absolute zero temperature (e.g., 4.2 K or
−270°C). Consequently, there would be almost no resistance in
their wires. This, in turn, allows us to use very strong electric
currents to generate a high magnetic field without generating
significant heat (hence the name superconducting). To achieve
these ultralow temperatures, cryogens (such as liquid nitrogen
and/or liquid helium) are required (which are very expensive).
Most of the scanners available today are superconducting magnets.

Magnetic Dipole Moment

From now on, we will just talk about protons of hydrogen nuclei. We
won't discuss any other nuclei. Let's take a number of protons. They all

have their own small magnetic fields and they are all spinning about
their own axes. Each one of the magnetic fields is called an MDM and is
denoted by the symbol µ. The axes of the MDMs are arranged in a
random way, and they all cancel each other out. If we add up all the
dipole moments, the net magnetic field will be zero (Fig. 2-13). This
result occurs in the absence of any external magnetic field (B0).
 Figure 2-13. In the absence of an external magnetic field B0, net
magnetization is produced.

Figure 2-14. In the presence of an external magnetic field B0, net

magnetization is produced.

What happens if we turn on an external magnetic field? What would

happen to the proton spins? They will act like bar magnets and line
themselves up with the large magnetic field, much like compass
needles in the earth's magnetic field (Fig. 2-14). However, they don't all
line up in the same direction. Approximately half point north and half
point south. Eventually, enough extra spins point north (about one in a
million1 to make the net magnetization point in the direction of B0.
Let's examine how this happens. At time t = 0, proton spins are
distributed randomly and the net magnetic field at t = 0 is zero.
 Immediately after being placed in a magnetic field, half the spins are
lined up in the direction of the magnetic field and half are lined up in
the reverse direction. Over time, more spins line up in the direction of
the magnetic field, creating net magnetization (Fig. 2-15). If we graph
the net magnetization versus time, it will look like the curve in Figure
2-16. This increase in magnetization follows an exponentially growing
curve that we talked about earlier (see Chapter 1). The time constant
of this curve depends on the following:
1. The kind of tissue we are imaging
2. The strength of the magnet

T1 Relaxation Time
The time constant of the curve in Figure 2-16 is denoted T1. Therefore,
the growth of magnetization M occurs with a time constant T1

described by the expression 1 − e−t/T1. Usually, when we talk about T1,

we're talking about recovery of magnetization along the axis of the B0
field. As time goes by, more and more spins are going to line up with
the external magnetic field. The net magnetization keeps growing
exponentially until it reaches a limit (Fig. 2-17). Remember how with
an exponential growth curve we would draw a tangent at the beginning
of the curve to establish T1, and then we would draw a tangent from
the curve at that time to establish 2 T1? After about four or five T1
times, we almost reach the plateau of the exponential growth curve.
Figure 2.15. Vector representation of net magnetization M0 a
certain amount of time after introduction of the external magnetic
field B0.
 Figure 2-16. The graphic representation of net magnetization over
time, which turns out to be an exponential function.

If we were to change the strength of the magnetic field, what would

happen to T1? If B0 decreases, the T1 of the tissue also decreases:
↓ B0 → ↓ T1
For instance, biologic tissues have shorter T1 values at 0.5 T than at 1.5
T.
 Figure 2-17. Net magnetization is described by the recovery curve
1 − e−t/T1, where T1 is the time constant (recovery rate).

Proton (Spin) Density

Magnetization also depends on the density of the protons (or spins),
that is, how many protons per unit volume there are in the tissue.
Certain tissues have more protons per unit volume than other tissues.
For example, air doesn't have a large number of protons in it, so it has
a very small proton density (spin density). We denote proton density
or spin density by N(H). It isn't just the absolute number of protons in
the tissue that's important; it is also the number of protons that are
mobile enough to be able to change direction and line up with the
external magnetic field.
N(H) = density of mobile protons
The net magnetization at a particular time is based both on the T1 of
the tissue and on the mobile proton density:
Magnetization ∝ N(H)(1 − e−t/T1)
If we redraw the T1 growth curve, the x-axis would be time, and the y-
axis would be
M = N(H)(1 − e−t/T1)

Precession
When a proton is placed in a large magnetic field, it begins to “wobble”
or precess. When we take a single proton spinning about its axis, but

not in an external magnetic field, it will generate its own small

magnetic field (Fig. 2-18). When we turn the external magnetic field
on, the proton behaves like a spinning top, which not only spins about
its own axis but also wobbles about the vertical axis as a result of
gravity (Fig. 2-19). The proton, likewise, not only spins about its own
axis, but also rotates or “precesses” about the axis of the external
magnetic field (B0).
 Figure 2-18. In the absence of an external magnetic field B0, a
proton rotating about its own axis generates a magnetic field.

Each proton spins much faster about its own axis than it rotates or
precesses around the axis of the external magnetic field.
Figure 2-19. In the presence of an external magnetic field B0, a
proton not only rotates about its own axis but also “wobbles”
about the axis of B0.
Larmor Equation
The rate at which the proton precesses around the external magnetic
field is given by an equation called the Larmor equation:
ω = γB0 (Eqn. 2-1)
where ω is the angular precessional frequency of proton, γ the
gyromagnetic ratio, and B0 the strength of external magnetic field.
The angular frequency ω can be expressed in Hertz (Hz) or radians per
second, depending on the units used for γ. If γ is in terms of MHz/T,
then ω (or, actually, the linear frequency f) is expressed in terms of
MHz. The gyromagnetic ratio γ is a proportionality constant that is fixed
for the nucleus with which we're dealing. For hydrogen protons, γ(H) =
42.6 MHz/T.

Example
If the magnetic field strength is 1 T, the precessional frequency of
hydrogen is
(42.6)(1) = 42.6 MHz
As the external magnetic field strength increases, the precessional
frequency of the hydrogen proton also increases, that is, at 1.5 T
the precessional frequency of hydrogen is
(42.6)(1.5) ≅ 64 MHz
Remember that MRI involves the RF portion of the electromagnetic
spectrum, in the range of 3 to 100 MHz. This range is caused by the
precessional frequency ranges of the hydrogen protons for the
magnetic field strengths we use clinically, that is,
for B0 from 0.2T → 3T, ω = 8.5 MHz → 128 MHz

Coils
A coil is an electrical device generally composed of multiple loops of
wire (Fig. 2-20) that can either generate a magnetic field (gradient
coil) or detect a changing (oscillating) magnetic field as an electric
 current induced in the wire (RF coil). Several types of coils are used in
MRI, including the following:
1. Gradient coils
a. Imaging gradient coil
b. Shim coil
 Figure 2-20. A: “Ideal” loops of coils positioned perpendicular to
changing transverse magnetization rotating in x-y plane. This
changing magnetization induces a voltage in the coils, which
causes current to flow (arrows). B and C: Loops of coils gradually
molded to fit inside bore of magnet while retaining sensitivity to
changing transverse magnetization.

2. Transmit and/or receive RF coils

a. Single phase or quadrature (receive or transmit)
b. Surface or volume (Helmholtz or solenoid)
c. Single or phased array
Transmit/Receive Coil. A transmitter coil sends or transmits an RF
pulse. A receiver coil receives an RF pulse. Some coils are both
transmitters and receivers (such as body coils and head coils). Others
are just receivers (e.g., surface coils). These coils act in much the same
way as do radio or television antennas.
The body coil is a fixed part of the magnet that surrounds the patient
and is used in a variety of applications as a transmitter and/or receiver.
A head coil is a helmetlike device that surrounds the patient's head and
may act as both a transmitter and receiver or just as a receiver. There
are a variety of surface coils (e.g., coils used in imaging the joints) that
serve as receivers, with the body coil employed as the transmitter.
Even when a surface coil is placed over the area of interest, the
received signals come from the entire body (this situation again differs
from computed tomography [CT] imaging); however, the signals
received in the region of the surface coil have a higher magnitude, that
is, a higher signal-to-noise ratio (SNR). In other words, surface coils
improve SNR in the region of interest.
Gradient Coils. Gradient coils cause an intentional perturbation in the
magnetic field homogeneity (usually in a linear fashion), which allows
one to decipher spatial information from the

received signal and localize it in space. This perturbation or variation in

magnetic field is several orders of magnitude smaller than the external
magnetic field but is significant enough to allow spatial encoding. To
achieve this, three orthogonal gradient coils are used (Fig. 2-21)
corresponding to the axes x, y, and z in a three-dimensional coordinate
system. This then allows encoding (or deciphering) of data in three
coordinates. These gradients are referred to as

Figure 2-21. The gradient coils exist in MRI, one along each
direction (x, y, and z).
1. The slice-select gradient
2. The phase-encoding gradient
3. The frequency-encoding or readout gradient
For an axial image, these would correspond to Gz, Gy, and Gx,
respectively. They are discussed at length in chapters to come.
Shim Coils. These coils are used to create a more uniform external
magnetic field B0. Keep in mind that inhomogeneities in the external
field are undesirable and can cause artifacts, especially when a
gradient-echo or a chemical fat suppression technique is used. Shim
coils help to minimize (although not totally eliminate) such variations.

Figure 2-22. An arbitrary designation of x, y, and z for a patient in

the scanner.

Quadrature Coils. In quadrature coil design, two receivers are present

90° to one another, capable of distinguishing real and imaginary
components of the received signal. This design can increase the SNR by

a factor of .
Solenoid Coils. These coils can be wrapped around the patient and
increase SNR. These coils are usually used in lower field magnets (e.g.,
open scanners), which have a vertical magnetic field orientation (rather
than a horizontal orientation in higher field scanners).
Phased-Array Coils. These coils contain multiple small surface coils
that are positioned on either side of the anatomy of interest. These
coils allow faster scanning with finer details. An example is the pelvic
array coil that allows exquisite visualization of pelvic structures.

Plane of Imaging
Selection of the gradient coils along x-, y-, or z-axis is arbitrary.
Imaging in different (axial, sagittal, coronal, or oblique) planes in MRI is
different from CT and is possible simply by appropriate assignment of
the gradient coils while the patient is always positioned in the magnet
with his or her long axis along the long axis of the scanner (Fig. 2-22).
For instance, assuming the z-axis to be along the long axis of the
magnet in the craniocaudal (CC) direction, the y-axis to be in the
posteroanterior (PA) direction, and the x-axis to be from right to left

(R-L),2 then axial images are obtained by allowing the slice-select

gradient to be along the z-axis. The possibilities are summarized in
Table 2-5. Oblique planes are obtained by combining the previously
mentioned gradients in a linear fashion.

Table 2-5

Phase- Frequency-
Slice-Select Encoding Encoding
Gradient Gradient Gradient

Axial z y x

Sagittal x y z

Coronal y x z
Key Points
In this chapter, we have discussed the basic
principles behind MR. We have talked about
electromagnetic waves, proton spins, external
magnetic fields, and longitudinal magnetization. We
briefly introduced the parameter T1 that is an
inherent property of a tissue. Let's summarize:
1 Electromagnetic waves (as the name applies) have
two components: an electric component (E) and a
magnetic component (H or B). These two
components are perpendicular to each other and 90°
out of phase.
2 The propagation component C is perpendicular to
both the E and B components, all traveling at the
speed of light (c = 3 × 108 m/sec).
3 In MRI, it is the magnetic component that interests
us. The electric component merely generates heat.
4 Electromagnetic waves are periodic functions of
time, oscillating at a frequency of
ω = 2πf
where ω is the angular frequency (in radians per
second) and f is the linear frequency (in cycles per
second or Hz).
5 Many types of electromagnetic waves exist
throughout the electromagnetic spectrum: X-rays,
visible light, microwaves, radiofrequencies, and so
on.
6 The frequencies used in MRI fall in the RF range (3
to 100 MHz). They are therefore called RF pulses.
7 Spinning charged particles generate an
electromagnetic field.
8 An example of the previous point in the body is the
hydrogen proton (1H).
9 The magnetic components of hydrogen protons
behave like bar magnets. This behavior is referred to
as a magnetic dipole moment (MDM).
10 In general, all particles with an odd number of
electrons in their covalent orbit have this property
(i.e., they can generate a magnetic field).
11 Although many different protons exist in the
body, hydrogen protons are dealt with in MRI
because they are the most prevalent in the body
(particularly in H2O, which comprises 60% of the
body).
12 The main magnetic field in MRI is denoted B0.
13 When a patient is placed in a magnetic field B0,
some of the protons are aligned parallel to B0 and
some antiparallel to it, but more are parallel than
antiparallel, producing a net magnetization
(longitudinal magnetization).
14 These protons also oscillate or precess about the
axis of the external magnetic field.
15 The frequency of precession of protons is
described by the Larmor equation ω0 = γB0 where γ
is the gyromagnetic ratio (in MHz/T). Therefore, the
stronger the magnetic field, the faster the protons
precess about it.
16 Magnetic susceptibility refers to the ability of a
substance to get magnetized when placed in a
magnetic field.
 P.30
17 Three types of substances with different magnetic
susceptibility effects were discussed: diamagnetic,
paramagnetic, and ferromagnetic.
18 There are five types of magnets based on field
strength: ultralow field, low field, midfield, high field,
and ultrahigh field.
19 There are three types of magnets based on
design: permanent magnets, resistive magnets, and
superconducting magnets.
20 Most existing scanners are high-field,
superconducting magnets (these require liquid
cryogens like liquid helium and nitrogen for cooling).
21 Most “open”-type MR scanners are permanent or
resistive magnets; they require no cryogens and thus
have low maintenance. However, they usually have
lower field strengths and thus generate less signal.
22 To create an image, RF pulses are transmitted
into the patient. These pulses flip the longitudinal
magnetization and generate a signal from the
patient.
23 RF pulses flip the longitudinal magnetization Mz
away from the z-axis: 90° pulses flip Mz by 90°; 180°
pulses by 180°; and partial RF pulses by α, which is
less than 90°.
24 The received signal has no spatial information.
Three types of gradient coils (sliceselect, readout or
frequency-encoding, and phase-encoding gradients)
are employed for the purpose of spatial
discrimination.
25 Different types of coils are used: body coil, head
 coil, and surface coil.
26 Surface coils are used for smaller body parts
(e.g., joints) to increase the signal and reduce the
noise (thus increasing the signal-to-noise ratio).
27 The rate at which the longitudinal magnetization
recovers from the transverse plane (after having
been flipped by a 90° pulse) is given by the time
parameter T1. This parameter also describes the rate
at which protons are magnetized when they are
placed in an external magnetic field.
28 The equation for this recovery at any time t is
given by
1 − e−t/T1
which is an exponential growth curve.
What do we need to create an image from the
information received from the patient? This process is
initiated by the use of an RF pulse and is discussed in
the next chapter.

Questions
2-1 Calculate the Larmor frequency of a proton at the following
magnetic field strengths:
(a) 0.35 T
(b) 0.5 T
(c) 1 T
(d) 1.5 T
(e) 2 T
(f) 3 T
(the gyromagnetic ratio, γ, of a proton ≅ 42.6 MHz/T).
2-2 T/F The rate at which protons are magnetized when placed in a
magnetic environment is the same as the rate of recovery of
longitudinal magnetization.

2-3 T/F Proton density represents the density of all the protons in
the tissue.

2-4 T/F When placed in a magnetic field, protons will line up with
that field immediately.

2-5 T/F The T1 of a tissue is larger at a stronger magnetic field

environment.

2-6 T/F Electromagnetic waves travel at the speed of sound.

2-7 T/F The main purpose of x, y, and z gradients is for slice

selection and spatial encoding.

2-8 T/F The rate at which protons precess about the main magnetic
field is faster than that about their own axes.

2-9 T/F When placed in a magnetic field, all the protons in the body
will line up with the field.

2-10 T/F Hydrogen protons are used in MRI because of their

abundance.

2-11 T/F In MRI, the imaging plane is determined by proper

assignment of the x, y, and z gradients.

1This number might appear insignificant. However, according to

Avogadro's law, there are over 1023 molecules per gram of tissue. Thus,
in each gram of tissue, there will be 1017 (i.e., 1023/106) excess
hydrogen protons pointing north.
2This is in fact the convention used throughout this book for simplicity.
3
Radio Frequency Pulse

Introduction
In the last chapter, we discussed the concept of longitudinal
magnetization. However, we have not yet addressed the issue of
receiving a signal from the patient. We can only transmit and receive
signals that oscillate (like an AC voltage). In addition, we're only
sensitive to oscillations along certain axes. Because the longitudinal
magnetization is not an oscillating function (like a DC voltage), it
cannot be read by a receiver. In addition, we're not sensitive to
oscillations along the z-axis. Consequently, this magnetization needs to
be “flipped” into the transverse x-y plane (where it can oscillate or
“precess” about the z-axis) to generate a readable signal. This is the
purpose of the radio frequency (RF) pulse.

RF Pulse
Suppose that a patient is in the magnet. Then we transmit an RF pulse.
What happens? Remember that an RF pulse is an electromagnetic
wave. Initially, all the spins are lined up along the axis of the external
magnetic field B0 about which they are precessing (Fig. 3-1). Then we
transmit an RF pulse. In a three-dimensional (x, y, z) coordinate
system, the direction of the external magnetic field always points in
the z direction. Thus, the net magnetization vector M0 will also point in
the z direction (Fig. 3-2).
 Figure 3-1. An RF pulse is transmitted after the protons have been
exposed to the external magnetic field B0.

One point of clarification about the magnetization vector M0: even

though all of the individual spins are precessing around the external
magnetic field axis, the net magnetization (which is made up of the
vector sum of all the individual spins) does not precess. The reason for
this is that all the individual spins are precessing, but they are all out
of phase with each other. Therefore, if we add them all up, they'll have
a large component along the z-axis; however, because of their phase
differences, they all cancel each other out and are left with no
component along the x- or y-axis (Fig. 3-3). (Note that in Fig. 3-3B,
while the two protons are precessing at the same rate, one is pointing
toward the right and the other is pointing toward the left.) Thus, the
net vector of magnetization does not precess (at least initially—it only
precesses in response to the RF pulse—see later text).
Now, let's transmit an RF pulse along the x-axis perpendicular to the
magnetization vector M0, that is, the axis of B0. (The RF pulse would be
along the axis “C” in Fig. 2-2.) Any proton

that is subjected to any sort of magnetic field starts to precess about

the axis of that magnetic field at a frequency ω0 given by the Larmor
equation (ω = γB), where B is the strength of that magnetic field.
Protons precess about the axis of B0 at a frequency ω0 = γB0. Now we
introduce a magnetic field (namely, the magnetic component of the RF
pulse) into the system with a direction along the x-axis. The protons
that were previously aligned with the external magnetic field B0 in the
z direction will now also begin to precess about the x-axis, that is,
about the axis of the new (RF) magnetic field.

Figure 3-2. The net magnetization vector M0 has the same

direction as the external magnetic field B0.
 Figure 3-3. Two protons rotating out of phase (A) will lead to a net
longitudinal magnetization but no component in the x-y plane (B).

At what rate will these protons precess around this new magnetic field?
The new precessional frequency will be
where B1 is the weaker magnetic field associated with the RF pulse.

We are now dealing with two different magnetic

fields:
B0 = a very strong external magnetic field (e.g., 1.5
T)
B1 = a very weak magnetic field generated by the RF
pulse (e.g., 50 mT)

B0 is a fixed magnetic field (much like a DC voltage). B1, however, is an

oscillating magnetic field (much like an AC voltage). It oscillates
because it is derived from the magnetic component of an oscillating
electromagnetic wave.
Because the magnetic field strength of B1 is much weaker than the
external magnetic field B0, the frequency of precession ω1 of the spins
around the axis of B1 is much slower than the precessional frequency
ω0 of the spins around the axis of the external magnetic field B0.
So, since B1 B0, then ω1 ω0.
The protons are precessing about the B0 field (z-axis) at frequency ω0
and about the B1 field (x-axis) at frequency ω1 at the same time. This
results in a spiral motion of the net magnetization vector from the z-
axis into the x-y plane. This spiral motion is called nutation.
Another thing to remember about the RF pulse is that, referring back to
Chapter 1, the RF pulse has a cos (ωt) waveform. The frequency ω of
the RF pulse should be identical to the Larmor frequency of the
precessing protons. Otherwise, the protons will not precess around the
B1-axis of the RF pulse. This point might be clarified if we first discuss
the concept of resonance.
Resonance. If the frequency ω of the RF pulse matches the frequency
of precession of the protons, then resonance occurs. Resonance results
in the RF pulse adding energy to the protons. A simple example of
resonance is the frequency of a child on a swing. Based on the length of
the swing and the weight of the child, a natural mechanical resonance
frequency exists. If the child is pushed faster or slower from this
frequency, the effort will be inefficient. If the child is pushed at his or
her resonance frequency, energy is added and he or she swings higher.
Similarly, if the proton precesses at frequency ω0 and the frequency of
the RF pulse is not ω0 (say it is ω2 instead), then the magnetic field B1
is oscillating at a different frequency than the protons, and the two
frequencies won't be matched. If the RF frequency does not match the
precessional frequency of the spins, the system won't resonate and no
energy will be added.
Consider this in the x-y plane. The protons are spinning at frequency
ω0. If we then have the B1 magnetic field oscillating at a frequency =
ω2 different from the proton precessional frequency ω0, then the
system won't resonate, that is, the protons won't “flip” into the x-y
plane. A point of clarification: the RF pulse is characterized by two
parameters, strength (B1) and frequency (ω2). The frequency of the RF
pulse must match the proton precessional frequency ω0 in order for
resonance to occur—and for the RF pulse to have any effect on the
protons at all. If the frequency ω2 is correct, then the strength of the
RF pulse (B1) results in precession of protons about the x-axis at
frequency ω1 (according to the Larmor equation ω1 = γB1).
If ω0 and ω2 match (i.e., ω2 = ω0), then the system resonates and the
protons flip into the x-y plane. In doing so, the protons precess around
the axis of the B1 magnetic field at a much lower frequency (ω1),
corresponding to the Larmor frequency associated with the RF

magnetic field B1 and not with the larger magnetic field B0.
Another point of clarification: remember that before the RF pulse, the
protons precess about the z-axis but they are out of phase and hence
have no net transverse component. After the RF pulse, the protons are
introduced to a new magnetic field B1 (also oscillating at frequency
ω0). Consequently, they will also tend to line up with the new magnetic
field and will then be in phase. This, in effect, creates transverse
magnetization. As more and more protons line up, phase “coherence”
increases, as does transverse magnetization. Simultaneously, as
previously discussed, the B1 field also causes a spiral downward motion
of the protons. These two factors explain the process of flipping.
Going back to the three-dimensional coordinate system (Fig. 3-4), the
vector M0 (the net magnetization in the direction of the protons aligned
along the external magnetic field) begins to precess about the x-axis in
the z-y plane. Depending on the strength of the RF pulse B1, and its
duration τ, we can determine the flip angle (i.e., the fractional angle
of a single precession):
 Figure 3-4. A certain amount of time after the application of the
RF pulse, the magnetization vector is partially “flipped” toward
the x-y plane, forming an angle θ with the z-axis.

According to Equation 3-2, the flip angle is proportional to

1. τ = the duration of the RF pulse
2. B1 = the strength of the RF magnetic field, that is, the strength of the
RF pulse
3. γ = the gyromagnetic ratio
We could have a very strong RF pulse applied over a short period of
time, or we could have a weak RF pulse applied over a longer time, and
still attain the same flip angle. The relationship between the flip angle
(θ) and frequency (ω1) is

Thus, flip angle = (frequency of RF precession) × (duration of RF pulse)

Rotating Frame of Reference

To simplify the concept of “flipping,” consider a new reference frame
that rotates at the Larmor frequency ω0. (If you wanted to study the
motion of someone riding a carousel, wouldn't it be easier to be on the
carousel than to watch it from outside?)
Suppose there were someone who was not in this coordinate system,
but rather on the outside looking in (Fig. 3-5). To this person, the
protons would be precessing simultaneously about the z-axis of the B0
field at frequency ω0, and about the x-axis of the B1 field at frequency
ω1. This outside observer would witness a rapid

precession around the z-axis that would slowly spiral down into the x-y
plane (Fig. 3-6). This nutational motion is the result of the two
precessional motions happening simultaneously.
Figure 3-5. An outside observer looking at the coordinate system
sees rapid precession of protons and B1 about the z-axis.
 Figure 3-6. The observer outside the coordinate system sees a
spiral motion of the magnetization vector toward the x-y plane.

If, however, the observer is located within the rotating coordinate

system, and moving at the same frequency as one of the oscillating
systems (B1 or B0), then she will see only the movement of the second
system. If, for example, she is rotating at the oscillation frequency of
the spins in the external magnetic field ω0, then she will only notice
the slow precession of the protons from the z-axis into the x-y plane as
if they were moving in a simple arc (Fig. 3-7). The above occurs if and
only if ω0 = ω2, that is, when the RF pulse frequency ω2 matches the
precessional frequency of the protons ω0. This condition will put the
system in resonance.

Figure 3-7. If the observer stands within the coordinate system,

she would then see a simple arc motion rather than a spiral
motion.

If we go back for a moment to the spiral motion of the protons seen

from an outside observation point (Fig. 3-6), the consecutive circles
around the z-axis represent the oscillation frequency ω0 of the spins in
response to the external magnetic field B0, and the slow downward
progress of the spiral to the x-y plane represents the oscillation
frequency of the spins in response to the magnetic field of the RF
pulse, B1. If the observer herself is oscillating within the system at a
frequency = ω0, then all she will see is the slow downward arc of the
protons precessing in response to the RF pulse. Because the magnetic
field created by the RF pulse (B1) is much smaller than the fixed
external magnetic field (B0), the precessional frequency around the z-y
plane is much slower than is the precessional frequency around the z-
axis.
90° RF Pulse. In response to a strong magnetic field in the z direction,
the spins line up. This results in net magnetization, M0. Next, we apply
an external RF pulse that flips the magnetization vector 90° into the x-
y plane. When the magnetization vector is in the x-y plane, we call
this Mxy. Mxy = component of M0 in the x-y plane If the entire vector
flips into the x-y plane, then the magnitude of Mxy equals the
magnitude of the vector M0. This is called a 90° flip (Fig. 3-8). The
pulse that causes the 90° flip is called a 90° RF pulse.

Figure 3-8. When the entire magnetization vector is flipped into

the x-y plane, it is called a 90° flip.
 Figure 3-9. When placed in a magnetic field B0, protons will fall
into one of two energy states: in the lower energy state, protons
are lined parallel to B0, whereas in the higher energy state they
are antiparallel to it.

The protons that are aligned with the external magnetic field are in
two energy states (Fig. 3-9). Those in the lower energy state (E1) are
lined up with (i.e., parallel to) the magnetic field B0, and those in the
higher energy state (E2) are aligned in the opposite direction. After a
90° RF pulse, some of the protons from the lower energy state are
boosted to the higher energy state. This happens only at the Larmor
frequency.
MATH: The Larmor equation can be derived from the following
principle. The energy difference between E1 and E2, denoted δE, is
given by

where µ is the magnetic dipole moment (MDM). In other words, to go

from one energy state to another, the energy required depends on the
magnetic dipole moment of the proton and the strength of the
magnetic field B0. By Planck's law:
where v = f denotes the linear frequency (in cycles per second or Hz),
ω denotes the angular frequency (in radians per second), h is the
Planck's constant (6.62 × 10-34 J/sec or 4.13 × 10-18 keV/sec), and =
h/2π. Then, combining Equations 3-4a and 3-4b, we can deduce that

and thus,

which is the Larmor equation, with

or, alternatively,
f = (2µ/h)B0
With
γ = 2µ/h (in Hz/T)
where

At equilibrium after the protons are placed in the magnetic field, the
number of protons in the low energy state (north-pointing) is greater
than the number in the high energy state (south-pointing), resulting in
the longitudinal magnetization vector M0 (Fig. 3-10A). As energy is
added by the RF pulse to flip the north-pointing protons to the higher
energy state, the number of protons in both states can be equalized.
When this occurs, a measurable longitudinal magnetization vector no
longer exists. In addition, the RF pulse causes the spins to begin
precessing in phase with each other. The vector sum of the in-phase
northand south-pointing precessing protons lies in the transverse plane
(Fig. 3-10B). This transverse magnetization precesses at the Larmor
frequency.
The angular frequency at which the protons rotate 90° about the x-axis
is given by the Larmor equation: ω1 = γB1 where, again, B1 is the
magnetic field associated with the RF pulse. As stated previously, the
phase, that is, the number of degrees of

precession, is related to the frequency ω1 and the duration τ of the RF

pulse:
Figure 3-10. The application of the RF pulse (B) can equalize the
number of northand south-pointing protons (A).
From this, we can calculate the time τ it would take to precess the
protons 90° (π/2), that is, the time it would take for an RF pulse to
“flip” the spins 90° into the x-y plane at a given RF strength B1. Setting
θ = 90° = π/2 = γB1τπ/2 then we obtain

This equation shows that if we keep the RF pulse on for time τπ/2, the
magnetization vector is flipped 90°.

180° Pulse
A 180° pulse has twice the power (or twice the duration) of a 90° pulse,
as shown in Equation 3-5. After a 180° RF pulse, the longitudinal
magnetization vector is inverted, and the spins begin to recover from
−M0. After a 180° RF pulse, the excess north-pointing spins are boosted
from the low energy state to the high energy state.

A 180° pulse exactly reverses the equilibrium northward-pointing excess

without inducing phase coherence, that is, transverse magnetization.
Using Equation 3-5, we can calculate the RF time duration required for
a 180° RF pulse at a given RF strength B1
180° = π = γB1τπ
resulting in
τπ = π/γB1
To recapitulate, to obtain a 180° RF pulse, we can use either an RF
pulse having the same strength as the 90° pulse but twice as long in
duration, or an RF pulse that's twice as strong for the same duration.

Partial Flip
In the case of a partial flip (<90°), the component of magnetization
ending up in the x-y plane (i.e., Mxy) is less than the magnitude of the
original magnetization vector M0 (Fig. 3-11).
In fact, Mxy = M0 sin θ A partial flip is achieved by decreasing either the
strength or the duration of the RF pulse, according to Equation 3-5.
Such flip angles are common in gradient echo (GRE) imaging, as will be
discussed in later chapters.

Figure 3-11. In a partial flip, transverse magnetization is smaller

than the original longitudinal magnetization. In fact, Mxy = M0 sin
θ.
 Auto RF (Prescan)
Prescan is the process of preparing the scanner for a specific patient.
This process is done via an auto RF pulse, which automatically does the
following:
1. It sets transmit gain (which determines the RF power and thus the
flip angle). In fact, the flip angle α is proportional to the square root
of the transmit power:

2. It sets the receive gain.

3. It sets the optimum ω0.

Key Points
1 An RF (radio frequency) pulse is a brief
electromagnetic burst with frequencies in the radio
frequency spectrum.
2 Like all electromagnetic waves, an RF pulse has
associated magnetic and electric fields. We are
interested in the magnetic component B1. (The
electric component causes tissue heating.)
3 The purpose of the RF pulse in MRI is to flip the
longitudinal magnetization.
4 This flip is done by first causing the protons to
precess in phase about the axis of the external field
(B0), as well causing them to precess about the axis
of the RF field (B1). The result is a spiral motion of
spins toward the x-y plane, called nutation.
 5 The flip angle is a function of the RF strength (B1)
and its duration (τ) and could be 180°, 90°, or a
fraction thereof (i.e., a partial flip 90°), depending on
the clinical application.

Questions
3-1 T/F The flip angle is determined by the duration of the RF pulse
and its power.

3-2 T/F The function of the RF pulse in MRI is to magnetize the

protons.

3-3 T/F The immediate action of the RF pulse is to cause the protons
to precess in phase.

3-4 T/F An RF pulse has a magnetic component.

3-5 T/F An RF pulse stands for radio frequency pulse, which is a form
of an electromagnetic wave with frequencies in the radio frequency
spectrum.

3-6 T/F A 180° pulse has 10 times the power of a 90° pulse.

3-7 T/F A partial flip has an angle between 0° and 90°.

4
T1, T2, and T2*

Introduction
We have already introduced relaxation times T1, T2, and T2*. In this
chapter, we discuss the physical properties behind them and see what
conditions cause them to be increased or decreased. As we have
discussed before, T1 and T2 are inherent properties of tissues and are
thus fixed for a specific tissue (at a given magnetic field strength). The
parameter T2*, however, also depends on inhomogeneities in the main
magnetic field, but again is fixed for a specific tissue within a given
external magnetic environment.

T1 Relaxation Time
The term relaxation means that the spins are relaxing back into their
lowest energy state or back to the equilibrium state. (Equilibrium by
definition is the lowest energy state possible.) Once the radio
frequency (RF) pulse is turned off, the protons will have to realign
with the axis of the B0 magnetic field and give up all their excess
energy.
T1 is called the longitudinal relaxation time because it refers to the
time it takes for the spins to realign along the longitudinal (z)-axis. T1
is also called the spin-lattice relaxation time because it refers to the
time it takes for the spins to give the energy they obtained from the RF
pulse back to the surrounding lattice in order to go back to their
equilibrium state.

T1 = longitudinal relaxation time

T1 = thermal relaxation time
T1 = spin-lattice relaxation time
 Immediately after the 90° pulse, the magnetization Mxy precesses
within the x-y plane, oscillating around the z-axis with all protons
rotating in phase (Fig. 4-1). After the magnetization has been flipped
90° into the x-y plane, the RF pulse is turned off. A general principle of
thermodynamics is that every system seeks its lowest energy level.
Therefore, after the RF pulse is turned off, two things will occur:
1. The spins will go back to the lowest energy state.
2. The spins will get out of phase with each other.
These events result from two simultaneous but separate processes
occurring after the RF pulse is turned off (Fig. 4-2):
1. The Mxy component of the magnetization vector decreases rapidly.

Figure 4-1. After the RF pulse, the longitudinal magnetization

vector is flipped into the x-y plane.
 Figure 4-2. Once the RF is turned off, the transverse
magnetization vector begins to decay while the longitudinal
component begins to recover.

2. The Mz component slowly recovers along the z-axis.

Question: What time constant characterizes the rate at which the Mz

component recovers its initial magnetization M0?
Answer: The T1 relaxation time.
When we first discussed magnetization, we said that the protons start
to line up with the external magnetic field at a rate given by T1.
(Because T1 is a time [in sec], the rate is 1/T1 [in sec-1].) The same
phenomenon occurs when we flip the magnetization M0 away from the
longitudinal z-axis and then allow it to realign with the main magnetic
field after an RF pulse. The rate at which Mz recovers to M0 is also
given by T1.
Immediately after a 90° pulse, all magnetization is in the x-y plane.
The Mz component then starts to grow at a rate characterized by T1
(Fig. 4-3):

Figure 4-3. The graph of recovery of longitudinal magnetization

with the growth rate of T1.
T2 Relaxation Time
Figure 4-2 also shows rapid decay of the Mxy component after the RF
pulse is turned off.
Question: What time constant characterizes the rate at which the Mxy
component decays?
Answer: The T2 relaxation time.
As the longitudinal magnetization vector Mz recovers, the transverse
vector Mxy decays at a rate characterized by T2 (Fig. 4-4):

Realize that the recovery of magnetization along the z-axis and the
decay of magnetization within the x-y plane are two independent
processes occurring at two different rates (Fig. 4-5). Take a simple
exponential process. We would expect the rate at which this process
decays in the x-y plane to be the same as that at which it grows along
the z-axis (Fig. 4-6). This is not the same in the MR system we are
discussing because this system involves a much more complicated
process. T2 decay occurs 5 to 10 times more rapidly than T1 recovery
(Fig. 4-7). To understand this, we need to understand the concept of
dephasing.
Dephasing. After the 90° RF pulse is turned off, all spins are in phase;
they are all lined up in the same direction and spinning at the same
frequency ω0. There are two phenomena that will make the spins get
out of phase: interactions between spins and external field
inhomogeneities.
Figure 4-4. The graph of transverse magnetization with the decay
rate of T2 and RF off.
 Figure 4-5. The recovery of longitudinal magnetization and the
decay of transverse magnetization occur at the same time but are
independent of each other with the RF off.

1. Interactions Between Individual Spins

When two spins are next to each other, the magnetic field of one
proton affects the proton next to it. Assume one proton is aligned with
the field and the other is against it (Fig. 4-8). The proton aligned with
B0 creates a slightly higher magnetic field for its neighbor so that
proton #1 is exposed to the magnetic field B0 plus a small magnetic
field created by the other proton (δB). The precessional frequency of
the proton then will increase slightly as ω (proton #1) = (B0 + δB)
On the other side, the other proton is exposed to a slightly weaker
magnetic field because the other proton points against B0. Therefore,
its overall magnetic field strength exposure will be slightly less. The
precessional frequency of proton #2 then will decrease slightly as
ω (proton #2) = (B0 − δB)

Figure 4-6. The rate of growth and decay of a simple exponential

process is expected to be similar.

The difference in the magnetic environment created by these proton-

proton interactions may be very small, but it makes a difference in the
overall homogeneity of the magnetic field to which the spins are
exposed. Thus, the first cause of dephasing is inherent in the tissue. It
is called a spin-spin interaction. This interaction is an inherent
property of every tissue and is measured by T2.

T2 = transverse relaxation time

T2 = spin-spin relaxation time

2. External Magnetic Field Inhomogeneity

This is the second phenomenon that makes spins get out of phase. No
matter how good a system we have, no matter how stable the external
magnetic field is, some variation in the homogeneity of the magnetic

field still exists (usually measured in parts per million).

Figure 4-7. Contrary to the previous figure, the rate of decay of

transverse magnetization is several times that of the recovery of
longitudinal magnetization.

External magnetic field inhomogeneity makes protons in different

locations precess at different frequencies because each spin is exposed
to a slightly different magnetic field strength. These varying
frequencies are very close to each other and very close to the true
Larmor frequency; however, these tiny differences in frequency result
in spin dephasing.

The two causes of spin dephasing:

1. Spin-spin interactions (internal inhomogeneities)
2. External magnetic field inhomogeneities

These two phenomena together cause protons to spin at slightly

different frequencies. Imagine that we have three protons:
1. One is precessing at the true Larmor frequency = ω0.
2. One, exposed to slightly higher magnetic fields, is precessing at a
frequency slightly faster than the Larmor frequency = ω0+.

Figure 4-8. Two adjacent protons, one aligned with the field and
the other against it.

3. One, exposed to slightly weaker magnetic fields, is precessing at a

frequency slightly slower than the Larmor frequency = ω0−.

If we wait long enough, the three protons in the x-y plane will get
completely out of phase. The net magnetic field within the x-y plane
will then go to 0.
Therefore, at time t = 0, all spins are in phase, and their vector sum
will be at maximum magnitude. As the spins begin getting out of phase
with each other, their summation vector will become smaller and
smaller. When all the spins are completely out of phase with each
other, their vector sum will be zero. The effect of spin-spin interaction
depends to a degree on the proximity of the spins to each other. For
example, in water (H2O), the protons are separated more widely than
they are in a solid tissue. Hence, the dephasing effect of spin-spin
interaction might not be as prominent in H2O as it is in a solid tissue.

The Received Signal

Let's go back to the x-y plane with the RF aligned along the x-axis. The
RF coil (e.g., head or body coil) is often both a transmitter and a

receiver. The signal is received at the same location at which it is

transmitted. Remember that a moving charged particle generates a
magnetic field. The reverse is also true. A magnetic field causes
movement of charged particles, that is, electrons. If we have a wire
with electrons running in one direction (away from us), the direction of
the magnetic field (according to the right-hand rule) can be determined
(in this case, it is pointing up; Fig. 4-9A). Similarly, if we have a straight
wire, and we have an oscillating magnetic field about this wire, the
magnetic field will induce a voltage and current in the wire (Fig. 4-9B).
 Figure 4-9. A, B: The righthand rule determines the direction of
generated magnetic field by an electric current in a wire.

The measured current is what we mean by a signal. Remember that

after a 90° pulse, the magnetization rotates in the x-y plane at
frequency ω0. This magnetization reflects a group phenomenon of
multiple precessing protons. Associated with each proton is a magnetic
field that is also precessing. Immediately after a 90° pulse, the protons
precess in phase. When the magnetic field of each spin (or each group
of spins) is in the same direction as the RF coil receiver, a very large
signal is induced in the RF receiver coil.
Thus, at time t = 0 (in Fig. 4-10A), all the protons are lined up in the
direction of the RF coil. As the spins rotate 90° at time t = t1, there
will be no component of the magnetization vector along the x
direction. All magnetization points in the y direction. However, this RF
coil can only detect the component of magnetization along the x-axis.
So at time t1, there is no signal. After another 90° rotation (at time
t2), a signal exists, but it is the negative of the original signal. At time
t3, there is again no magnetization in the x direction, and, therefore,
no signal. At time t4, spins are again lined up with the receiver coil and
signal is maximal. A graph of the received signal will then look like a
sinusoidal curve (Fig. 4-10B). The frequency of the received signal is ω0
because the protons are spinning at frequency ω0.
However, is this really the received signal, or is there more to it?

Free Induction Decay

In an ideal situation, if we had a perfectly uniform magnetic field, the
received signal would, in fact, have looked like Figure 4-10B. However,
this is not what's really happening. What really happens is the following
(Fig. 4-11): We start at time t = 0 in the x direction. However, because
of spin dephasing (namely, spin-spin interactions and external magnetic
field inhomogeneities), by the time the spins reach t4, they will have
dephased slightly and the signal coming from the spins will be slightly
less than it was originally. The signal becomes weaker and weaker as
time goes by, and it spirals to the center of the x-y plane.
 Figure 4-10. The relationship between transverse magnetization
(A) and the received signal (B) at different points in time.

The signal vector is continuously decaying in magnitude as it is

precessing around the x-y plane. How would the signal look to the RF
receiver? The answer lies in Figure 4-12. This figure shows the shape of
the signal picked up by the receiver; it is an oscillating, decaying
signal. It is called a free induction decay (FID) because after we turn
off the RF pulse:
 Figure 4-11. The spiral-like decay of transverse magnetization.

1. The spins begin to precess freely.

2. The signal starts to decay with time.
3. The spins induce a current in the receiver coil.
Thus, the FID results from an oscillating magnetic field generated by
the oscillating spins, which induces a current in a receiver coil. This
decaying oscillating signal is described mathematically as
 We've seen these terms before (refer to Chapter 1):
1. (cos ω0t): This is the formula for an oscillating wave, with a
frequency of ω0.
2. (e-t/T2*): Because the signal is decaying we have to include an
exponential function. The time constant of this exponential function
is given by T2*.
Therefore, the general form of the received signal is based on:
1. An oscillating signal, which varies as cos ω0t

Figure 4-12. The decaying sinusoidal waveform of the received

signal (the FID).

2. A decaying signal, which decays with time constant T2* as given by

the exponential e-t/T2*
Differences Between T2 and T2*. T2* decay depends on both
1. External magnetic field
2. Spin-spin interactions
T2 decay depends primarily on
1. Spin-spin interactions1
T2 of a tissue, because it depends only on spin-spin interactions, is
fixed—we have no control over what the spins do to each other. T2*
depends on the homogeneity of the external magnetic field, so it is not
fixed. It varies depending on how uniform the main magnet is. T2* is
always less than T2. T2* decay is always faster than T2 decay (Fig. 4-
13). The following equation relates the two together:

The term 1/T is the relaxation rate with units of sec−1 (recall that 1/T
is a frequency).
The relaxation rate (1/T2*) depends on the relaxation rate of the tissue
(1/T2) plus the magnetic field inhomogeneity of the external magnet. If
we have a perfect magnet that does not introduce any inhomogeneity,
then δB = 0 and T2* = T2. The newer systems have less magnetic field
inhomogeneity, thus making the T2* effects less strong; however,
complete homogeneity is not possible. Hence, there will always be
some T2* effect.
Figure 4-13. T2 and T2* decay curves.

Key Points
1 The rate of recovery of the longitudinal
magnetization is given by T1.
2 The rate of decay of the transverse magnetization
is given by T2.
3 The rate of decay of the FID is given by T2*.
4 T1 is 5 to 10 times greater than T2.
5 T2* is always less than T2.
6 T2 is a result of spin-spin interactions (internal
tissue inhomogeneities), whereas T2* is dependent
on both internal and external (main magnetic field)
inhomogeneities.
 7 FID is produced by a rotating magnetic field, which
induces an electric current in a stationary coil.

Questions
4-1 True/false questions:
(a) T2* depends on the external magnetic field inhomogeneity.
(b) T2 depends on the external magnetic field inhomogeneity.
(c) T2 depends on T2*.
(d) T2* depends on T2.

4-2 The recovery of longitudinal magnetization is proportional to

(a) e-t/T1
(b) 1 - e-t/T1
(c) 1 - e-t/T2
(d) e-t/T2
(e) none of the above

4-3 The decay of transverse magnetization is proportional to

(a) e-t/T1
(b) 1 - e-t/T1
(c) 1 - e-t/T2
(d) e-t/T2
(e) none of the above

4-4 T/F The rate of decay of the FID is given by T2.

4-5 Which one of the following equations is correct?

 (a) T2 > T2* > T1
(b) T2* > T2 > T1
(c) T1 > T2 > T2*
(d) T1 > T2* > T2

4-6 Match the following: (i) T1; (ii) T2 with

(a) recovery of longitudinal magnetization
(b) decay of transverse magnetization

1T2 also depends on diffusion (i.e., how rapidly spins spread out and

leave the lattice); however, this is a minor factor in comparison with

spin-spin interactions.
5
TR, TE, and Tissue Contrast

Introduction
In previous chapters, we discussed the roles of T1 and T2, longitudinal
and transverse magnetization, and the radio frequency (RF) pulse.
Obviously, by doing the procedures described in the previous chapters
only once, we won't be able to create an image. To get any sort of
spatial information, the process must be repeated multiple times, as
we shall see shortly. This is where TR and TE come into play. The
parameters TR and TE are related intimately to the tissue parameters
T1 and T2, respectively. However, unlike T1 and T2, which are inherent
properties of the tissue and therefore fixed, TR and TE can be
controlled and adjusted by the operator. In fact, as we shall see later,
by appropriate setting of TR and TE, we can put more “weight” on T1
or T2, depending on the type of clinical application.
How do we actually measure a signal? With the patient in a large
magnetic field (Fig. 5-1A), we apply a 90° RF pulse, and the
magnetization vector flips into the x-y plane (Fig. 5-1B). Then, we turn
off the 90° RF pulse, and the magnetization vector begins to grow in
the z direction and decay in the x-y plane (Fig. 5-1C). By convention,
we apply the RF pulse in the x direction, and for that reason, in a
rotating frame of reference, the vector ends up along the y-axis (Fig. 5-
1B).
 Figure 5-1. A: Longitudinal magnetization before the RF pulse. B:
Immediately after the RF pulse, the magnetization vector is flipped
into the x-y plane. C: After a certain time period, Mz has recovered
by a certain amount while Mxy has decayed by a different amount.

After a 90° RF pulse, we have decaying transverse magnetization Mxy

(which is the component of the magnetization vector in the x-y plane)
and recovering longitudinal magnetization Mz (which is the component
of the magnetization vector along the z-axis). Remember that the
received signal can be detected only along the x-axis, that is, along the
direction of the RF transmitter/receiver coil. The receiver coil only
recognizes oscillating signals (like AC voltage) and not nonoscillating
voltage changes (like DC voltage). Thus, rotation in the x-y plane
induces a signal in the RF coil, whereas changes along the z-axis do not.
 Figure 5-2. The received signal (the FID) has a decaying sinusoidal
waveform.

At time t = 0, the signal is at a maximum. As time goes by, because of

dephasing (see Chapter 4), the signal becomes weaker in a sinusoidal
manner (Fig. 5-2). The decay curve of the signal is given by the
following term: e−t/T2*
The sinusoidal nature of the signal is given by the equation cos ωt
Therefore, the decaying sinusoidal signal is given by the product (e
−t/T2*)(cos ωt)

When t = 0: cos ωt = cos ω(0) = cos 0 = 1 e−t/T2* = e0 = 1

When t = 0, (e−t/T2*)(cos ωt) = 1. Thus, at time t = 0, the signal is
maximum (i.e., 100%). As time increases, we are multiplying a
sinusoidal function (cos ωt) and a decaying function (e−t/T2*), which
eventually decays to zero (Fig. 5-3).

Figure 5-3. The product of a sinusoidal signal and an exponentially

decaying signal results in a decaying sinusoidal signal.
 Figure 5-4. Immediately after transmission of the RF pulse, an FID
is formed.

When we put a patient in a magnet, he or she becomes temporarily

magnetized as his or her protons align with the external magnetic field
along the z-axis. We then transmit an RF pulse at the Larmor frequency
and immediately get back a free induction decay (FID; Fig. 5-4). This
process gives one signal—one FID—from the entire patient. It doesn't
give us any information about the location of the signal. The FID is
received from the ensemble of all the different protons in the patient's
body with no spatial discrimination. To get spatial information, we have
to specify somehow the x, y, and z coordinates of the signal. Here,
gradients come into play. The purpose of the gradient coils is to
spatially encode the signal.
To spatially encode the signal, we have to apply the RF pulse multiple
times while varying the gradients and, in turn, get multiple FIDs or
other signals (e.g., spin echoes). When we put all the information from
the multiple FIDs together, we get the information necessary to create
an image. If we just apply the RF pulse once, we only get one signal
(one FID), and we

cannot make an image from one signal. (An exception to this statement
is echoplanar imaging [EPI], which is performed after one RF pulse—see
Chapter 22.)
 Figure 5-5. The time interval between two successive 90° RF
pulses is denoted TR.

TR (The Repetition Time)

After we apply one 90° pulse (the symbol we'll use for a 90° RF pulse is
in Fig. 5-5), we'll apply another. The time interval between applications
is called TR (the repetition time).
What happens to the T1 recovery curve during successive 90° pulses
(Fig. 5-6)?
1. Immediately before time t = 0, the magnetization vector points along
the z-axis. Call this vector M0 with magnitude M0.
2. Immediately after t = 0, the magnetization vector Mxy lies in the x-y
plane, without a component along the z-axis. Mxy has magnitude M0
at t = 0+.

Figure 5-6. The recovery curves after successive RF pulses.

3. As time goes by and we reach time t = TR, we gradually recover some
magnetization along the z-axis and lose some (or all) magnetization
in the x-y plane. Let's assume at time TR the transverse
magnetization Mxy is very small. What happens if we now apply
another 90° RF pulse? We flip the existing longitudinal magnetization
vector (Mz) back into the x-y plane. However, what is the magnitude
of the magnetization vector Mz at the time TR? Because Mz(t) = M0(1
− e−t/T1) then at t = TR,

4. As we see in the T1 recovery curve, the magnetization vector (Mz) at

time TR is less than the original magnetization vector M0 because the
second 90° RF pulse was applied before complete recovery of the
magnetization vector Mz.
5. After the magnetization vector is flipped back into the x-y plane, it
will begin to grow again along the z-axis (according to the T1
recovery curve) until the next TR, when it will again be flipped into
the x-y plane. We now have a series of exponential curves that never
reach full magnetization.

Received Signal
Let's now take a look at the signal we are receiving (S). Because we are
only applying a series of 90° pulses, the signal will be a series of FIDs:
1. At time t = 0, the initial signal will be a strong FID similar to that
shown in Figure 5-7A.
2. At time t = TR, the signal will be slightly less in magnitude but will
also be an FID (Fig. 5-7B).
3. At time t = 2TR, the signal will be equal in magnitude to that in
Figure 5-7B (Fig. 5-7C).
Because the T1 recovery curve is given by the formula 1 − e−t/T1, if we
could measure the signal immediately after the RF pulse is given with
no delay, then each FID signal would be proportional to 1 − e−TR/T1
(This cannot really happen in practice.) Up to now, the signal S is given
by the formula S ∝ 1 − e−TR/T1

Figure 5-7. The FIDs after successive RF pulses: A: at t = 0; B: at t

= TR; C: at t = 2TR.

Remember that the word “signal” is really a relative term. The signal
that we get is a number without dimension, that is, it has no units. If
we are dealing with a tissue that has many mobile protons, then,
regardless of what the TR and T1 of the tissue are, we'll get more
signals with more mobile protons (see Chapter 2). Thus, when
considering the signal, we must also consider the number of mobile
protons N(H).

For a given tissue, the T1 and the proton density are constant, and the
signal received will be according to the above formula. If we measure
the FID at time TR immediately after the application of the second 90°
RF pulse, it will measure maximal and be equal to N(H) (1 − e−TR/T1).
Therefore, the FIDs that are acquired at TR intervals (i.e., 1TR, 2TR)
are maximal if they can be measured right after the 90° pulse, that is,
right at the beginning of the FID. However, in reality, we have to wait a
certain period until the system electronics allows us to make a
measurement.

TE (Echo Delay Time or Time to Echo)

TE stands for echo delay time (or time to echo). Instead of making the
measurement immediately after the RF pulse (which we could not do
anyway), we wait a short period of time and then make the
measurement. This short time period is referred to as TE.
Let's go back to the T2* decay curve and see what happens. In the x-y
plane, the FID

signal decays at a very rapid rate because of two factors:

 Figure 5-8. The value of the FID at time 0 is M0, whereas at time
TE it is M0·e−TE/T2*.

1. External magnetic field inhomogeneities

2. Spin-spin interactions
The signal decays at rate T2* according to the decay function e−t/T2*
From this we see that if we take the signal measurement right away,
before there is any chance for signal decay, the signal will be equal to
the original magnetization (M0) flipped into the x-y plane (point 1 in
Fig. 5-8). However, if we wait a short time period (TE) before we make
a measurement, the signal will look like point 2 in Figure 5-8.

Now we have to put the two curves together because both T1 recovery
and T2 decay processes are occurring simultaneously (Fig. 5-9).

Figure 5-9. The recovery and decay curves plotted on the same
 graph.

Let's go back to the T1 recovery curve. After the 90° RF pulse, the spins
are flipped into the x-y plane. After a time interval TR, the amount of
received longitudinal magnetization is

Superimposed on this T1 recovery curve, we'll draw another curve,

which is the T2* decay curve, with two new axes. The T2* decay curve
starts out at the value of M0 (1 − e−TR/T1) on the T1 recovery curve and
then decays very quickly. The decay rate of the new curve is given by
T2* according to the formula e−t/T2*
After a period of time TE, we can measure the signal. The value of the
signal at TE will be a fraction of the maximum signal intensity on the T1
recovery curve. In other words, it will be the product of Equations 5-3a
and 5-3b: Signal = S ∝ M0(1 − e−TR/T1)(e−TE/T2*)
The confusing thing about the diagram is that there are two sets of axes
(Fig. 5-9):
1. The first set of axes is associated with TR.
2. The second set of axes is associated with TE.
If we draw them to scale, the T2 decay curve (time scale TE) will be
decaying much faster than the T1 curve is recovering (time scale TR).
However, the graph does give us a visual concept of what the final
signal intensity is going to be.

Because the initial longitudinal magnetization M0 is proportional to the

number of mobile protons, that is, M0 ∝ N(H) then, in general, the
signal intensity we measure is given by

(The difference between T2 and T2* is the correction for external

 magnetic field inhomogeneity achieved with spin-echo techniques.)

Tissue Contrast (T1 and T2 Weighting)

Let's see what happens when we deal with two different tissues. So far
we have been dealing with a single tissue, but now we'll consider two
tissues: tissue A and tissue B.
Question: Of the two tissues in Figure 5-10A, which one has the longer
T1?
Answer: Tissue A has the longer T1 (it takes longer to recover).
If we draw just a tangent along each curve at the origin, tissue A has a
longer T1 than tissue B. However, just by looking at the curves, it takes
tissue A longer to reach equilibrium than tissue B. Let's say we have two
different TRs:
1. Short TR = TR1
2. Long TR = TR2

Figure 5-10. A: Two tissues A and B with different T1s. Which

tissue has the longer T1? B: Consider two different TRs on a
recovery curve. Which TR provides better tissue contrast between
A and B?
Question: Which one of the TRs in Figure 5-10B gives better tissue
contrast?
Answer: TR1 gives the better contrast.
Let's go back to Equation 5-4 and see if this answer makes sense: SI =
N(H)(e−TE/T2*)(1 − e−TR/T1) where SI stands for signal intensity. If TR goes
to infinity, then 1 −e−TR/T1 becomes 1. If TR → ∞, then 1 − e−TR/T1 → 1
and SI → N(H) (e−TE/T2*). If TR is very long, we can get rid of the T1
component in the equation. What this means in practice is that we
eliminate (or, more realistically, reduce) the T1 effect by having a very
large TR.

Long TR reduces the T1 effect.

We can't really achieve a long enough TR in practice to eliminate totally
the T1 effect 100%, but we can certainly minimize the T1 effect with a
TR of 2000 to 3000 msec (in general, if TR is four to five times T1, then
the T1 effect becomes negligible). Let's go back to Figure 5-10B and see
what happens at TR = TR1. At this point, the TR is not long enough to
eliminate the T1 term in the equation (1 − eTR/T1). So we have: signal
intensity (tissue A)/signal intensity (tissue B) = (1 − e−TR1/T1(tissue A))/(1 −
e−TR1/T1(tissue B))

Because the T1s of tissue A and tissue B are different, the short TR
brings out the difference in contrast between tissue A and tissue B.
Thus, for short TR, the two tissues can be differentiated on the basis of
different T1s. In other words, we get T1 tissue contrast with short TR.

Short TR enhances the T1 contrast.

We don't want TR to be very long when we're evaluating T1 because, as
we've already learned, when TR →∞, then (1 − e−TR/T1) approaches 1,
thus eliminating the T1 effect. However, we also don't want TR to be
too short. If TR is close to 0, then e−0/T1 = e0 = 1 and 1 − e−TR/T1 = 1 − 1
= 0
In this situation, with very short TR, we end up with no signal. Ideally,
 we would like to have a TR that is not much different from the T1 of
the tissue under study.

T2* Tissue Contrast

Let's consider T2* contrast between two tissues.
Question: In Figure 5-11A, which tissue has a longer T2*?
Answer: Again, if we graphically draw a tangent at t = 0 for each
curve, we see that tissue A has a longer T2*. Put differently, it takes
the signal from tissue A longer to decay than that from tissue B.

Figure 5-11. A: Two tissues A and B with different T2*s. Which

tissue has the longer T2? B: Consider two different TEs on a decay
curve. Which TE provides better tissue contrast between A and B?

Let's pick two different TEs (Fig. 5-11B). Here we have two TEs:
1. Short TE = TE1
2. Long TE = TE2

Question: Which one of the TEs in Figure 5-11B results in more tissue
contrast between tissue A and tissue B?
Answer: TE2 gives us more contrast.
Let's again look at the formula for signal intensity (Equation 5-4): SI =
N(H)(e−TE/T2*)(1 − e−TR/T1)
If TE is very short (close to zero), then e −TE/T2* approaches 1. TE →0 →
e−TE/T2*→e0 = 1 Then signal intensity = N(H)(1)(1 − e−TR/T1) = N(H)(1 − e
−TR/T1)

This means that, with a very short TE, we get rid of the T2* effect in
the equation. Therefore, we eliminate (or, again, in reality, reduce) the
T2* effect by having a very short TE.

Short TE reduces the T2* effect.

We can see this graphically from the graph (Fig. 5-11B) and
mathematically from the equation (Equation 5-3). When we have a long
TE, we enhance T2* contrast between tissues. Even though the signal-
to-noise ratio is low (because there is greater signal decay for a longer
TE), the tissue contrast is high.

Key Points
1. Long TR reduces T1 effect.
2. Short TR enhances T1 effect.
3. Short TE reduces T2* (T2) effect.
4. Long TE enhances T2* (T2) effect.

Questions
5-1 In the graph in Figure 5-12, the T1 and T2 curves are plotted
simultaneously for convenience. Assume the following values for T1
and T2: for white matter (WM): T1 = 500, T2 = 100 msec for
cerebrospinal fluid (CSF): T1 = 2000, T2 = 200 msec Also assume a
spin density N = 100 for both WM and CSF.
(a) For a TR = 2000 msec, find the relative signal intensities for
WM and CSF (i.e., points A and B on the graph).
(b) Calculate the crossover TE where WM and CSF have identical
T2 weighting (point C).
(c) Now, calculate the signal intensities of WM and CSF for TE =
25 (first echo) and TE = 100 msec (second echo), and the ratio
CSF/WM.
(d) Repeat (a) to (c) for TR = 3000 msec and observe how one
gets more T2 weighting in the second echo (higher ratio
CSF/WM).
(e) Now, calculate the signal intensities for TR = 3000 and TE =
200 msec.
Notice that despite relative loss of signal for both WM and CSF,
the ratio CSF/WM actually increases, indicating more T2
weighting (i.e., CSF gets brighter on the images).

Figure 5-12. T1 and T2 curves are plotted

simultaneously. Use this graph to answer the
questions.

The following values may be helpful for those of you without a

sophisticated calculator: e = 2.27, e−1 = 1/e = 0.37, e−2 = 1/e2 =
 0.14, e−3 = 0.05, e−4 = 0.02, e−5 = 0.01, e−6 [all equal to] 0, e−0.5
= 0.61, e−1.5 = 0.22, e−0.13 = 0.88; ln 0.64 = loge 0.64 = −0.45, ln
0.78 = −0.25

5-2 Suppose that at 1.0 T, the approximate T1 and T2 values for the
following tissues are as follows.

Tissue T1 (msec ) T2 (msec )

H2O 2500 2500

Fat 200 100

CSF 2000 300

Gray matter 500 100

(a) Calculate the signal intensity ratios for

1. H2O/fat
2. CSF/gray matter

for the following pulse sequences:

1. T1WI/SE with TR = 500, TE = 25 msec
2. T2WI/SE with TR = 2500, TE = 100 msec
Note: Assume similar spin densities for these tissues.
(b) Demonstrate the above graphically.
Hint: e−1 = 0.37, e−5 = 0.01, e−0.04 = 0.96, e−1.25 = 0.29, e−12.5 [all
equal to]0 e−2.5 = 0.08, e−0.25 = 0.78, e−0.2 = 0.82, e−0.01 = 0.99, e
−1/3 = 0.72, e−0.25/300 = 0.92.

5-3 A longer TR
(a) increases T1 weighting
(b) reduces T1 weighting
 (c) increases T2 weighting
(d) reduces T2 weighting

5-4 A longer TE
(a) increases T1 weighting
(b) reduces T1 weighting
(c) increases T2 weighting
(d) reduces T2 weighting

5-5 Calculate the signal N(H)(1 − e−TR/T1) e−TE/T2 for the following
theoretical situations:
(a) TR = ∞
(b) TE = 0
(c) TR = ∞ and TE = 0

5-6 Match the following: (i) reduces T1 effect, (ii) enhances T1

effect, (iii) reduces T2 effect, (iv) enhances T2 effect with
(a) short TR
(b) long TR
(c) short TE
(d) long TE
 6
Tissue Contrast: Some Clinical
Applications

Introduction
In the previous chapter, we talked about T1 and T2 weighting in terms
of the time parameters TR and TE. Now let's discuss the T1 and T2
characteristics of the following tissues and see what physical properties
affect them:
1. H2O
2. Solids
3. Fat
4. Proteinaceous material

T2 Characteristics
The T2 characteristics of a tissue are determined by how fast the
proton spins in that tissue dephase. If they dephase rapidly, we get a
short T2. If they dephase more slowly, we get a longer T2.
H2O. Because of the structure of the water molecule (H-O-H) and
because of the sparsity of these molecules, spin-spin interaction among
the hydrogen protons is minimal. Therefore, dephasing occurs at a
much slower rate in water compared with other tissues. The T2
relaxation time for H2O is, therefore, long. Remember that T2 decay is
caused either by external magnetic field inhomogeneities or by spin-
spin interactions within or between molecules. In H2O, the effect of
one hydrogen proton on another is relatively small. The distance
between hydrogen protons both within each molecule and between
adjacent molecules is relatively large, so there is little spin-spin
interaction and, therefore, less dephasing.
Solids. The molecular structure of solids is opposite to that of pure
water. It is a very compactly structured tissue, with many interactions
between hydrogen protons. This large number of spin-spin interactions
results in more dephasing. Thus, the T2 for solids is short.
Fat and Proteinaceous Material. The structure of these materials is
such that there is less dephasing than in solids but more dephasing than
in water. Therefore, T2 for proteinaceous material or fat is
intermediate.

T1 Characteristics
The T1 of a tissue has to do with the way the protons are able to give
off their energy to the surrounding lattice, or to absorb the energy from
the lattice. It turns out that the most efficient energy transfer occurs
when the natural motional frequencies1 of the protons are at the
Larmor frequency (ω0). Recall that the Larmor frequency is
proportional to the strength of the magnetic field: ω0 = γBo

For hydrogen, γ = 42.6 MHz/Tesla

In other words, the precessional frequency of a hydrogen proton is 42.6

MHz in a 1-Tesla

magnetic field. However, the natural motional frequency of hydrogen

protons depends on the physical states of the tissue. It is influenced by
the atoms to which they are attached or to which they are proximal.
H2O. Hydrogen protons in the small H2O molecule have higher natural
motional frequencies than, for example, hydrogen protons in a solid
structure. The natural motional frequency of hydrogen protons in water
is also much faster than the Larmor frequency for hydrogen.

ω (H2O) ω0
Solids. Hydrogen protons in solids have lower natural motional
frequencies than do water protons. The natural motional frequencies of
hydrogen protons in solids are somewhat slower than the Larmor
frequency for hydrogen.
ω (solids) < ω0

Figure 6-1. Hydration layer water.

Fat. Hydrogen protons in fat have natural motional frequencies that are
almost equal to the Larmor frequencies used for MRI.
ω (fat) ≈ ω0
This result is caused by the rotational frequency of the carbons around
the terminal C-C bond. Because this frequency is near the Larmor
frequency, the efficiency of energy transfer from the protons to the
lattice or from the lattice to the protons is increased, thus decreasing
 T1.
Proteinaceous Solutions. The foregoing discussion on the T1 and T2
characteristics of fluids such as water applies only to pure water (or
bulk phase water). However, most of the water in the body is not in
the pure state but is bound to a hydrophilic macromolecule such as a
protein.
Such water molecules form hydration layers around the macromolecule
and are called hydration layer water (Fig. 6-1). These bound H2O
molecules lose some of the freedom in their

motion. As a result, the natural motional frequencies of the H2O

molecules get closer to the Larmor frequency, thus yielding a more
efficient energy transfer. The net result is a shortening in the T1
relaxation. Therefore, proteinaceous fluids, that is, hydration layer
water, are brighter than pure water on T1-weighted images.
If the protein content is high enough, hydration layer water can cause
some T2 shortening. This shortening is generally seen in gels and in
mucinous fluids. Such proteinaceous fluids may be darker than pure
fluids on T2-weighted images.
For H2O and solid tissue, the energy transfer is not efficient, and the T1
for H2O and solid tissue is long. Also, the T1 for H2O is longer than the
T1 for solid tissue because the difference between the Larmor
frequency and the natural motional frequencies of hydrogen protons in
H2O is much greater than the difference between the Larmor frequency
and the motional frequencies of hydrogen protons in solid tissue.
Let's now draw the T1 and T2 curves for these different tissues (Fig. 6-
2):
1. Fat has the shortest T1 and will have the steepest T1 recovery curve.
2. Proteinaceous fluid also has a short T1.
3. H2O has the longest T1 and will have the slowest T1 recovery curve.
4. Solid tissue has intermediate T1.
For the sake of argument, we'll assume that they all have the same
 proton density. Actually, the proton density of H2O is higher because
there are more hydrogen protons per volume in water than either fat or
solid tissue, and intensity is based not only on T1 and T2 but also on the
proton density N(H):

Figure 6-2. T1 recovery curves of fat, water, and a solid tissue.

SI ∝ N(H) (e-TE/T2) (1 - e-TR/T1)

At a time TR, we transmit another radio frequency (RF) pulse. Let's
superimpose the T2 decay curve on the T1 recovery curve (Fig. 6-3):
1. H2O has a very long T2, so it will have a very shallow T2 decay curve.
2. Solid tissue has short T2 and will thus decay fairly rapidly.
3. Fat has an intermediate T2.
4. Proteinaceous fluid may have a short or intermediate T2 depending
on the protein content.
 Therefore, we see that if we pick a long enough TE (TE3) in Figure 6-3,
the signals that we measure from each tissue show the following:
1. H2O has the highest signal intensity (point a, Fig. 6-3).
2. Solid tissue has the lowest signal intensity.
3. Fat has an intermediate signal intensity.
4. Proteinaceous fluid has an intermediate or low signal intensity
depending on its protein content.
If we take a shorter TE (TE2), we might pick a point where fat and H2O
might have the same signal intensity. This is a crossover effect (point b,
Fig. 6-3).

Figure 6-3. T2 decay curves of fat, water, and a solid tissue.

 If the TE is really short (TE1), we just get a T1 or proton density effect
(it depends on the TR), where
1. Fat has the highest intensity (point c, Fig. 6-3).
2. Proteinaceous fluid also has high intensity similar to fat.
3. Solid tissue has intermediate intensity.
H2O has the lowest intensity.

So, we can see from the curve that

1. If TR and TE are short, we get T1 weighting.
2. If TR and TE are long, we get T2 weighting.
3. If TR is long and TE is short, we get proton density weighting.
Let's now look at three different tissues in the brain: (i) gray matter, (ii)
white matter, and (iii) cerebrospinal fluid (CSF; Fig. 6-4). On a T1
recovery curve:
Figure 6-4. T1 recovery curves of CSF, white matter, and gray
matter.
 Figure 6-5. T2 decay curves of CSF, white matter, and gray matter.

1. White matter is bright. The myelin sheath acts like fat; with more
efficient energy exchange, it has a shorter longitudinal relaxation
than does gray matter.
2. Gray matter is intermediate: without myelin, it acts more like a
typical solid tissue.
3. CSF is dark: like water, it has inefficient energy exchange and thus
the same long longitudinal relaxation, T1.
Let's add the T2 decay curves to the T1 recovery curves (Fig. 6-5):
1. CSF, like H2O, has the least dephasing, and thus the longest T2.

2. White matter has a slightly shorter T2 than gray matter.

If we use a long TE (TE3), then we'll get a typical T2-weighted image.
Therefore, at TE = TE3, we have (Fig. 6-6):
 Let's pick a shorter TE = TE2 (Fig. 6-6). At this point, white matter and
CSF are isointense (crossover point). We want to achieve this
isointensity on a proton density image. We can see the advantages by
considering what tumors and demyelinating plaques do on a T1
recovery curve, or on a T2 decay curve. Most pathologic lesions have a
slow T1 recovery curve because of their vasogenic edema (which
contains H2O). However, their T1 recovery curve is not as slow as pure
water. Most pathologic lesions (e.g., tumor, edema, multiple sclerosis
[MS] plaque) also have a long T2 but not as long as that of CSF.
In Figure 6-6, we've included a T1 recovery curve and a T2 decay curve
for a pathologic lesion. If we are looking for MS plaques, let's first look
at a T2-weighted image (long TE = TE3 in the graph):

1. White matter is dark.

2. CSF is bright.
3. MS plaque is also bright.
 Figure 6-6. Recovery and decay curves of CSF, WM, GM, and a
lesion.

Even though brightness may be different between the CSF and MS

plaque, the ratio is not great enough to discern a difference (e.g., the
lesion is adjacent to a lateral ventricle).
If we now look at the intensities at a shorter TE (TE2) corresponding to
the CSF and white matter crossover point, then CSF and white matter
will be isointense.
The pathologic lesion (e.g., MS plaque) will be brighter than both CSF
and white matter, and it can thus be detected more easily. Remember
also that if we choose a long TR and a very short TE (TE1 on the graph),
the TE occurs before the crossover points of either CSF, gray matter, or
white matter, resulting still in a proton density-weighted image.
This is a good time to bring up the proton density factor: N(H). We've
been, to a certain extent, ignoring it. We've talked about T1 and T2,
and we've been assuming that all the tissues have almost the same
proton density. However, in Table 6-1, we can see the differences in the
proton densities of various tissues. Also, T1 is defined as the time when
63% of the longitudinal magnetization has recovered (with 3T1 = 95%
recovery), whereas T2 is defined as the time when 63% of the
 transverse magnetization has decayed (with 3T2 = 95% decay).
For instance, if the CSF has a proton density of 1 (or 100%), then white
matter has a proton density of 0.61 (61% of CSF) and edema has a
proton density of 0.86 (86% of CSF). How does this difference in proton
density affect the

graphs of T1 and T2? Let's talk about two different tissues (Fig. 6-7):

Table 6-1 T1, T2, and proton density of brain tissues at

1.5 Ta

T1 (msec ) T2 (msec ) N(H)

White matter 510 67 0.61

Gray matter 760 77 0.69

Edema 900 126 0.86

CSF 2350 180 1.00

aStark and Bradley, p.44 .

1. CSF
2. White matter
CSF has a higher proton density than white matter, so it has a higher
maximum limit on the T1 recovery curve. White matter has a lower
proton density than CSF, but its T1 is shorter. The two recovery curves
cross at the point where white matter and CSF have the same intensity
(TR ≈ 2500 msec).
 For the mathematically interested reader, this TR is
the solution to the following equation: 1.0 (1 − e-
TR/2650
) = 0.61 (1 − e-TR/510) or e-TR/2650 −0.61 e-TR/510
−0.39 = 0 using the T1 and N(H) values for white
matter and CSF from Table 6-1, resulting in a TR of
approximately 2500 msec (2462 msec to be exact!).

Figure 6-7. The plateau of the recovery curve of a tissue is

determined by the proton density of that tissue N(H). For instance,
N(CSF) is larger than N(WM).

Let's now consider two situations:

1. Short TR
2. Long TR
1. First, draw the T1 recovery curves for white matter and CSF (Fig. 6-
8). Now consider a short TR (say, 300 msec). White matter is initially
brighter than CSF because of its shorter T1. However, CSF has a longer
T2 than white matter. Therefore, after the T2 crossover point, CSF will
become brighter than white matter (e.g., at TE2). Thus, at long TE, we
get T2 contrast. If we pick a short TE (TE1), we get T1 contrast. Thus,
with a short TR, which maximizes T1 contrast, we want to choose as
short a TE as possible to maximize the T1 contrast.

Figure 6-8. Recovery and decay curves for WM and CSF for a short
TR.
 T1W: Short TR/Short TE

2. Now, draw the T1 and T2 curves again and this time pick a long TR.
Remember that CSF has a greater proton density than white matter, so
it will have a higher plateau value than white matter, which is brought
out by the long TR (Fig. 6-9). Then draw the T2 decay curves, keeping
in mind that CSF has a longer T2 than white matter. If we now pick a
very short TE (TE1), the two signals are driven by their respective
proton densities: CSF will have greater intensity than white matter
(i.e., 39%; Table 6-1). At this point, the difference in intensity reflects
their (true) proton density differences (assuming a very short TE).

Figure 6-9. Recovery and decay curves of WM and CSF for a long
TR.

PDW: Long TR/Short TE

If TE is long (TE2), increase the signal intensity differences between
white matter and CSF. This increased intensity difference reflects the
T2 difference.

T2W: Long TR/Long TE

Let's now introduce an abnormality— namely, edema—and incorporate it

with CSF and white matter (Fig. 6-10). We know that the T1 recovery
curve for edema is in between CSF and white matter—it has a shorter
T1 than CSF and a longer T1 than white matter. We also know that the
plateau for edema is less than that for pure CSF, but more than that for
white matter. Again, if we choose a TR that is long enough for white
matter to reach its plateau, and then choose a TE that is short (either
at or

before the crossover point for CSF and white matter T2 decay), then
edema has the highest signal intensity.
 Figure 6-10. Recovery and decay curves of WM, CSF, and edema for
a long TR.

Therefore, for “proton density” images (long TR/short TE):

1. Edema is bright.
2. CSF and white matter are isointense.
Now, pick a TR at the point of intersection of the T1 recovery curves for
CSF and white matter—a point at which white matter has almost
reached its peak intensity, but CSF has not (similar to the previous
graphs, but TR is now longer in order to reach the crossover point; Fig.
6-11). Now, apply the 90° pulse, and follow the T2 decay curves. With a
short TE, CSF is brighter than white matter. With a long TE, CSF is still
brighter than white matter, but the difference in brightness gets
magnified. On the long TR/short TE image, the difference in intensity
reflects only the differences in proton densities between the two
tissues, whereas the long TR/long TE image incorporates both proton
densities and T2 differences between the two tissues.
Parenthetically, in a true protein density image (as in Fig. 6-9), CSF or
H2O has the highest signal (because water has more protons than any
other tissue). Therefore, to minimize the T1 and T2 influences on what
should be a true proton density-weighted image, we need to make the
TR long enough to allow the T1 recovery curves to reach their plateaus
and then make the TE short enough to minimize T2 decay. (Actually,
this may not be a desirable image because lesions and normal fluid may
be indistinguishable.)

Figure 6-11. Recovery and decay curves of WM and CSF for a TR

corresponding to the crossover point of CSF and WM.
Key Points
Table 6-2 summarizes the T1 and T2 properties of
three tissues: water, solids, and fat/proteinaceous
material. Table 6-3 contains the relative T1 and T2
values (short, intermediate, or long) for several
tissues. Figures 6-12 through 6-20 show examples of
tissue contrast.

Table 6-2 T1 and T2 as a function of natural motional

frequencies (ω) vs. the Larmor frequency (ω0) for different
tissues

Fat and
Proteinaceous
H2O/Fluids Solids Material

T1 ω ω0 ω < ω0 ω ≈ ω0

Inefficient Inefficient
energy energy Efficient energy
transfer transfer transfer

Very Long T1 Long T1 Short T1

Less Most Intermediate

T2 dephasing dephasing dephasing

Long T2 Short T2 Intermediate T2

 Table 6-3 Relative T1 and T2 values for several tissuesa

Long T1 (low Short T1

SI) Intermediate (high SI)

Long T2 Water/CSF d (EC

(high SI) Pathology metHgb)
Edema

Intermediate Muscle
GM a
(oxyHgb)
WM

Short T2 Air Fat

(low SI) Cortical bone Proteinaceous
Heavy Ca++ solution
b (deoxyHgb) c (IC met
e Hgb)
(hemosiderin) Paramagnetic
Fibrosis materials
Tendons (Gd, etc.)

a
a-d represent breakdown products of hemoglobin (a,
oxyhemoglobin; b, deoxyhemoglobin; c, intracellular
methemoglobin; d, extracellular methemoglobin; e,
hemosiderin). Abbreviations: GM , gray matter; WM , white
matter; SI, signal intensity; Hgb, hemoglobin; IC, intracellular;
EC, extracellular.
 P.66

Figure 6-12. Axial T1 (A), proton density (B), and fast spin echo
T2 (C) images of the brain. Note that on the T1 image the white
matter is brighter than the CSF due to the shorter T1 of white
matter. However, the CSF is brighter on the proton density image
due to its higher proton density. The white matter becomes even
darker on the T2 compared with the CSF secondary to additional
T2 differences. The arrow points to a subarachnoid hemorrhage,
whereas the arrowhead points to a small subdural hemorrhage.
Note that the hemorrhage is isointense to CSF on the T2 due to a
crossover point being achieved, whereas the hemorrhage is
brighter on the T1 and on the proton density-weighted images
due to the shorter T1 versus CSF.

P.67
Figure 6-13. Axial T1 (A), proton density (B), and T2 (C) images
of the brain again show the relative signal differences in normal
structures between the three different sequences. The arrow
points to a small intraventricular meningioma that has typical
signal close to gray matter on all sequences.
P.68
Figure 6-14. Sagittal T1 (A), proton density (B), and T2 (C)
images of the lumbosacral spine show dark signal in the L5 and
S1 vertebral bodies on the T1 sequence due to the longer T1 of
edema compared with the shorter T1 of the fatty marrow at the
other levels. The CSF and the intervertebral discs are normally
dark on T1 due to fluid and/or dessication. The proton density
and T2 images show the L5/S1 disc and a few others to be
bright, as is the CSF. Note that the CSF is brightest on the T2.
The combination of both abnormal bone marrow signal and the
adjacent bright disc represent osteomyelitis and discitis,
whereas bright discs at other levels with adjacent normal
marrow represent hydrated discs.

P.69
Figure 6-15. Sagittal T1 (A), proton density (B), and T2 (C)
images of the lumbosacral spine show normally bright signal in
all the vertebral bodies through L4; however, the L5 vertebral
body and the sacrum have very bright signal on all sequences.
The proton density and T2 images were acquired with a fast spin
echo technique. This patient had radiation therapy for cervical
cancer with subsequent complete fatty replacement of the
marrow at L5 and the sacrum as opposed to the normal fatty-
containing marrow at other levels.

P.70

Figure 6-16. Axial T1 (A) and T2 (B) images demonstrate a large,

right basal ganglia acute hypertensive hemorrhage (isointense on
T1 and bright on T2—oxyhemoglobin). There is a rim of bright T1
and dark T2 signal (arrows) that represents a characteristic rim of
the more temporally advanced intracellular methemoglobin.
Figure 6-17. Axial T1 (A) and T2 (B) images demonstrate an
acute (dark on T1 and T2—deoxyhemoglobin) left medial
temporal intraparenchymal hematoma (arrows). Additional axial
T1 (C) and T2 (D) images in the same patient at another level
show a late subacute (bright on both T1 and T2—extracellular
methemoglobin) right occipital hematoma (arrows) and acute
(isointense on T1 and dark on T2) left medial temporal hematoma
(arrowhead—best seen in image D). This patient had amyloid
angiopathy.

P.71
Figure 6-17. (continued)

Figure 6-18. Axial T1 (A) and T2 (B) images show an early

subacute epidural hematoma (bright on T1 and dark on T2—
intracellular methemoglobin) along the right frontal lobe
(arrows).

P.72
Figure 6.19. Axial proton density (A) and T2 (B) images
demonstrate superficial gyriform dark signal from superficial
siderosis (hemosiderin) in a patient with a history of
subarachnoid hemorrhage (arrows).

P.73
 Figure 6-20. Axial T1 with fat saturation without gadolinium (A)
and fast spin echo T2 (B) images show bright T1 and relatively
dark T2 signal in the lumen of the small bowel consistent with
proteinaceous solutions (arrows). Additionally, the patient has a
right endometrioma (arrowhead) with predominantly bright T1
and dark T2 signal secondary to recurrent hemorrhage.

Questions
6-1 T/F Hydration layer water has a shorter T1 than bulk water.

6-2 Match (i) short T1 and T2; (ii) short T1, long T2; (iii) long T1,
short T2; and (iv) long T1, long T2 with
(a) air
(b) fat
(c) water
(d) intracellular methemoglobin
(e) extracellular methemoglobin

6-3 T/F The most efficient energy transfer occurs at the Larmor
frequency.

6-4 Match the following: (i) short TR and TE; (ii) long TR and TE; (iii)
short TR, long TE; (iv) long TR, short TE with
(a) T1 weighted
(b) T2 weighted
(c) intermediate weighted

1Translation, rotation, and vibration.

7
Pulse Sequences: Part I (Saturation,
Partial Saturation, Inversion Recovery)

Introduction
A pulse sequence is a sequence of radio frequency (RF) pulses applied
repeatedly during an MR study. Embedded in it are the TR and TE time
parameters. It is related to a timing diagram or a pulse sequence
diagram (PSD), which is discussed in Chapter 14. In this chapter, we
discuss the concepts of saturation and consider pulse sequence partial
saturation, saturation recovery, and inversion recovery (IR). In the next
chapter, we'll talk about the important spin-echo pulse sequence.
Figure 7-1 illustrates the notations used for three types of RF pulses
throughout this book.

Saturation
Immediately after the longitudinal magnetization has been flipped into
the x-y plane by a 90° pulse, the system is said to be saturated.
Application of a second 90° pulse at this moment will elicit no signal
(like beating a dead horse). A few moments later, after some T1
recovery, the system is partially saturated. With complete T1 recovery
to the plateau value, the system is unsaturated or fully magnetized.
Should the longitudinal magnetization only be partially flipped into the
x-y plane (i.e., flip angles less than 90°), then there is still a
component of magnetization along the z-axis. The spins in this state are
also partially saturated.
 Figure 7-1. The notation for 90°, 180°, and partial flip pulses.

Partial Saturation Pulse Sequence. Start with a 90° pulse, wait for a
short period TR, and then apply another 90° pulse. Keep repeating this
sequence. The measurements are obtained immediately after the 90°
RF pulse. Therefore, the signal received is a free induction decay (FID).
Let's see how this looks on the T1 recovery curve (Fig. 7-2). At time t =
0, flip the longitudinal magnetization 90° into the x-y plane. Right after
that, the longitudinal magnetization begins to recover. Wait a time t =
TR, and repeat the 90° pulse. Initially, at time t = 0, the longitudinal
magnetization is at a maximum. As soon as we flip it, the longitudinal
magnetization goes to zero and then immediately thereafter begins to
grow. At time t = TR, the longitudinal magnetization has grown but has
not recovered its plateau before it is flipped into the x-y plane again.
(Note that the length of the longitudinal magnetization vector before
the second 90° RF pulse is less than the original longitudinal
magnetization vector.)
Now, with a third 90° RF pulse, we again flip the longitudinal
magnetization into the x-y plane. Again, the longitudinal magnetization
goes to zero and immediately begins to recover.
Again, at time 2TR, it is less than maximum but is equal to the previous
longitudinal magnetization (at time TR). Each subsequent recovery time
TR after each subsequent 90° pulse will also be the same. Thus, the
maximum FID occurs at time t = 0 after the first 90° RF pulse, and all
subsequent FIDs will have less magnitude but will have the same value.

Figure 7-2. The recovery curves following the RF pulses in a partial

saturation sequence.

Question: Is there a residual transverse magnetization Mxy at time TR

just before the next 90° RF pulse?
Answer: No! Because T1 is several times larger than T2, after a time
TR has elapsed, the magnetization in the x-y plane has fully decayed to
zero.
In partial saturation, TE is minimal. The signal is measured immediately
after the 90° RF pulse.
Partial saturation: TR is short, TE is minimal.
Question: With short TR and minimal TE, what kind of an image would
we be getting?
Answer: T1-weighted image.
A partial saturation pulse sequence generates a T1-weighted image.

Saturation Recovery Pulse Sequence

The previous sequence is called partial saturation because at the time
of the second 90° RF pulse (at time TR), we haven't yet completely
recovered the longitudinal magnetization. Therefore, only a portion of
the original longitudinal magnetization (M0) is flipped at time TR (and
subsequent TRs). Hence the name partial saturation.
In saturation recovery, we try to recover all the longitudinal
magnetization before we apply another 90° RF pulse. We have to wait a
long time before we apply a second RF pulse. Thus, TR will be long (Fig.
7-3).
After each 90° RF pulse, we measure it and an FID right away. Because
we allow the longitudinal magnetization to recover completely before
the next 90° pulse, the FID gives the maximum signal each time. In
other words, we have recovered from the state of saturation.
In saturation recovery, TR is long and TE is minimal.
Question: With long TR and minimal TE, what kind of image do we get?
Answer: Proton density weighted (PDW).
 Figure 7-3. In a saturation recovery sequence, TR is long and
longitudinal magnetization vectors are near maximal.

The saturation recovery pulse sequence results in a proton density-

weighted image.
Neither of these sequences is really used any more, but they are so
simple to understand that they are good springboards from which to
learn about other, more complex pulse sequences. These pulse
sequences are not used because it is very difficult to measure the FID
without a delay period. Electronically we have to wait a certain period
of time to make the measurements. Also, external magnetic
inhomogeneity becomes a problem; that's why spin-echo sequences
(which we will discuss in the next chapter) are used to eliminate this
problem.

Inversion Recovery Pulse Sequence

In IR, we first apply a 180° RF pulse. Next, we wait a period of time
(the inversion time TI) and apply a 90° RF pulse. Then we wait a period
of time TR (from the initial 180° pulse) and apply another 180° RF
pulse (Fig. 7-4), beginning the sequence all over again.

Figure 7-4. In inversion recovery, the time between the 180° pulse
and the 90° pulse is denoted TI.

Before we apply the 180° pulse, the magnetization vector points along
the z-axis. Immediately after we apply the 180° pulse, the
magnetization vector is flipped 180°; it is now pointing south (−z),
which is the opposite direction (Fig. 7-5).
We then allow the magnetization vector to recover along a T1 growth
curve. As it recovers, it gets smaller and smaller in the −z direction
until it goes to zero, and then starts growing in the +z direction,
ultimately recovering to the original longitudinal magnetization.
After a time TI, we apply a 90° pulse. This then flips the longitudinal
magnetization into the x-y plane. The amount of magnetization flipped
into the x-y plane will, of course, depend on the amount of longitudinal
magnetization that has recovered during time TI after the original 180°
RF pulse. We measure this flipped magnetization. Therefore, at this
point we get an FID proportional to the longitudinal magnetization

flipped into the x-y plane. Also, at this point, we begin the regrowth of
the longitudinal magnetization. Recall that for a typical T1 recovery
curve, the formula for the exponential growth of the curve is 1 − e−t/T1
However, when the magnetization starts to recover from −M0 instead of
zero (Fig. 7-6), the formula for recovery is 1 − 2e−t/T1

Figure 7-5. The recovery curves in inversion recovery. After the

180° pulse, the longitudinal magnetization vector is flipped 180°
and starts to recover from a value that is the negative of its initial
maximal value.

Exercise
Verify the above formula mathematically.
At time t = 0, Signal intensity (SI) = 1 − 2e−0/T1 = 1 − 2(1) = − 1 So at
time t = 0,
Magnitude of signal intensity = − 1.
 At t = ∞ (infinity), Signal intensity = 1 − 2e−∞/T1 = 1 − 2(0) = + 1
So at time t = ∞, the signal is maximal. These values correspond to the
graph in Figure 7-6.
Null Point. The point at which the signal crosses the zero line is called
the null point. At this point, the signal intensity is zero. The time at
this null point is denoted TI(null). We can solve the equation
mathematically for TI(null), at which point the signal intensity is zero:
Signal intensity = 0 = 1 − 2e−T1/T1
The solution to this equation is (see Question 7-1 at the end): TI(null) =
(loge2) T1 = (ln 2) T1 ≈ 0.693 T1
Let's go back and re-examine the recovery curves. Actually, there are
two different exponentially growing curves occurring sequentially (Fig.
7-7):
1. Recovery after the 180° RF pulse
2. Recovery after the 90° RF pulse
1. The T1 recovery curve following the 180° pulse starts at −M0 and
grows exponentially according to the formula: M0 (1 − 2e−T1/T1)
 Figure 7-6. The recovery curve in IR is given by the formula 1 −
2e−t/T1.

Figure 7-7. There are two recovery curves in one IR cycle.

2. The T1 recovery curve following the 90° RF pulse, after the

longitudinal magnetization, flips into the x-y plane, starts at 0, and
grows exponentially according to the formula: M0 (1 − e−TR/T1)

If we combine both of these T1 recovery curves, we get a relationship

that combines both T1 and TR together. The result is the product of the
above two formulas: SI ∝ M0 (1 − 2e−T1/T1)(1 − e−TR/T1)
Assuming that TI « TR, the product of the terms within parentheses can
be simplified to (see Question 7-2 at the end): (1 − 2e−T1/T1) + (e−TR/T1)
Clinical Applications of Inversion Recovery. In an IR pulse sequence,
we start out with a 180° RF pulse followed, after a time interval TI, by
a 90° RF pulse. Next, after a certain time interval TR, the sequence is
repeated with another 180° pulse.
TI = inversion time, which represents the time interval between the
180° pulse and the 90° pulse
TR = time interval between successive 180° pulses (or successive 90°
pulses)
Consider graphically what happens to two tissues: edema and white
matter (Fig. 7-8). In IR, we first flip the longitudinal magnetization
180° with a 180° RF pulse. Subsequently, the magnetization vector still
runs along the z-axis but points in the negative (south) direction. Next,
the longitudinal magnetization vectors begin to grow according to the
T1 growth curves of edema and white matter. Edema has a greater
proton density than white matter, so its maximum magnetization along
the z-axis will be greater than the maximum for white matter.
Likewise, after they have flipped 180°, the T1 recovery curve for
edema begins lower, that is, it is more negative along the z-axis than is
white matter.
From this initial position along the z-axis, the longitudinal
magnetization for edema grows along its T1 recovery curve until it
reaches its maximum. It starts decreasing in the negative z direction
until it reaches zero (the null point), and then it continues increasing in
the positive z direction until it reaches its maximum. The T1 growth
curve for white matter, because of its lower proton density, starts
closer to zero on the negative z-axis than edema after the 180° flip.
Because of its shorter T1, it recovers more rapidly along its T1 curve
than does edema to reach its maximum.
At time TI, a 90° excitation pulse is applied. Edema and white matter
will have recovered their longitudinal magnetization at different rates
depending on their individual T1s. When both longitudinal
magnetization vectors are flipped into the x-y plane by the 90° pulse,
they generate FID signals. Immediately after the 90° pulse, each
longitudinal magnetization vector goes to zero but begins recovering
according to its T1 recovery curve. Then, after a time interval TR, the
process is repeated with another 180° inverting pulse.
We saw earlier that the T1 growth curve after the 180° pulse is given
by the formula: M0 (1 − e−TR/T1)(1 − 2e−T1/T1)

Figure 7-8. Recovery curves for WM and edema.

Magnitude Reconstruction. Magnitude reconstruction is another

variable in IR. If we want to increase the signal-to-noise ratio by about
40% (more precisely, by a factor of ), we can add the x and y
channels of the coil together as the root mean square (rms), that is,
.
This gives us a magnitude image, which is always positive. It appears
like the mirror image of the negative growth curves, flipped about the
time axis (Fig. 7-9). The dashed lines going from the positive z-axis,
down to the zero point, are actually the mirror image of the two T1
growth curves of edema and white matter, “flipped” so that we only
register their magnitude, not their positive or negative phase. This new
method of displaying the IR process is called magnitude
reconstruction. Although it has more signal to noise than the
original phase construction, its dynamic range is less than the original,
that is, 0 to M0 versus −M0 to M0. Thus, magnitude reconstruction is
used whenever the signal-to-noise ratio is limited, and phase
reconstruction is used when greater contrast is needed.
In IR, TR is always long. By picking a long TR, we reach a steady state
with maximum value on the T1 growth curve of each tissue after the
90° pulse. What happens when TI is chosen to null white matter (Fig. 7-
9)? If we consider just the magnitude of the signal, which is the
distance of the T1 recovery curve above or below the time line, then
edema has a greater magnitude than white matter. Because edema has
a longer T2 than white matter, if we choose a long TE, then we will
magnify the difference in intensity between the two tissues. In fact,
the longer the TE, the greater the contrast difference will be between
edema and white matter (Fig. 7-10).

Fat Suppression: STIR Imaging

STIR stands for short TI (or Tau) inversion recovery. Let's draw two T1
recovery curves, after the 180° RF pulse, for two tissues—fat and H2O
(Fig. 7-11). Pick the TI at the point where fat crosses the zero point.
(The null point is equal to ln 2 [or 0.693] multiplied by the T1 of fat
[see Question 7-1 at the end].)
Figure 7-9. Recovery curves, magnitude recovery curves (mirror
image curves to make everything positive), and associated decay
curves for WM and edema.
Figure 7-10. Tissue contrast for two different TEs.
 Figure 7-11. In the STIR fat suppression technique, TI is chosen so
that the T1 recovery curve for fat crosses zero at the time of the
90° pulse.

At this null point for fat, if we draw the T2 decay curves, fat starts at
zero and will stay at zero. There will be no transverse magnetization
from fat in the x-y plane, and water will have its usual T2 decay curve.
In effect, we have suppressed the fat signal. Therefore, after a 180°
inverting pulse, we wait a time TI = 0.693 T1 (fat) and we give the 90°
pulse. All other tissues will have longitudinal magnetization that will
flip into the x-y plane and give off a signal according to their T2 curves.
However, at its null point, fat will not have any longitudinal
magnetization to flip into the x-y plane and thus will not have any
signal.
The term STIR is called short TI inversion recovery because fat has a
very short T1; therefore, a very short TI must be chosen to null it (at
high field [1.5 T] this TI is 140 msec, whereas at midfield [0.5 T] it is
100 msec). Fat will reach its null point before white matter, gray
matter, H2O, or edema (Fig. 7-11).

Key Points
We have discussed three types of pulse sequences:
saturation recovery, partial saturation, and inversion
recovery (IR). The latter is very important because it
allows suppression of any tissue by selecting TI to be
0.693 times the T1 of that tissue: TI(null) = 0.693 ×
T1 This subject is further elaborated in Chapter 25 on
tissue suppression techniques.
A partial saturation sequence results in T1 weighting
(short TR and TE). A saturation recovery, however,
results in PD weighting (long TR, short TE).

Questions
7-1
(a) Given an inversion recovery (IR) pulse sequence (Fig. 7-6),
prove that TI that “nulls” or “suppresses” a certain tissue is
equal to 0.693 × T1 (tissue), that is, TI(null) = 0.693 × T1

Hint: The IR curve is proportional to SI ∝ 1 − 2e−t/T1, where t =

TI.
(b) Assuming a T1 = 180 msec for fat, what TI would “suppress”
the fat?

7-2 Consider the inversion recovery (IR) pulse sequence shown in

Figure 7-12. Prove that the signal measured after each 90° pulse
(i.e., at A, A′) is given by N(H) (1 − 2e−TI/T1 + e−TR/T1) assuming that TI
is much smaller than TR (i.e., TI « TR).
7-3 Match
(i) partial saturation
(ii) saturation recovery with
(a) PD weighted
(b) T1 weighted

7-4 T/F In an IR sequence, a 180° pulse is followed TI millisecond

later by a 90° pulse.

Figure 7-12.
8
Pulse Sequences Part II (Spin Echo)

Introduction
This chapter focuses on the most frequently used pulse sequence—the
spin echo (SE) pulse sequence. When the concept of dephasing was
discussed in previous chapters, we brought up two main causes: (i)
external magnetic inhomogeneity, and (ii) inherent spin-spin
interactions. The SE pulse sequence eliminates the former by an
additional refocusing or rephasing 180° radio frequency (RF) pulse. By
using the SE pulse sequence, we can eliminate dephasing caused by
fixed external magnetic field inhomogeneities. (We can't eliminate
spin-spin interactions because they are not fixed, i.e., they fluctuate
randomly.)

Spin-Echo Pulse Diagram

As a result of the 90° pulse, the magnetization vector Mz is flipped into
the x-y plane. Consider the precession of three different magnetization
vectors in the transverse plane, each in a slightly different magnetic
environment (Fig. 8-1A). Initially, all these vectors are in phase and
they are all precessing at frequency ω0.
In Figure 8-1A, say one group of spins is exposed to the magnetic field
B0, which causes them to precess at frequency ω0. The adjacent group
of spins sees a slightly higher field B0 + and precesses at a slightly
higher frequency ω0 +, and another sees a slightly lower field B0 − with
a precessional frequency of ω0−. After the 90° pulse, the three spins
will begin to get out of phase with each other (Fig. 8-1B). Eventually,
the fast vector and the slow vector become 180° out of phase and
cancel each other out (Fig. 8-1C).

Analogy
Let's consider the analogy of three runners running around the track
(Fig. 8-2). Initially, they start out at the same point. After they run for
a time τ, they are no longer together—one is running faster and gets
ahead of the others, and one is running slower, falling behind the
others.
At this time, if we make the runners turn around and run the opposite
way, each one will still be running at the same speed (precessing at the
same frequency in the case of the spins). They have just changed
direction and are running back to where they started. Each one will
then run the same distance if they run the same amount of time τ.
Therefore, at time 2τ, they will all come back at the starting point at
the same time and will be back together in phase. The action of making
the runners change direction is done with the use of a 180° refocusing
pulse in the case of the spins.
Thus, at a certain time τ after the 90° pulse, when the spins have
gotten out of phase, a 180° pulse is applied. Now all the spins flip 180°
in the x-y plane and they continue precessing, but now in the opposite
direction (Fig. 8-3). Let's look at the pulse sequence diagram (Fig. 8-4).
We start off with a 90° RF pulse to flip the spins into the x-y plane. We
wait a time τ and apply a 180° RF pulse. Then we wait a long time, TR,
and repeat the process.
If we draw the free induction decay (FID) after the 90δ pulse, we see
that the FID dephases

very rapidly due to the T2* effect related to external magnetic field
inhomogeneities and spin-spin interactions. The spins get out of phase.
After time τ we apply the 180δ refocusing pulse. After an equal time τ
they will be completely in phase again, and the signal will reach a
maximum.
 Figure 8-1. Three magnetization vectors in three slightly magnetic
environments. In (A) they are in phase and their vector sum is
three times each individual vector. In (B) they are slightly out of
phase, yielding a smaller net vector. In (C) vector 1 and 3 cancel
each other out because they are 180δ out of phase, leaving only
vector 2.

Figure 8-2. Analogy of three runners on a track. At time τthey are

made to turn around and run back toward the starting point.
Because the slowest runner is now in the lead, they will all reach
the starting point at exactly the same time (at time 2τ).

1. Time τ is the time from the 90° RF pulse to the 180° RF pulse.
2. Time τ is also the time from the 180° RF pulse to the point of
 maximum rephasing, that is, the echo.
3. We call 2τ the echo delay time (time to echo)—TE: the time after
the 90° pulse when we get maximum signal again.
4. The 180° pulse is, therefore, called a refocusing or rephasing pulse.
We can apply a second 180° pulse. Now, instead of one 180° pulse
following the 90° pulse, we have two 180° pulses in sequence after a
90° pulse (Fig. 8-5). After the first echo, the spins will begin to dephase
again. A second 180° pulse applied at time τ2 after the first echo will
allow the spins to rephase again at time 2τ2 after the first echo and a
second echo is obtained. Each echo has its own TE.
1. The time from the 90° pulse to the first echo is TE1.
2. The time from the 90° pulse to the second echo is TE2.

Ideally, we would like to regain all the signal from the original FID. In
practice, it can't happen. We are able to regain the signal lost due to
fixed external magnetic field inhomogeneities by applying a refocusing
180° pulse, but dephasing caused by spin-spin interaction of the tissue
cannot be regained. If we join the points of maximum signal due to
rephasing as a result of the 180° pulses, we will get an exponentially
decaying curve with a time constant given by T2. Therefore, the decay
of the original FID and the decay of each subsequent echo is given by e
−t/T2*, whereas the decay of the

curve describing the maximum signal reached by each echo is given by

e−t/T2. This is the difference between T2* and T2.
Figure 8-3. The vectors in Figure 8-1 are reversed 180° in direction
at time τ (A), so that at time 2τ they'll get in phase again (B).
Figure 8-4. In a spin-echo pulse sequence, a 180° pulse is applied
at time τ, causing the spins to get in phase at time 2τ. This leads to
the formation of an echo from the FID.

Figure 8-5. An example of a dual-echo, spin-echo pulse sequence

in which two echoes are formed via application of two 180° pulses.
Symmetric Echoes
In Figure 8-5, if τ1 = τ2, then we get symmetric echoes.

Example
Take TR = 2000, and TE = 40 and 80 msec. Here, τ1 = 20, so that TE1 = 2
τ1 = 40, and TE2 = 80 = TE1 + 2τ2 = 40 + 2τ2; then 2 τ2 = 40 and τ2 = 20.
So τ1 = τ2 in symmetric echoes.

Asymmetric Echoes
If τ1 ≠ τ2, then we get asymmetric echoes.

Example
Take TR = 2000, TE = 30 and 80 msec.
Here TE1 = 2(τ1) = 30 msec, so τ1 = 15 msec.
TE2 = 80 msec = TE1 + 2 (τ2) = 30 msec + 2τ2 = 80 msec.
Then 2τ2 = 50 and τ2 = 25 msec.
So τ1 ≠ τ2 in asymmetric echoes.
Question: What does the 180° pulse do to the longitudinal
magnetization?
Answer: It inverts it. However, at time TE/2 (on the order of 10 msec),
the recovered longitudinal magnetization is negligible and its inversion
does not cause any significant signal loss. In fact, at t = TE/2, we have
Mz = M0 (1 - e-TE/2TR) ≅ 0
because TE/2 TR, so that e-TE/2TR ≅ 1.

Tissue Contrast
As discussed in Chapter 6, tissue contrast in SE depends primarily on TR
and TE. There are three types of tissue contrast:
1. T1 weighted (T1W)
2. T2 weighted (T2W)
3. Proton density weighted (PDW; also called “balanced,”
“intermediate,” and “spin density”)
Let's see what TR and TE must be for these three imaging scenarios
(Table 8-1):
1. For T1 weighting, we want to eliminate the T2 effect and enhance
the T1 effect.
a. To eliminate (reduce) the T2 effect, we want a short TE.
b. To enhance the T1 effect, we want a short TR.
c. The signal is then proportional to N(H) (1 - e-TR/T1).
2. For T2 weighting, we want to eliminate the T1 effect and enhance
the T2 effect.
a. To eliminate (reduce) the T1 effect, we want a long TR.

Table 8-1

TR TE Signal (Theoretical)

T1W Short Short N(H) (1-e−TR /T1 )

T2W Long Long N(H) (e−TE /T2 )

PDW Long Short N(H)

b. To enhance the T2 effect, we want a long TE.

 c. The signal is, therefore, proportional to N(H) (e−TE/T2).
3. For proton density weighting, we want to eliminate the T1 and T2
effects.
a. To eliminate (reduce) the T1 effect, we want a long TR.
b. To eliminate (reduce) the T2 effect, we want a short TE.
c. The signal is then proportional to N(H).

Remember that in practice we never totally eliminate any of these

factors. We would have to have an infinitely long TR to eliminate all T1
effects, and we would need a TE of 0 to eliminate all T2 effects.
Therefore, all T1-weighted images in practice have some T2 influence
(Fig. 8-6); all T2-weighted images have some T1 influence; and proton
density-weighted images have some influence from both T1 and T2.
This is why we use the terms T1 weighting, T2 weighting, and proton
density weighting:

Figure 8-6. An out-of-phase spoiled gradient echo T1 (A with TE

1.8 msec) and an in-phase spoiled gradient echo T1 (B with TE 4.2
msec) demonstrate bright gallstones in (A), but dark gallstones in
(B). This is due to the short T2* time of the gallstones with rapid
signal loss between 1.8 and 4.2 msec. Also note the signal loss of
the bile in the gallbladder on the out-of-phase image indicating a
mixed amount of fat/cholesterol signal and water signal.
1. We put more weight on the differences in T1 by shortening the TE
and the TR.
2. We put more weight on the differences in T2 by lengthening TR and
the TE.
3. We put less weight on T1 and T2 by lengthening TR and shortening
TE, thus giving more weight to proton density.

Key Points
1 The SE (spin echo) pulse sequence is composed of
a 90° excitation pulse followed by one or more 180°
rephasing pulses.
2 The purpose of the 180° pulse is to eliminate the
dephasing effects caused by external magnetic field
inhomogeneities by rephasing the spins at the time
of echo (TE).
P.89
3 The resultant echo then depends on T2 decay
rather than T2* decay, as seen with the FID (free
induction decay).
4 Table 8-2 summarizes the tissue contrast in SE
with respect to TR and TE.

Table 8-2

Short TE Long TE

Short TR T1W Mixed

Long TR PDW T2W

Questions
8-1 Consider a dual-echo SE sequence as in Figure 8-5:
(a) What are the received signals at the first echo and at the
second echo?
(b) What would the signal at TE1 be without a 180° refocusing
pulse?
(c) Calculate the ratio of the signals at point A without a
refocusing pulse to that with a refocusing pulse for TE1 = 25, TE2
= 50, T2 = 50, and T2* = 25 msec.

8-2 Match
(i) T1W (ii) T2W (iii) PDW with
(a) short TR and short TE
(b) long TR and short TE
(c) long TR and long TE

8-3 T/F
The 180° pulses totally eliminate the dephasing of spins in the
transverse plane.
9
Fourier Transform

Introduction
Fourier was an 18th century French mathematician. His picture, along
with the Fourier transform of his picture, is shown in Figure 9-1. The
Fourier transform (FT) is a mystery to most radiologists. Although the
mathematics of FT is complex, its concept is easy to grasp. Basically,
the FT provides a frequency spectrum of a signal. It is sometimes easier
to work in the frequency domain and later convert back to the time
domain.
Let's start by saying that we have a signal g(t), with a certain waveform
(Fig. 9-2). This signal is basically a time function, that is, a waveform
that varies with time. Now, let's say we have a “black box” that
converts the signal into its frequency components. The conversion that
occurs in the “black box” is the Fourier transform. The FT converts the
signal from the time domain to the frequency domain (Fig. 9-2). The FT
of g(t) is denoted G(ω). (The frequency can be angular [ω] or linear
[f].)
The FT is a mathematical equation (you don't have to memorize it). It is
shown here to demonstrate that a relationship exists between the
signal in the time domain g(t) and its Fourier transform G(ω) in the
frequency domain:

where ω = 2πf.
We are already familiar with the term (e−iωt) from Chapter 1. This is
the term for a vector spinning with angular frequency ω. The formula
integrates the product of this periodic function and g(t) with respect to
time. It also provides another function, G(ω), in the frequency domain
(Fig. 9-2).
One interesting thing about the Fourier transform is that the Fourier
transform of the Fourier transform provides the original signal. If the
Fourier transform of g(t) is G(ω), then the Fourier transform of G(ω) is
g(t):

FT provides the range of frequencies that are in the signal. Here are
some examples of functions and their Fourier transforms:

Example 1
The cosine function: cos (ω0t).
Obviously this signal (Fig. 9-3) has one single frequency. The frequency
could be any number. The Fourier transform is a single spike
representing the single frequency in the frequency domain (because of
its symmetry, we also get a similar spike on the opposite side of zero1).
The Fourier transform in this case tells us that there is only a single
frequency because it shows only one frequency spike at ω0 and is zero
everywhere else on the line. We can, for simplicity, ignore the
symmetric spike

on the negative side of zero and just consider the single spike on the
positive side to tell us that there is a single frequency (ω0). The spike
represents frequency and amplitude of the cosine function.
Figure 9-1. A: Picture of the French mathematician Fourier. B: The
magnitude of Fourier's two-dimensional Fourier transform. C: The
phase of his Fourier transform. (Reprinted with permission from
Oppenheim AV. Signals and systems. Prentice Hall; 1983.)
 Figure 9-2. The FT of g(t), designated G(ω).

Example 2
The sinc wave: sinc(ω0t) = sin(ω0t)/(ω0t)
The Fourier transform of this signal (Fig. 9-4) has a rectangular shape
and shows that the signal

contains, not just a single frequency, but a range of frequencies from

−ω0 to + ω0.

Figure 9-3. The FT of cos(ω0t) consists of two spikes, one at ω0

and one at −ω0.
 The bandwidth of this range of frequencies is from − ω0 to + ω0.
Bandwidth = ±ω0 = 2ω0 (We'll discuss bandwidth again later.) So the
Fourier transform tells us the range of frequencies that there are in a
signal, as well as the amplitude of the signals at those frequencies.
What's nice about the Fourier transform is that if we have the range of
frequencies and the amplitudes, we can reconstruct the original signal
back.

Example 3
Let's consider two frequencies:
1. cos ωt and
2. cos 2ωt (which is twice as fast as cos ωt)
The signal cos(2ωt) oscillates twice as fast as cos(ωt) (Fig. 9-5). If we
add them up, we get a complex signal (Fig. 9-6). If we were just given
the signal in Figure 9-6, we would have no idea that it is the sum of two
cosine waves.
Question: What is a good way to figure out the frequencies of which the
signal is composed?

Figure 9-4. The FT of a sinc function (sinc ω0t = sin ω0t/ω0t) has a
rectangular shape. The two ends of this rectangle are at ω0 and −
ω0 (where ω0 is the frequency of the sinc function).
 Answer: The Fourier transform of this complex signal (which we know is
the sum of two cosine waves, one twice as fast as the other) contains
two spikes, one twice as far from the origin as the other (Fig. 9-6). This
FT, then, demonstrates the composition of the signal in terms of its
frequencies.

Example 4
Let's now have a complex signal with two cosine waves, with the second
cosine wave not only twice as fast, but with twice the amplitude as
well. Again, by looking at the signal in Figure 9-7, we have no idea what
it is composed of, but by looking at the frequency spectrum (the FT of
the signal), we can tell the composition of the signal: in this case, two
separate cosine waves with differing frequencies and differing
amplitudes. The FT provides the frequency spectrum of a signal with its
amplitudes.

Example 5
Let's consider the FT of a sine wave: sin ωt. The FT of a sine wave is
different from the FT of a cosine wave:
1. FT of a cosine wave is symmetric with two symmetric spikes on either
side of zero (Fig. 9-8a).
Figure 9-5. The FT of cos 2ωt has two spikes: one at 2ω and one
at −2ω. (Here, only the positive frequencies are shown.)

Figure 9-6. The FT of cos ωt + cos 2ωt has two sets of spikes at
±ω and ±2ω. By looking at the FT signal, it is easy to figure out
its composition in the time domain.
Figure 9-7. The signal cos ωt + 2cos 2ωt and its FT. The FT has
again two sets of spikes: one at ±ω and one at ±2ω (but with
twice the magnitude).

Figure 9-8. A: FT of cos ωt. B: FT of sin ωt. Here the spikes have
opposite polarities and are also imaginary [iA in (B) as opposed to
(A)].
 Figure 9-9. sin ωt is an example of an odd function where the
value of the signal at −t is the negative of its value at +t.

2. FT of a sine wave is antisymmetric. It has a positive spike to the right

of zero and a negative spike to the left of zero2 (Fig. 9-8b).
The reason for this is that a sine function is an odd function. In other
words, if we take a time interval of t to the right of zero, the sine
value is positive, whereas, if we take a similar interval in the other
direction, the sine value is negative (Fig. 9-9). The cosine function,
however, is an even function. If we go a certain interval to the right or
left of zero, the cosine value is the same (Fig. 9-10). The FT of an even
function is real, whereas the FT of an odd function is imaginary. Thus,
cosine functions have FTs that are real (Fig. 9-8a). Sine functions have
FTs that are imaginary (Fig. 9-8b).

Fourier Transform Versus Fourier Series

There is a difference between a Fourier transform and a Fourier
series. Admittedly this is confusing. Let's talk about the original
 function g(t). This function can be represented by an infinite number of
sine and cosine waves:

What does this mean? Let's say that we have the rectangular function
(Fig. 9-11a). We said that this is composed of an infinite number of sine
and cosine waves:

Figure 9-10. cos ωt is an example of an even function where the

values at the signal at ±t are the same.

1. If we start with a single cosine wave, the signal will be as in Figure 9-

11b.
2. As we add sine and cosine, the signal will be as in Figure 9-11c.
3. As we continue to add sine and cosine, signal will be as in Figure 9-
11d.
4. The more cosine and sine we add, the more the signal approximates a
square wave (Fig. 9-11e).
It is impractical to go to infinity (∞). However, by eliminating the higher
frequencies, we get a “ring-down” effect.
If we do an FT of the Fourier series of g(t) above, we get a series of
spikes (Fig. 9-12). The envelope of this FT is the sinc wave.
The foregoing has emphasized temporal frequencies in cycles per
second or hertz. This would apply to spectroscopy but not imaging. For
imaging, we are interested in “spatial frequencies” with units of cycles
per millimeter. If image data g(r) is distributed within a 3D volume, it is
then possible to define the following complementary distribution G(k)
using a Fourier transform:
G(k) = ∫vg(r) exp [2 πi(k • r)]dr
Here k denotes a complex vector composed of three components and dr
is a volume element integrated over the volume V described by the
vector r = {x, y, z}. The vector k = {kx, ky, kz} (the origin of the term “k-
space”) is complementary to the Euclidean r-space. The two
distributions g(r) and G(k) carry exactly the same

information. In fact, knowing G(k), the function g(r) can be computed

from
G(r) = ∫G(r) exp [- 2 πi(k • r)]dk
These are known as Fourier transforms and reverse Fourier transforms.
Technically, acquiring an imaging in the presence of a gradient breaks
the data down into its component frequencies (in k-space), which is the
essence of a Fourier Transform. Turning the k-space data back into an
image is technically a reverse Fourier transform.
Figure 9-11. A: A square or rectangular function can be
approximated as the sum of a finite number (N) of sine and cosine
functions. B: N = 1. C: N = 3. D: N = 7. E: N = 20. The signal in (E)
more closely approximates a rectangle, except for the presence of
a ring-down effect.
 Figure 9-12. The FT of the signal in Equation 9-3 is a set of spikes
whose envelope is a sinc function.

In summary, the Fourier series tells us that a signal can be represented

by a series of sine and cosine waves (in the time domain). The Fourier
transform, however, gives the frequency spectrum of the function (in
the frequency domain).

Key Points
The FT, as intimidating as it may appear, represents
a simple concept. Every signal (in the time domain)
is composed of a series of frequencies. The FT is a
way of representing that signal in terms of its
frequencies. The FT also allows mathematical
manipulations performed in the frequency domain,
which are sometimes easier than in the time domain.
The one-to-one relationship between a signal and its
FT allows reconstruction of the original signal from its
FT.
In other words, the FT represents a function in the
frequency domain whose amplitude varies with the
frequencies present in the signal. The bandwidth
 (BW) is simply a measure of the range of frequencies
present in the signal (in Hz or in radians/sec).

Questions
9-1 T/F The FT of an FT equals the original signal.

9-2 T/F
(a) It is always easier to perform calculations in the frequency
domain.
(b) It is always easier to perform calculations in the time
domain.

9-3 T/F The FT of a cosine function consists of two spikes, one on

each side of the zero separated by the frequency of the cosine.

9-4 T/F The FT represents the frequency spectrum of a signal,

whereas the Fourier series decomposes the signal into a series of
sine and cosine waves.

1One can think of a negative frequency as oscillation in the opposite

direction. For instance, if clockwise rotation is regarded as a positive

frequency, then counterclockwise rotation would constitute a negative
frequency. Also, see the discussion below regarding even and odd
functions and their Fourier transforms.
2Actually, the negative spike has an imaginary rather than a real

amplitude (in the form of iA, where i is the imaginary unit and A is
the amplitude of the spike). This concept is rather important in
understanding the symmetry that exists in k-space, discussed in
Chapters 13 and 16.
 10
Image Construction: Part I (Slice
Selection)

Introduction
The signals received from a patient contain information about the
entire part of the patient being imaged. They do not have any
particular spatial information. That is, we cannot determine the
specific origin point of each component of the signal. This is the
function of the gradients. One gradient is required in each of the x, y,
and z directions to obtain spatial information in that direction.
Depending on their function, these gradients are called
1. The slice-select gradient
2. The readout or frequency-encoding gradient
3. The phase-encoding gradient
Depending on their orientation axis they are called Gx, Gy, and Gz.
Depending on the slice orientation (axial, sagittal, or coronal), Gx, Gy,
and Gz can be used for slice select, readout, or phase encode.
A gradient is simply a magnetic field that changes from point to point—
usually in a linear fashion. We temporarily create magnetic field
nonuniformity in a linear manner along all three axes to obtain
information about position.
 Figure 10-1. Selecting a slice of a certain thickness.

First, we'll consider the slice-select gradient, which is the easiest of all
to understand. Once a slice has been selected, we worry about the
problem of in-plane spatial encoding, that is, discriminating position
within the slice. As we'll see shortly, the principles behind slice
selection and spatial encoding in MRI are different from the principles
used in computerized tomography (CT).

How to Select a Slice

Suppose that we have a patient on the table and we want to select a
slice at a certain level and of a certain thickness (Fig. 10-1). Remember
that the patient is lying in the external magnetic field B0, which is
oriented along the z-axis. If we transmit a radio frequency (RF) pulse
and get a free induction decay (FID) or an echo back, the received
signal would be from the entire patient. There is no spatial
discrimination. All we get is a

signal, and we have no idea yet from where exactly in the body the
signal is coming.
The frequency of the RF pulse is given by the Larmor frequency:
ω0 = γB0
If we transmit an RF pulse that does not match the Larmor frequency
(the frequency of oscillation at magnetic field B0), we won't excite any
of the protons in the patient.
However, if we make the magnetic field vary from point to point, then
each position will have its own resonant frequency. We can make the
magnetic field slightly weaker in strength at the feet and gradually
increase in strength to a maximum at the head (Fig. 10-2). This effect
is achieved by using a gradient coil.
Let's say the magnetic field strength is 1.5 T at the center, 1.4 T at the
feet, and 1.6 T at the head. Then, the foot of the patient will
experience a weaker magnetic field than the head. Therefore, a
gradient in any direction (x, y, or z) is a variation in the field along that
axis in some fashion (the most common form of which is linearly
increasing or decreasing). If we now transmit an RF pulse of a single
frequency into the patient, we will receive signals corresponding to a
line in the patient at the level of the magnetic field corresponding to
that frequency (according to the Larmor frequency), but it will be an
infinitely thin line. What we need to do is transmit an RF pulse with a
range of frequencies—a bandwidth of frequencies.
Figure 10-2. Slice thickness is determined by the slope of the
gradient.
 Figure 10-3. The relationship between field strength and Larmor
frequency in determining slice thickness and position.

What will the RF pulse look like in the frequency domain? First, note
that for a 1.5-T magnet, the Larmor frequency is 64 MHz for hydrogen
protons:

64 MHz corresponds to B0 of a 1.5-T magnet.

To see how this is derived, recall that

ω0 = γB0
where γ ≅ 42.6 MHz/T and B0 = 1.5 T This results in ω0 = 42.6 × 1.5 = 64
MHz The graph in Figure 10-3 shows the magnetic field strength and the
corresponding Larmor

frequency range. We are concerned here about a magnetic field

strength ranging from 1.4 to 1.6 T because, in our example, that is the
magnetic field strength range to which we have exposed the patient.
Let's excite one slice with an RF pulse. For example, let's excite a slice
extending from 1.55 to 1.57 T. This corresponds to a frequency range
from 66 to 67 MHz.

Figure 10-4. Comparison of wide and narrow waveforms and their

FTs. The narrower the waveform in time domain, the wider its FT
will be.

If the RF has a square shape in the frequency domain with a range of

frequencies that correspond to a range of magnetic field strengths,
then we will excite only the protons that are in the slice containing
that range of magnetic field strengths. The other protons in the rest of
the body are not going to get excited because the range of frequencies
from which we are transmitting the RF pulse does not match the
Larmor frequencies of the other protons. The range of frequencies in
the RF pulse will only match the Larmor frequency of a single slice.
So, we transmit an RF pulse with a range of frequencies that we know
will correspond to a range of magnetic field strengths in a particular
slice. This range of frequencies determines the slice thickness and is
referred to as the bandwidth.
Bandwidth = range of frequencies (determines the slice thickness)
We can measure the bandwidth (range of frequencies) by looking at the
Fourier transform of the RF pulse. Let's now compare the RF signal with
its Fourier transform. The RF pulse is generally a sinc wave and looks
like Figure 10-4a with the Fourier transform that has a square shape.
If we have a narrower signal, we get a wider frequency bandwidth (Fig.
10-4b). The narrower signal reflects the fact that there are more
oscillations in a given period of time so that the maximum frequency of
the signal is greater. Because the Fourier transform depicts an infinite
number of frequencies from zero to maximum, the bandwidth gets
wider to depict a greater maximum frequency.
Let's apply this example to a cosine wave (Fig. 10-5a). The Fourier
transform for this cosine wave contains two spikes, one on each side of
zero (Fig. 10-5b). If we then take a cosine wave with twice the
oscillation frequency and look at its Fourier transform, we see that the
spikes are farther apart (Fig. 10-6). Thus, as the cosine wave goes two
times faster, the Fourier transform shows that the maximum frequency
(in this case, the only frequency) is two times farther away from the
zero point. Furthermore, if we look at the diagram of the cosine waves,
the faster cosine wave has a narrower oscillating waveform—the
narrower the waveform, the faster the oscillation frequency. Again, the
narrower the waveform of the RF signal, the wider its bandwidth (the
greater the maximum frequency; Fig. 10-6).
 Figure 10-5. The FT (B) of (A) a cos signal (cos ω0t) has two spikes
at ± ω0.

Figure 10-6. The FT of cos 2ω0t has two spikes at ±2ω0.

Slice Thickness
Let's now talk about how we establish slice thickness (Fig. 10-7). The
same principle would apply if we were to image from the base of the
skull to the vertex rather than from the patient's head to toe. We
establish a magnetic field strength gradient so that at the midpoint of
the field of study (in this instance, the entire body), the field strength
is 1.5 T; at the low end of the gradient (the foot), the field strength
will be 1.4 T; and at the high end of the gradient, the field strength will
be maximum at 1.6 T. These magnetic field strengths also correspond
to different frequencies. Using the Larmor equation, we can calculate
that approximately:
1.6 T ˜ 68 MHz
1.5 T ˜ 64 MHz
1.4 T ˜ 60 MHz

Figure 10-7. An example of the relationship between slice

thickness and position, frequency, and field strength.
 Figure 10-8. Ideal contiguous slices correspond to ideal
rectangular-shaped FTs that are positioned side by side.

If we pick a frequency bandwidth of a certain range, we would then get

a slice of a certain thickness. Therefore, we transmit an RF pulse with a
specific frequency bandwidth and no frequencies outside of this range
(ideally). The frequency bandwidth will match the Larmor frequencies
of the protons only in a section of the patient of a certain thickness,
which corresponds to the range of magnetic field strengths
corresponding to the Larmor frequencies. The magnetic field strength
everywhere outside this slice is going to be either more or less than the
magnetic field strengths that correspond to the Larmor frequencies of
the RF bandwidth.
Question: What happens if we put one slice right next to another?
Answer: Ideally, the contiguous slices are right next to each other and
the Fourier transform has a rectangular shape (Fig. 10-8). In other
words, we want to have frequency ranges that are discrete and next to
each other, so that each range of frequencies excites a different slice,
and we can obtain contiguous slices.
Figure 10-9. A more realistic FT has side lobes such as a Gaussian
curve.
 Figure 10-10. Cross-talk: in the absence of ideal rectangular-
shaped FTs, the side lobes of the FTs may overlap.

Cross-Talk. In reality, the frequency spectrum of the RF pulse does not

have a rectangular shape. Instead, it may have a bell shape or
“Gaussian” curve (Fig. 10-9). If we place these frequency spectrums
close together, they will overlap; these areas of overlap cause “cross-
talk” (Fig. 10-10).
Remember that cross-talk is best explained in the frequency domain. It
is more difficult to comprehend this in the time domain. To avoid this
cross-talk created by the overlap of adjacent frequency bandwidths, we
create a gap between the consecutive bandwidths (in the frequency
domain), thus creating a gap between consecutive slices in the actual
image (Fig. 10-11). This will minimize or eliminate “cross-talk.”
How to Change the Slice Thickness. There are two ways to change the
slice thickness:
1. The first way to decrease the thickness is to use a narrower
bandwidth. A narrower frequency bandwidth will excite protons in a
narrower band of magnetic field strengths (Fig. 10-12a).

Figure 10-11. To minimize cross-talk, the slices are set farther

apart (by introducing gaps between successive slices).
 Figure 10-12. To decrease slice thickness, either a narrower BW
(A) or a steeper gradient (B) is used.

2. The second way to decrease slice thickness is to increase the slope of

the magnetic field gradient (Fig. 10-12b), that is, by increasing the
gradient strength.
Slice-Select Gradient. The change in the magnetic field strength along
the z-axis is called the z gradient (Gz). For an axial slice in a
superconducting magnet, it is also called the slice-select gradient. If
we increase the gradient and keep the frequency bandwidth the same,
we get a thinner slice.
The slice thickness can be decreased by
1. decreasing the bandwidth of the RF pulse, or
2. increasing the slice-select gradient.
There is an electronic limitation as to how much we can decrease the
bandwidth. There is also a machine limitation as to how much we can
increase the gradient. These factors set an absolute limit on how thin a
slice can be.
By the foregoing procedure, we select a slice with a certain thickness.
The slice is selected with a frequency range in the RF pulse that
corresponds to the slice location and its thickness. The echo signal that
we get back from the slice is from the entire slice. We have no way yet
of discriminating points within the slice. This is where the frequency-
encoding and the phase-encoding steps come into play.

Review
RF Pulses. There are two types of RF pulses:
1. Nonselective
2. Selective
 Figure 10-13. The sinc function and its FT. The bandwidth (BW) is
2fmax where fmax is the frequency of the sinc function.

By selective, we mean an RF pulse that is slice selective—an RF pulse

whose frequency bandwidth corresponds to a specific band of magnetic
field strengths along a magnetic field gradient. Ideally, this RF pulse
will select only a certain slice of the body that we are imaging (used in
two-dimensional [2D] imaging).
A nonselective RF pulse excites every part of the body that is in the
coil (used in threedimensional [3D] imaging).

Sinc RF Pulse
Earlier, we talked about one type of RF pulse in which we had a sinc
wave in the time domain. The sinc function is mathematically
expressed as
sinc (t) = sin (t)/t
This simply means that the oscillating function sin t is divided by t.
Therefore, because t goes into an oscillating function (sin t), the result
will be an oscillating function. As t goes from a large value to a small
value, the result of (sin t/t) will get larger and will reach maximum
when t approaches zero. The rectangular transform in the frequency
domain has a positive maximum frequency (fmax) and a negative
maximum frequency (-—fmax), as shown in Figure 10-13. The bandwidth
is thus 2 × (maximum frequency), that is,
BW = 2fmax
This is only one type of selective RF pulse. There are other types of
selective RF pulses.

Gaussian RF Pulse
The first generation of MR machines used an RF pulse that had a
Gaussian shape (Fig. 10-14a).

Figure 10-14. The FT (B) of a Gaussian signal (A) is itself Gaussian.

The Gaussian RF pulse has a bell shape in the time domain.

The Fourier transform of a Gaussian function is also a Gaussian curve
(Fig. 10-14).
If we take another Gaussian curve in the narrower time domain, its
Fourier transform will be wider (Fig. 10-15). An inverse relationship
exists between the range of frequencies and the duration of the RF
pulse, as we have discussed before.
A short pulse results in a wider bandwidth. The MR machine uses
different types of RF pulses for different purposes, but for the sake of
discussion, let's assume that we are dealing with an ideal sinc wave RF
pulse that has an ideal square Fourier transform.
Remember that the RF pulse is an electromagnetic wave generated by
an electric current through a coil. If we use the body coil to generate
the RF pulse, we generate the RF throughout the body. If we use a
transmit/receive coil, then the coil transmits the RF pulse and also
receives the signal from the body resulting in less RF energy in the body
and more signal. Most coils are receive only which do not change the RF
energy placed in the body, but increase the signal.

Figure 10-15. A narrow Gaussian signal has a wide Gaussian FT,

and vice versa.

Going back to the sinc wave, in reality we can't go all the way to
infinity in time when transmitting a signal. We have to truncate the
signal and deal with a certain finite time domain (Fig. 10-16). The
Fourier transform of this truncated signal gives the “ripples” effect on
the square wave as we have seen in Chapter 9. The more we truncate
the signal, the more “ripples” we get. Ripples are also called
“overshoot” and “undershoot” artifacts.
Bandwidth. Bandwidth is a measure of the range of frequencies. We
know that for a 1.5-T magnet, the Larmor frequency is about 64 MHz.
The RF pulse that we generate is in the RF range, but its bandwidth is
in the audible frequency range.
This is analogous to a radio. When we tune into the FM station KSON
97.3 FM in San Diego, are we really getting 97.3 MHz of sound waves? If
the signal you receive on your radio

has this high a frequency, you can't hear it (we can't hear a MHz
frequency signal [maybe dolphins can, but humans can't!]). However,
the frequency that we receive is actually in the audible range. Each
station has a certain frequency range that they deal with, and the
bandwidth that they use is nearly the same for all radio stations in the
audible frequency range; it is about 1 to 2 kHz (Fig. 10-17).

Figure 10-16. The FT of a truncated sinc function has a

rectangular-like profile containing rings or ripple effects.

Now, the bandwidth of 1 kHz is 100,000 times smaller than the 97.3
MHz frequency. The audible frequency range gets modulated to the
center frequency (e.g., FM 97.3 MHz). The modulated frequency is
then transmitted. The antenna receives this modulated signal. The
signal then goes into the radio and is demodulated (Fig. 10-18).
Depending on which station we tune into (e.g., KSON FM at 97.3 MHz),
we get that range of frequencies demodulated back to zero frequency.
Thus, the bandwidth of a signal transmitted by a radio station always
stays at about 1 kHz, but it is transmitted as a 1-kHz bandwidth
modulated to, say, 97.3 MHz. Then, in the radio, it is demodulated back
to zero center frequency, still with a bandwidth of about 1 kHz. What
we send and what we receive are in the same frequency range. In
between, the frequency gets modulated from 0 to 97.3 MHz. We are
just changing the center frequency. Everything else stays the same.

Figure 10-17. An example of frequencies of FM radio stations.

The reason radio transmission does this is that each radio station is only
allowed a narrow bandwidth (in the kHz range) for transmission.
However, we cannot transmit a kHz frequency. It just doesn't travel
very far, and radio transmissions have to travel for miles. Besides, the
signals from different stations will get all mixed up if they all work in a
small frequency range in the order of a few kilohertz. Therefore, the
kHz frequency bandwidth is modulated to the MHz range, still
maintaining the same kHz bandwidth. Now with the MHz frequency as a
carrier, the narrow 1-kHz bandwidth can be transmitted over long
distances.
In MRI, the center frequency is the Larmor frequency. Thus, the RF
pulse that we transmit into the patient is centered at the Larmor
frequency of 64 MHz at 1.5 T (Fig. 10-19). However, the bandwidth of
frequencies in the RF pulse is very narrow. Again, for simplicity, we will
assume that the bandwidth of the RF pulse has been demodulated to a
center frequency of zero.
Slice-Select Gradient. Let's go back to the slice-select gradient. We
purposely create a linear magnetic nonuniformity, so the foot will
experience a weaker magnetic field than the head. The slope of
magnetic field versus distance is called the gradient. A gradient is a
measure of change of magnetic field with distance. We can have a
linear gradient or a nonlinear gradient. The gradients that we use in
MRI are usually linear. (In fact, nonlinearities may be present in the
gradients that cause geometric distortion artifacts [see Chapter 18].)
By creating this gradient, different magnetic fields are experienced
from the foot to the head

in an increasing order. The protons in the body will also experience a

gradient in terms of their precessional frequency in that the protons in
the foot are going to precess slower than the protons in the head.

Figure 10-18. Each radio station has a limited audible BW. To

transmit this to the listeners, this BW is “modulated” via a carrier
frequency (which is several orders of magnitude higher). In the
radio, this frequency is demodulated back to the audible range.
We then transmit an RF pulse that matches the proton precessional
frequency in a certain section of the body. Then, only the protons in
this section will resonate; none of the other protons in any other
portion of the body will resonate (i.e., flip into the transverse plane).
If we want to select a specific slice, then we transmit an RF pulse with
a bandwidth that has the appropriate center frequency. This gradient is
turned on only when we transmit the RF pulses. This makes sense
because we only want the RF pulse to excite a thin slice of the body.
When we transmit the 180° pulse for the same slice, we activate the
same gradient.
When we study another slice, the gradient stays the same. We just alter
the center frequency of the RF pulse. In this manner, we can excite
different slices in any order desired.

Figure 10-19. The process of demodulation in MRI.

Key Points
We have seen how one goes about selecting a slice in
the body. This is done via a slice-select gradient. To
vary the slice thickness, we can either vary the
bandwidth of the transmitted RF pulse or the slope of
the gradient. Thus, we can decrease the slice
 thickness by
1 decreasing the bandwidth of the RF pulse, or
2 increasing the slice-select gradient.
Because the RF profiles are not ideal and may have
side lobes or tails, if you try to have contiguous
slices, you'll run into a problem called cross-talk.
Basically, the transmitted signals in the frequency
domain (i.e., their Fourier transforms) will overlap
and “cross-talk.” To avoid this, you must introduce
gaps between the slices. This is done by excluding a
certain range of frequencies (i.e., bandwidths) in the
transmitted RF pulses. The larger the gaps, the less
cross-talk you get, but the chance of missing a lesion
within the gaps is greater.
The center frequency for each transmitted bandwidth
is like a carrier frequency around which the desired
bandwidth is centered, much like what goes on in
radio-communication. Once a slice is selected, the
question arises as to how to determine the pixels
within that slice. This is the topic of the next chapter.

Questions
10-1 (a) The range of frequencies included in an RF pulse is referred
to as its bandwidth (BW). Suppose that an RF pulse has frequencies
ranging from −500 to 500 Hz (i.e., BW = 1,000 Hz = 1 kHz). Now, to
achieve a slice thickness of 5 mm, determine the amplitude of the
slice-selection gradient.
(b) What is the minimum achievable slice thickness given a minimum
RF BW = 426 Hz and a maximum gradient Gz = 10 mT/m?
Hint: ω = γB so Δω = BW = γΔB. Now, B = Gzz so that ΔB = GzΔz. Thus
BW = γΔB = γGzΔz or Δz = BW/(γGz), where Δz = slice thickness and γ
= 42.6 MHz/T.

10-2 Thinner slices can be achieved by

(a) decreasing the transmit (RF) BW
(b) decreasing the receive (signal) BW
(c) increasing the slice-select gradient strength
(d) all of the above
(e) only (a) and (b)
(f) only (a) and (c)

10-3 T/F The FT of a sinc function is rectangular.

10-4 T/F The FT of a bell-shaped Gaussian function is also bell-

shaped.
11
Image Construction: Part II (Spatial
Encoding)

Introduction
In the last chapter, we learned how to select a slice and how to adjust
its thickness. However, we did not address the question of from where
within a particular slice each component of the signal comes. In other
words, we still don't have spatial information regarding each slice. To
create an image of a slice, we need to know how much signal comes
from each pixel (picture element) or, more accurately, each voxel
(volume element). This is the topic of spatial encoding, of which there
are two parts: (i) frequency encoding and (ii) phase encoding.

Frequency Encoding
After selecting a slice, how can we get information about individual
pixels within that slice? As an example, consider a slice with three
columns and three rows, for a total of 9 pixels. This slice is selected
using a selective 90° pulse (Fig. 11-1). We turn the Gz (slice-select
gradient) on during the 90° pulse and turn it off after the 90° pulse.
 Figure 11-1. Spin-echo pulse sequence diagram.

We also send a selective 180° RF refocusing pulse, and we again turn

the Gz gradient on during the 180° pulse. The echo is received after a
time TE. The echo is a signal from the entire slice. To get spatial
information in the x direction of the slice, we apply another gradient Gx
called the frequency-encoding gradient (also called the readout
gradient) in the x direction (Fig. 11-2). With this gradient in the x
direction, the center of this 3 × 3 matrix (the center volume) is not
going to experience the gradient, that is, it's not going to experience
any change in magnetic field from that prior to turning on the Gx
gradient. The column of pixels to the right of midline will experience a
higher net magnetic field. The column of pixels on the left will have a
lower net magnetic field.
 Figure 11-2. Frequency-encoding gradient along the x-axis.

The Gx gradient is applied during the time the echo is received, that
is, during readout.
Let's now assign some magnitude numbers to the pixels in the matrix
(Fig. 11-3).
The numbers in each pixel and their specific location is what we
ultimately want to discover because this corresponds to an image. We
want to recreate this image using MRI.
Initially, all the protons in this section experience the same frequency
of precession. Let's call that frequency ω0. Now let's assign each pixel
its frequency at a specific point in time before turning on the Gx
gradient, while they all still have the same frequency, and combine it
with the magnitude we've assigned to each pixel (Fig. 11-4). For
simplicity, we'll use a cosine wave as the received signal. In reality, the
received signal is a more complicated one, such as a sinc wave.

Figure 11-4. Each pixel is also assigned a frequency ω0 and is

represented as A cos ω0t, where A is the magnitude.
 Figure 11-3. In the previous example of a 3 × 3 matrix, each pixel
is assigned a value (magnitude).

Each pixel has a designated magnitude (amplitude), and they all have
the same precessional frequency ω0 (except those pixels that have zero
signal amplitude). Without any gradient in the x direction, this is the
signal we're going to get. The signal will be the sum of all the signals
from each pixel.
The sum of the amplitudes = (0) + (1) + (-2) + (1) + (2) + (0) + (1) + (0) +
(1) = 4 The frequency is the same for each pixel (i.e., ω0), as is the
shape of the signal (i.e., cos ω0t). So the sum of the pixels = signal
from whole slice = 4 cos ω0t We know that, in reality, the signal is more
complex. For example, the signal is a decaying signal with time such as
a sinc wave, but, for simplicity, let's accept that we are dealing with

a simple cosine wave as our signal with an amplitude of 4.

 In summary, when we transmit an RF pulse with frequencies
appropriate for a particular slice, all the protons in that slice will start
to precess in phase at the Larmor frequency (ω0). Each pixel contains a
different number of protons designated by a number. For purposes of
illustration, the number we have assigned to each pixel is proportional
to the number of protons in each pixel. This number corresponds to the
amplitude of the signal. The signal is designated as a cosine wave
because it oscillates as a result of the precessing protons. However, we
still do not have any spatial information; all we have at this point is a
signal coming from the entire slice without spatial discrimination. What
we want to be able to do is to separate the summed signal into its
components and to tell, pixel by pixel, where each component of the
received signal originated.
Let's now apply the frequency-encoding gradient in the x direction and
see what happens to the pixels (Fig. 11-5). Let's look at the three
columns in the matrix:
1. The pixels in the center column will not feel the gradient. Thus, they
will remain with the same frequency (ω0). (And, of course, the
amplitude of each pixel is constant because the number of protons
doesn't change.)
2. The column of pixels to the right of midline is going to have slightly
higher frequency. We'll call this (ω0+). This is because at a higher
magnetic field strength, the protons in this column will oscillate at a
higher frequency.
3. The column of pixels to the left of midline will experience a slightly
lower field strength and thus have a precessional frequency a little
lower than the other columns. We'll call this (ω0-).

The signal that we get now is still the sum of all the individual signals;
however, now each column of pixels has a different frequency, so we
can algebraically only add up the ones that have the same frequency,
as follows: Column #1: 0 + (cos ω0−t) + (−2 cos ω0−t) = −cos ω0−t Column
#2: (cos ω0t) + (2 cos ω0t) + 0 = 3 cos ω0t Column #3: (cos ω0+t) + 0 +
(cos ω0+t) = 2 cos ω0+t
 Sum of signals = (−cos ω0-t) + (3 cos ω0t) + (2 cos ω0+t)
Let's look at the Fourier transform of the signal before the Gx gradient
is applied (Fig. 11-6A) and look at it again after the Gx gradient is
applied (Fig. 11-6B). For a cosine wave, the Fourier transform is a
symmetric pair of spikes at the cosine frequency, with an amplitude
equal to the magnitude of the signal. (Remember that this is simplified.
Usually, we deal with a band of frequencies, i.e., the bandwidth, as
opposed to a single frequency. However, right now, for simplicity, we
leave it as a single frequency, with its Fourier transform as a single
spike.)
Now the computer can look at the Fourier transform and see that we
are now dealing with three different frequencies:
1. The center frequency comes from the central column, and the
amplitude of the frequency spike represents the sum of the
amplitudes of the pixels in that column, that is, (3 cos ω0t).
2. The higher frequency comes from the column to the right, and the
amplitude of that frequency spike represents the sum of the
amplitude of the pixels in that column, that is, (2 cos ω0+t).
3. The lower frequency comes from the column to the left, and the
amplitude of that frequency spike represents the sum of the
amplitude of the pixels in that column, that is, (−cos ω0−t).

The way frequency encoding works is that frequency and position have
a one-to-one relationship:
frequency ↔ position
So far we have done some spatial encoding and have extracted some
information from the slice.

We are now able to decompose the slice matrix into three different
columns (Fig. 11-7). That is, we have three different shades of gray
corresponding to the three columns.
Figure 11-5. When the matrix is exposed to the frequency
gradient, different frequencies result in each column: ω0, ω0+, and
ω0−.
 So, now we've done our job in the x direction. The next thing we want
to do is to decompose the individual columns into their three individual
pixels (i.e., work in the y direction). The two ways of doing this are as
follows:
1. Back projection
2. Two-dimensional Fourier transform (2DFT)
Back Projection. If we think in terms of CT imaging and apply
gradients, we start out with an area we want to image and apply a
gradient (Fig. 11-8A). Then we can rotate the gradient by an angle θ
and reapply the gradient (Fig. 11-8B). We can continue this to complete
360°, and

each time we do this we get different numbers. At the end, we end up

with a set of equations, which can be solved for values of the pixels in
the matrix. This is the back projection approach performed by rotating
the frequency gradient.
Figure 11-6. A: The signal and its Fourier transform (FT) prior to
the application of the gradient. The signal has a single frequency
ω0. B: After application of the frequency gradient Gx, the resulting
composite signal will be composed of three frequencies with a
more complicated waveform and FT.
 Figure 11-7. The sum of the signals in each column. Since the
signals belonging to the same column have the same frequency,
they are additive.

Advantage
1. It is possible to pick a small field of view (FOV).
 Disadvantages
1. This technique is very dependent on external magnetic field
inhomogeneities (i.e., it is sensitive to ΔB0).
2. This technique is also very sensitive to the magnetic field gradients.
If the gradient is not perfect, you get artifacts.
Because of these disadvantages, this technique was given up.

Figure 11-8. Back projection. By gradually rotating the gradient

(from A to B), we get a set of equations the solution to which gives
the pixel values.

2DFT: Two-dimensional Digital. Fourier transform (2DFT) is the

method currently used and is the topic of the remainder of this chapter.

Advantages
1. Lack of sensitivity to external magnetic field inhomogeneities
2. Lack of sensitivity to gradient field inhomogeneities
 Figure 11-9. The phase-encode gradient Gy is applied along the y-
axis. It is usually applied between the 90° and the 180° pulse or
between the 180° pulse and the echo. The first Gx gradient is used
for offsetting any phase shift induced during frequency readout
(further discussion in Chapter 14).

Phase Encoding
In the 2DFT technique, in addition to using the Gz gradient for slice
selection and the Gx gradient for encoding in the x direction, we add
another gradient Gy in the y direction. This is called the phase-
encoding gradient (Fig. 11-9).
We turn on the Gy gradient before we turn on the Gx (readout)
gradient. It is usually applied right after the RF pulse or just before the
Gx gradient or anywhere in between.

Gy is usually applied between the 90° and the 180° RF pulses or

between the 180° pulse and the echo.
So now let's again look at the slice with its 9 pixels before either the Gx
(frequency-encoding) gradient or the Gy (phase-encoding) gradient is
applied (Fig. 11-10A). The designation of the pixels on the right in
Figure 11-10A is a way of denoting phase and frequency. The arrow
represents the hand of a clock and denotes the position of precession
(i.e., the phase) at a given point in time. After the 90° RF pulse, all the
protons in the selected slice precess at the same frequency (ω0). At any
point in time before being exposed to a magnetic field gradient, the
protons in all the pixels will all be pointing in the same direction (e.g.,
north) without any phase difference among them.
 Figure 11-10. Analogy of clock handle. A: Before application of Gx
or Gy, the handles point north. B: After application of Gy (right),
the handles in different rows get out of phase. (Continued on next
page.)

This is precisely what the clock diagram shows: Prior to being exposed
to a magnetic field gradient, all the protons in each pixel are in phase
with each other, oscillating at the same frequency. (Note that we have
eliminated magnitude from the clock diagram. All we are concerned
about for

now is the phase and frequency of the direction of the spins at a

particular point in time.)

Figure 11-10. (continued) C: After application of Gx (right) and Gy

(left), each pixel has a different frequency and phase (i.e., the
handles rotate at different speeds and with different phase,
keeping in mind that the speed is the same for all the elements
belonging to the same column).

Now let's expose the slice to a Gy gradient— a magnetic field gradient in

the y direction (Fig. 11-10B). With the gradient now applied in the y
direction, the pixels in the upper row will experience a higher net
magnetic field, the pixels in the middle row will experience no change
in the magnetic field, and the pixels in the lower row will experience a
lower net magnetic field.
Consequently, the pixels in the middle row, because they experience no
change in the magnetic field, will have no phase change after the
gradient is turned on. They will continue to point in the same direction
as they did before activation of the gradient.
Protons in the pixels on the top row will all start precessing faster
because they are now experiencing a stronger magnetic field.
Therefore, they will all remain in phase with one another but will be
out of phase with the protons in the middle row.
The pixels in the bottom row will all start precessing more slowly
because they are now experiencing a weaker magnetic field. They will
all be in phase with one another but will be out of phase with the
protons in the middle and upper rows.
Once the Gy gradient is turned off, all the protons will now be at the
same magnetic field strength once again, so they will all again precess
at the same frequency.
However, look at what has happened. A permanent phase shift has
occurred in the protons of each row. True, they are now all spinning at
the same frequency; however, those that were previously exposed to
the higher magnetic field and were out of phase with the protons in the
middle row will continue to be out of phase now that all protons are
spinning at the same frequency again. Likewise, those protons
previously exposed to the lower magnetic field that went out of phase
with the protons in the middle now will continue to be out of phase
now that all protons are spinning at the same frequency again.
Now we have caused a difference in the rows of pixels based on phase
(designated by θ in Fig. 11-10B). Differences in spatial position

up and down are reflected in that phase value. Hence the term phase
encoding.
Remember that the Gy gradient is turned on before reading out the
signal. So when we read the signal, we turn on the Gx gradient, which,
as we learned from our earlier discussion, allows us to frequency
encode in the Gx direction (Fig. 11-10C). With the Gx gradient on, the
middle column protons won't experience any change in their
precessional frequency, so their frequency is unchanged. However, as
you can see, each pixel in the middle column already has a distinct
phase shift, which had occurred when the Gy gradient was on, and this
phase shift persists.
With the Gx gradient turned on, the protons in the columns to the right
of midline will experience a greater magnetic field so that all the
protons in this column will have a faster precessional frequency.
However, we notice that each pixel in this column was already out of
phase with the other pixels in the column due to the phase shift that
had occurred when the Gy gradient was on. Therefore, protons in each
pixel of the right column shift the same amount (because they all have
the same increased frequency). However, because they shift from a
unique position, they will then each move to a phase shift that is
different for each pixel.
Likewise, the protons in the column to the left of midline will
experience a lower precessional frequency with the Gx gradient on,
but, again, we notice that each pixel in this column is already out of
phase with the other pixels in the column due to the phase shift that
had occurred when the Gy gradient was on. So, again, after the Gx
gradient is turned on, each pixel will move to a specific phase shift
distinct for each pixel. In summary, x position is represented by a
unique frequency and y position by a unique phase.
 Figure 11-11. The received signal after application of both Gx and
Gy for the 3 × 3 matrix used in previous examples.

The protons in each pixel have a distinct frequency and a distinct

phase, which are unique and encode for the x and y coordinates for
that pixel.
Question: How does one determine the phase shift between adjacent
rows?
Answer: First, to figure out the phase shift, we divide 360° by the
number of rows: Δθ = 360/Number of rows
Because we have three rows, the phase shift between rows is (Fig. 11-
11): Δθ = 360/3 = 120° (or 2π/3).
Therefore, in the middle row, there will be no phase shift. In the upper
row, the phase shift will be 120°. In the lower rows, the phase shift will
be +240° (which is the same as −120°).
Each row has its own unique phase shift caused by the Gy gradient:
Row 1: phase shift of + 120°
Row 2: no phase shift
Row 3: phase shift of − 120°
Also, each column has its own unique frequency caused by the Gx
gradient.
Column 1: frequency of ω0−
Column 2: frequency of ω0

Column 3: frequency of ω0+

When combined, as it is during the readout of the signal, we see that
each pixel has its own unique phase shift and frequency.
Question: Why does it take time to do phase encoding?
Answer: It takes time because we need to do a separate phase encode
for each row of pixels that we need to discriminate in the slice. In this
case, we have three rows of pixels, so we would do three phase-
encoding steps. Each time we do a separate phase encode, it is a new
spin echo taking time TR after a new 90° RF pulse. With each new
phase-encoding step (taking time TR), we change the magnetic gradient
Gy.
TR#1: no gradient—no phase shift between rows
TR#2: gradient with 120° phase shift between rows
TR#3: gradient with 240° (or -120°) phase shift between rows
We wouldn't need another TR because 240° + 120° = 360° phase shift,
and this would give us the same information as a 0° phase shift. What
would happen to the signals with the three different phase-encoding
steps (Fig. 11-12)?
During TR#1, where we don't apply a gradient Gy in the y direction, we
get no phase shift between the rows. Then, when we apply the
frequency-encoding gradient Gx in the x direction, we get the
 frequency difference between the columns.
During TR#2, we apply a Gy gradient so that there is a 120° phase shift
between the rows. Then, when we apply the frequency-encoding
gradient Gx in the x direction, we get the frequency difference in the
columns on top of the 120° phase shift between the rows.
During TR#3, we apply a steeper Gy gradient so that now there is a 240°
(or -120°) phase shift between the rows. Then, when we apply the
frequency-encoding gradient in the x direction, we get the frequency
difference in the columns on top of the 240° phase shift between the
rows.
Notice that in each case the middle row never experiences any phase
shift, and the middle column never experiences any frequency change.
Therefore, the center pixel never experiences any frequency or phase
shift. Also notice that we need a separate TR for each phase-encoding
step. That is why phase encoding takes time. We need one TR time
period to perform each phase-encoding step. Hence, part of the
formula describing the acquisition time for a sequence includes TR and
the number of phase-encoding steps (as well as the number of
excitations).
As an example, if we need to discriminate 256 rows, then we need to
perform 256 phaseencoding steps, each with a different gradient Gy,
taking time 256 × TR. The difference in the phase shift between the
rows in this case would be 360°/256 phase-encoding step ≈ 1.40°.
1. TR#1 = no gradient, no phase shift
2. TR#2 = gradient to allow a phase shift = 1.40° between rows
3. TR#3 = steeper gradient to allow a phase shift = 2 × (1.40°) between
rows
4. TR#4 = steeper gradient to allow a phase shift = 3 × (1.40°) between
rows
5. TR#256 = the steepest gradient to allow a phase shift = 255 × (1.40°)
between rows
If we do these steps once more, we get the same information as the
 first phase-encoding step, which is redundant. Each time we do a
phaseencoding step followed by frequency encoding, we get a signal.
The first signal we get is without any phase shift (TR#1). Then we
activate the phase-encode gradient, add some phase shift, and get
another signal (TR#2), and so on. Each signal is different because it has
a different phase shift.
Data Space. Each of these signals fills one line in a set of rows referred
to as the data space (Fig. 11-13). k-space can be thought of

as a digitized version of the data space (more on this in Chapter 13).

Let's see how this is done within each TR period:
1. With TR#1, we have no phase shift. After the frequency-encoding
step, a signal is received and placed into one row of the data space.
In the previous example, we happened to put it into the center row
of the data space (this is arbitrary, though).
Figure 11-12. The received signal during each cycle.
 Figure 11-13. Each row in the data space (analog k-space)
contains the received signal corresponding to a particular phase-
encode gradient.

2. With TR#2, we add a phase shift, and, after the frequency-encoding

step, a signal (which will be different from the signal from the first
TR) is received and put into another row in the data space. In the
previous example, we put it into the top row of the data space
(again, this is arbitrary).
3. With TR#3, we add more phase shift. After the frequency-encoding
step, a third signal (which is different from the other two signals
because of the increased phase shift) is received and put into another
row of the data space. In the previous example, we put it into the
bottom row of the data space.
The interval between the rows in the data space is given by TR msec,
going through a cycle from one 90° pulse to the next 90° pulse.
Summary
TR#1: place the signal into the center of the data space
TR#2: place it one above the center of the data space
TR#3: place it one below the center of the data space
If we had a TR#4, we would place it two above the center, and if we
had a TR#5, we would place it two below the center. We could continue
along filling the data space in this manner.
Remember that in our example, we start in the center of the data
space with TR#1, which has no phase shift. In each subsequent row (as
we go farther out in the data space) there is progressively greater and
greater phase shift in each phase-encoding step. Also remember that
each phase-encoding step has a different magnetic field gradient; thus,
its phase shift will be different from each preceding and each
subsequent phase-encoding step.
However, also note that the selection of phase-encoding steps, that is,
the order in which we perform them, is arbitrary. We can start with no
phase shift and go progressively to maximum phase shift, or we can
start with maximum phase shift and go down to no phase shift.
Likewise, the position to which they are assigned in the data space is
also arbitrary.
When we discuss fast spin echo in later chapters, we will see the
arbitrariness of this

phase-encoding assignment. Even in conventional spin-echo imaging,

the placement of signals into the data space can vary. For example, two
different patterns of the data space filling with 256 phase-encoding
steps are shown in Figure 11-14. We can put the first signal in the
bottom of the data space and work up with successive phase-encoding
steps. Or, as in our example, we can start in the center and go
alternately up and down. Usually, the center row of the data space
corresponds to no phase gradient.
 Figure 11-14. Various k (or data) space trajectories.

Note, however, that the center of the data space does not represent
the center of your picture. Each signal has information in it about the
entire picture. Remember that each signal that goes into each row of
the data space is the sum of all the signals from individual pixels in the
slice.
The information in the data space is in the time domain (so it is not as
scary as it looks). In fact, it is in the time domain in both directions:
the received signal is displayed over a period of time (t), and signals in
two successive rows are obtained at every TR (Fig. 11-15).
The information in this data space has not yet been digitized. In fact, a
digitized version of this information is the true k-space. We will see in
a later chapter (Chapter 16) that the “digitized” k-space is in a spatial
frequency domain. But let's not get confused now! Let's see how we
can digitize this information in the data. This is accomplished via
sampling.

Sampling
The signal that goes into the data space has been phase encoded and
frequency encoded, but it has not yet been sampled (more on this in
the next chapter).
When we describe a matrix of, say, 256 × 192, what do we mean by it?
If we only have 192 phase-encoding steps, why do we have 256 (instead
of 192) frequency encodes if we only need one frequency encode for
every encoding step?
Actually, the 256 number refers to the different number of frequencies
we have for each phase-encoding step. These two steps are thus
independent from one another. For example, let's look at a 4 × 5
matrix. This means that we do five different frequencies for each
phaseencoding step, with the center column having no change in
frequency and two different frequencies for the columns on either side;
there will be

four different phase-encoding steps. Therefore, the result will be

asymmetric pixels. Let's go back to our original 3 × 3 matrix, with its
data space as in Figure 11-15.
Figure 11-15. The data space is in the time domain.
 Figure 11-16. The Fourier transform of the data space (Fig. 11-5)
is the desired image, shown here in three dimensions.

How do we go from the data space to the desired image? The answer is
via Fourier transformation. The Fourier transform of the signals in
Figure 11-15 is a set of spikes, which in a three-dimensional space looks
like Figure 11-16. More on this in chapters to come.

Key Points
In the previous chapter, we learned how to select a
slice using a slice-select gradient Gz. In this chapter,
we saw how to determine the pixel values in a slice
using two gradients: (i) a frequency-encoding (or
readout) gradient Gx and (ii) a phase-encoding
gradient Gy.
The Gx gradient is the same strength for each echo,
and since it is applied during the readout (i.e., during
reception of the echo), it alters the Larmor frequency
along the x-axis. This provides specific information
 along the x-axis. Each TR interval contains one
readout (Gx) per slice.
The Gy gradient, however, is applied in increments
between the 90° RF pulse and each echo. Since this
gradient is applied at a time apart from the echo, it
does not change the echo frequency, but merely
induces a phase shift. Thus, for every slice, each TR
interval contains one phase-encoding step (i.e., one
unique value of Gy). This process completes one line
in k-space corresponding to the selected Gy. This
process is repeated Ny times to fill all of k-space.
We have not yet discussed the mechanism of
performing frequency- and phase-encoding
operations. This is the topic of the next chapter.

Questions
11-1 Match
(i) Gx (ii) Gy (iii) Gz with
(a) applied during the echo
(b) applied during the RF transmission
(c) applied between the RF and the readout

11-2 T/F The purpose of the gradients is to determine the position of

the originating signals from the patient (i.e., spatial encoding).

11-3 What is the phase increment for 128 phase-encode steps (i.e.,
Ny = 128)?

11-4 Match
 (i) position along x-axis
(ii) position along y-axis with
(a) phase-encode gradient Gy
(b) frequency-encode gradient Gx
(c) absolute phase φy
(d) absolute frequency fx

11-5 T/F In conventional spin-echo imaging, during each cycle (one

TR period), only a single value of the phase-encode gradient
strength Gy is applied.
 12
Signal Processing

Introduction
Signal processing refers to analog and/or digital manipulation of a
signal. The signal could be an electric current or voltage, as is the case
in magnetic resonance imaging (MRI). Image processing is a form of
signal processing in which the manipulations are performed on a
digitized image. Analog-to-digital conversion (ADC) is a process by
which a time-varying (analog) signal is converted to a digitized form
(i.e., a series of 0s and 1s) that can be recognized by a computer. An
understanding of signal processing requires a basic understanding of the
concept of frequency domain and Fourier transform (FT) because most
of the “processing” of a signal is accomplished in the frequency domain
and, at the end, the results are converted back into the time domain.
One of the key concepts in signal processing, as we shall see shortly, is
the Nyquist sampling theorem. An understanding of the sampling
procedure allows one to appreciate the relationship between the
samples of a signal (in the time domain) and its bandwidth (in the
frequency domain). Once this concept is grasped, the issue of aliasing
(wraparound) artifact can be explained very easily. A knowledge of
signal processing will also help the reader understand the more
complicated, newer fast scanning pulse sequences presented in later
chapters.

Sequence of Events
First, let's summarize what has been discussed so far. Figure 12-1
illustrates a summary of a spin-echo pulse sequence. The following is a
summary of the sequence of events:
1. We have 90° and 180° pulses separated by a time of TE/2 msec.
2. After a time of TE millisecond after each 90° radio frequency (RF)
 pulse, we get an echo.
3. We turn on the slice-selective gradient (Gz) during the two RF
transmissions. This causes a linear gradient of magnetic field along
the z-axis. By choosing an RF pulse with an appropriate frequency
and bandwidth, we can select a slice at a particular position with a
particular thickness.
As an aside, consider Figure 12-2. Plotting field strength versus position
(Fig. 12-2), the gradient is represented as a sloped line. The slope of
the line is a constant that we call G. The value y along this line with
slope G at point x is y = Gx. This is a simple linear equation. Figure 12-
2B plots gradient strength versus time. Figures 12-2A and 12-2B are
used interchangeably to illustrate a linear gradient with strength G.
4. Right before we receive the echo, we apply a phase-encoding
gradient (Gy). The symbol for the phase-encoding gradent is in Figure
12-3. This symbol denotes the multiple phase-encoding steps that are
necessary as we cycle through the acquisition. Remember that one of
the phaseencode steps may be performed without any gradient.
 Figure 12-1. Spin-echo pulse sequence diagram.

5. The frequency-encoding gradient (Gx) is turned on during the time

period during which the echo is received.
6. Time requirements
a. The frequency-encoding step takes about 10 msec (4-8 msec at high
field; 16-30 msec at lower fields).
b. The phase-encoding step takes 1-5 msec.
c. (c) Each RF pulse (with a Gz gradient) takes 2-10 msec.
Then we repeat the whole sequence of events after time TR. The time
spent from the center of the 90° pulse to the end of the echo readout
is

Figure 12-2. A linear G represents a linear function Gx, which can

be represented either as a linear line with slope G (A) or as a
rectangle with height G (B).

TE + 1/2 (sampling time) = (TE + Ts/2)

The sampling time Ts is the time it takes to sample the echo, which is
the time that the Gx gradient (readout or frequency-encode gradient) is
on. The Gx gradient is on throughout the echo readout—from beginning
to end (Fig. 12-4). The time from the beginning of the 90° pulse to the
midpoint of the echo is TE. Because half of the sampling time continues
past the midpoint of the echo, we need to add half the sampling time
to TE to account for the entire “active time” from the beginning of the
RF pulse to the end of the sampling time:

Figure 12-3. The symbol for phase-encode gradient.

Active time = TE + Ts/2

There may also be time taken up by other events that occur before the
RF pulse (such as presaturation pulses) that we include as overhead
time (To). Thus,

Active time = TE + Ts/2 + To

Let's assume a TE of 40 msec, a sampling time Ts = 10 msec, and an

overhead time To = 5 msec. Then,
Active time = 40 msec + 10 msec/2 + 5 msec = 50 msec
Therefore, it takes 50 msec to read out the signal from one echo. We
then place this signal into the data space (Fig. 12-5). We have
designated 256 rows in the data space (from −127 to +128), and in this
instance we've placed the first signal at position −127.

Figure 12-4. The frequency gradient is turned on during readout

(i.e., during the echo).
 7. In the next TR cycle, we do exactly the same thing, except this time,
the phaseencoding step will be done with a slightly weaker magnetic
gradient and will be one step higher in k (data) space.
Question: Why does the magnitude of the signal differ with each
phase-encoding step? For example, in Figure 12-5, the signal from the
echo of TR#2 seems to have higher maximum amplitude than the signal
from the echo of TR#1.
Answer: The strength of the phase-encoding gradient affects the
magnitude of the signal.
When the phase-encode gradient is at a maximum (when we have the
largest magnetic field gradient), we have the maximum dephasing of
proton spins. Recall that we are dealing with proton spins that have
been flipped into the transverse plane by a 90° pulse, and the signal is
maximum as long as these proton spins stay in phase. By using a
magnetic field gradient, we are introducing an artificial external means
of dephasing. Also, remember that when we flip the protons into the
transverse plane, they are initially in phase and then rapidly go out of
phase due to external magnetic inhomogeneities and spin-spin
interactions (Fig. 12-6A). Next, the protons are flipped with a 180°
pulse and, after a period of time = TE/2, they go back in phase (Fig. 12-
6B).
On top of this, we introduce a magnetic inhomogeneity by way of a
linear gradient in which the inhomogeneity increases linearly, causing
additional dephasing of proton spins. We realize that we have to accept
this additional dephasing because this is the way we obtain spatial
information along the axis of this gradient. This process is called phase
encoding.
However, this explains why, when we use the magnetic field gradient
for phase encoding, the additional dephasing caused by the gradient
will necessarily decrease the overall signal we receive during that
phase-encoding step. It then allows us to conclude as follows:
a. The largest magnetic gradient we use for maximum phase encoding
will give us the lowest magnitude signal.
 Figure 12-5. For each phase-encoding step, a signal is obtained
that is placed in the data space.

b. When the magnetic gradient in the phaseencode direction is zero, we

will not introduce any additional dephasing, and we will get the
largest magnitude signal.
Therefore, if we go back and examine the signal from TR#1 and TR#2,
we see that
a. TR#1 at position -127 in the data space has a lower amplitude signal
than TR#2 at position -126. This is because during TR#1 a larger
phase-encoding gradient is used than during TR#2. (Keep in mind,
however, that assignments of gradients to different TR intervals in
the data space are arbitrary.)
 Figure 12-6. The spins are initially in phase (A) and then get out of
phase at time TE/2. At this point, a 180° pulse is applied, reversing
the vectors (B) so that they come back in phase after another
period TE/2 (i.e., at time TE).

k-space, as we mentioned in the last chapter (with more to come in

Chapters 13 and 16), can

be thought of as a digitized version of the data space. The signal in the

center of k-space (at position 0) has the maximum amplitude. This is
because this signal is obtained at a phase-encoding step at which no
magnetic field gradient is used (i.e., no phase gradient and hence no
extra dephasing due to phase encoding). In fact, the center line of k-
space will always be occupied by the phase-encoding step that uses no
gradient.

The center line of k-space will always contain the

phase-encoding step with the weakest gradient and
thus with the most signal.

The most peripheral lines of k-space will be occupied by the phase-

encoding steps that use the strongest gradients.

The periphery of k-space will contain the phase-

encoding steps with the largest gradients and thus
with the least signal.

Figure 12-7. Multislice acquisition. During each TR cycle, there is a

certain “dead time” that can be used to acquire other slices.
8. Multislice technique
Remember that TR is much longer than the active time needed to
perform all the functions necessary to select a slice, phase encode, and
frequency encode. The TR might be, say, 1000 msec. The active time in
our example was 50 msec. There is a lot of “dead time” in between
the 50 msec of active time and the next 90° pulse. We can take
advantage of this “dead time” in order to get information regarding
other slices.
As an example, in Figure 12-7 we will have time for two additional
slices to be studied during the dead time within one TR period. After
we obtain the signal from slice number 1, we can apply another 90°
pulse of a different center frequency ω, but with an identical transmit
bandwidth to specify the next slice (Fig. 12-8). Here we are talking
about the transmitted bandwidth (of the RF pulse) that determines
slice thickness, not to be confused with the receiver bandwidth (of the
echo) that determines noise. More on this later.
 Figure 12-8. During each TR period, several RF pulses with
different frequencies (and BWs) corresponding to different slices
are transmitted.

To choose the next slice, we keep the same magnetic gradient Gz, but
we choose a bandwidth at a higher or lower center (or Larmor)
frequency to flip the protons 90° in a different slice. The bandwidth is
the same as the first slice, but the center frequency is different.
We choose the same phase-encoding gradient Gy so we get the same
amount of dephasing in this next slice. We sample the echo with the
same frequency-encoding gradient as we did with the first slice.
Because we still have time during the dead time to acquire a third
slice, we repeat everything and again choose a 90° RF pulse of a
different Larmor frequency so we can flip the protons of a different
slice into the transverse plane. The signal from each slice will be
placed in a different k-space.

Each slice has its own k-space.

Slice selection can be performed in several different ways. We can have

contiguous slices; sequential slices with a gap between slices; and
interleaved slices, where we do odd-numbered slices (i.e., 1, 3, 5) and
then go back and do evennumbered slices (i.e., 2, 4, 6).
9. Number of slices (coverage)
The number of slices we can do within any TR is limited by the dead
time after (TE + Ts/2 + To). Furthermore, if we choose to have two
echoes per TR (as in a dual-echo, spin-echo sequence), the number of
slices we could have would be cut back more. The formula for the
maximum number of slices one can obtain is

If we do multiple echoes (or just one long echo), the formula is

governed by the longest TE.

Examples
1. TR = 1000, TE = 35 msec, Ts = 10 msec, To = 10 msec (short TE)

2. TR = 1000, TE = 75, Ts = 10 msec, To = 10 msec (long TE)

We usually don't know what the sampling time (Ts) or overhead time
(To) is; therefore, a rough approximation is

Maximum number of slices < TR/TE.

It is said that in a double-echo sequence, the first echo is “free.” This

means that the number of slices in a double-echo sequence is
determined only by the TE of the second echo
TR = 1000, TE1 = 30, TE2 = 80
Maximum number of slices

In fact, the maximum number of slices is determined by TR, and the

time it takes to get one line in the data space as follows:
Maximum number of slices < TR/time to get one line of signal into the
data space = TR/(TE + Ts/2 + To)
Also remember that each slice has its own data space (and, thus, its
own k-space), and if we are doing a double-echo sequence, each echo
has its own data space (k-space). The signal obtained from each slice
during the same TR will be obtained with the same phase-encoding
step. Thus, the equivalent line in each data space of the slices for a
given TR will be subject to the same dephasing effects of the phase-
encoding gradient (Fig. 12-9).

Figure 12-9. Each slice has its own k (data) space.

Figure 12-10. During slice selection, the transmit BW of RF pulses

 remains the same but their center frequencies vary.

10. Center frequency

A few more words about the center frequency and transmit bandwidth.
As we go from slice to slice, the RF center frequency (i.e., the Larmor
frequency) changes, but the bandwidth remains the same. In Figure 12-
10, the bandwidth (the range of frequencies) remains the same. The
center frequency of the bandwidth changes: it increases as we go
higher on the magnetic field gradient and it decreases as we go lower
on the magnetic gradient.
The bandwidth (range of frequencies) should be constant because we
want each slice to be of the same thickness. At the low end of the
magnetic gradient the frequencies are lower, and at the top end of the
gradient the frequencies are higher. Thus, as we go up the gradient, we
approach higher center frequencies for the RF pulse; however, the
bandwidth won't change.

Figure 12-11. An inverse relationship exists between the duration

of a pulse and its BW.
If the duration of the RF pulse is short, then its bandwidth will be wide
and vice versa (Fig. 12-11). In our pulse sequence, we use the same RF
bandwidth (i.e., the same RF duration) each time, but the center
frequency is changed. Think of the center frequency as a carrier
frequency. We are sending the same signal, but we are carrying it at a
different center frequency, and the bandwidth around it is the original
signal (Fig. 12-12).
If we use a wider bandwidth with the same center frequency, then we
get a thicker slice; a narrower RF pulse whose FT has a wider
bandwidth gives us a thicker slice. A carrier is a signal that determines
the center frequency. For more details, refer to the discussion on
bandwidth in Chapter 10.
Figure 12-13A displays a typical signal from the echo. After the signal is
received at the receiver coil, it is digitized (through signal sampling)
because the computer analyzing the signal can only work with digitized
numbers. The vertical bars in Figure 12-13A demonstrate the sampling
procedure. Instead of having a continuum of signal amplitudes, samples
of the signal at certain time intervals (usually equidistant intervals) are
taken.
The computer now only has these discrete (as opposed to analog)
values. Each value is represented in the computer as a binary number
(0s and 1s) so the computer doesn't deal with the whole signal—just
discrete samples of the signal. A computer bit is either a 0 or a 1. A
byte consists of 8 bits and represents the basic building block of
computer characters. Each sample of an analog signal is encoded into a
series of bytes, which is recognized by every computer. This process is
the basis for ADC.
 Figure 12-12. The FT of RF pulses—the BWs are the same, but the
center frequencies are different.

The idea behind ADC is to just use the samples of a

signal and be able to reconstruct the original signal
from its samples.

Let's consider a sinc signal with frequency ω. As we have seen in

previous chapters, this signal has a rectangle-shaped FT (Fig. 12-13B).
The time between successive sampling points is called the sampling
interval δTs (or dwell time).
δTs = sampling interval
After it is sampled, the aforementioned signal will look like Figure 12-
14A. The envelope of this function is a curve connecting the sample
points (which resembles the original sinc function). The FT of this
sampled signal is given in

Figure 12-14B. This FT looks like a periodic version of the original FT

(Fig. 12-13B) because the FT of a discrete and periodic function is also
periodic (see later text). If δTs is the sampling interval, the first center
frequency is given by the following formula (Fig. 12-14B):
 Figure 12-13. The FT of a sinc function (A) contains a single
rectangle (B).

Center frequency = 1/δTs

If we make the sampling interval very short, we will spread out the
square waves in the FT, that is, if we take several samples per cycle,
the square waves spread out (Fig. 12-15). If we make the sampling
interval very wide, the square waves of the FT are going to get closer,
that is, if we take few samples per cycle, the square waves get closer
(Fig. 12-16).

Aliasing
We don't want to take too few samples per cycle, that is, we don't want
the sampling interval to be too wide because then the square waves
will overlap and aliasing will result (Fig. 12-17).
Why is the FT of a continuous sinc function a single wave, whereas that
of its sampled version shows repetitive square waves (Fig. 12-18)? The
reason for this is mathematical. We will try to give an explanation that
doesn't involve the complicated mathematics behind it.

Figure 12-14. The FT of a sampled (discrete) sinc function (A)

 contains a series of rectangle. B: The midpoint of each rectangle is
a multiple of 1/δTs.

To sample the signal, we have to multiply the signal by a sequence of

spikes (each spike is called a delta function). Each spike is separated
from the next by the sampling interval (δTs) (Fig. 12-19).
Because the value of the series of spikes is zero everywhere between
the spikes and is positive where the spikes are located, then
multiplying the signal by the spikes will result in a value of 0
throughout the signal except where the spikes are located (Fig. 12-20).
Now, the FTs of a series of spikes are also a series of spikes (Fig. 12-21).
Whereas the spikes are separated by δTs, the FTs of the spikes are in
the frequency domain and are separated by 1/δTs (Fig. 12-21). The FT
of the signal multiplied by the spikes is the FT of the signal convolved
with the Fourier transfer of the spikes. (This has to do with a
mathematical operation called convolution.) Convolution basically
means that we take the FT of the signal and center it around each spike
(Fig. 12-22).

Figure 12-15. A short sampling interval δTs (i.e., taking more

samples) causes the rectangles to spread out.
Figure 12-16. A longer sampling interval (i.e., taking fewer
samples) causes the rectangles to move closer to one another.

Figure 12-17. When the sampling interval is too long (i.e., not
enough samples are taken), the rectangles may overlap (causing
aliasing).
Figure 12-18. The FT of a sinc function is a rectangle, but that of
its sample variant is a series of rectangles.
Figure 12-19. To sample a continuous, analog signal, a series of
spikes (called delta functions) is multiplied by the signal.

Figure 12-20. After multiplying the signal by a series of spikes, the

result is zero except at the points of the spikes. The value of each
resulting spike is equal to the value of the signal at that point.
Figure 12-21. The FT of a series of spikes (separated by δTs) is also
a series of spikes (separated by 1/δTs).
Figure 12-22. The FT of the product of a signal and a series of
spikes is the FT of that signal (e.g., a rectangle) “convolved” with
a series of spikes. The result is the FT of that signal replicated an
infinite number of times.
Figure 12-23. When the previous FT is passed through a low-pass
filter, the result is the desired FT.

Figure 12-24. FT of cos ω0t.

 Figure 12-25. A, B: The FT of a sampled cosine signal.

Eventually, we want to get back our original signal. If we take these

repetitive FTs and pass them through a low-pass filter (LPF), we will
ultimately retrieve our original signal (Fig. 12-23). But remember, we
really don't need to know the above mathematics to understand the
principles of sampling.
Let's consider an easier signal: a cosine function cos ω0t and its FT (Fig.
12-24). The sampled version of this signal (with four samples per
interval) is shown in Figure 12-25A with its FT in Figure 12-25B. Notice
how the FT of this digitized cosine function has multiple replicas.
Briefly,
1. The cycle is 4-sec long.
2. δTs is 1 sec.
3. Frequency = number of cycles/sec = 1/4 cycle/sec = 1/4 Hz.
4. 1/δTs = 1/1 sec = 1 Hz.

On the FT (Fig. 12-25A), we have the center frequency 1/4 Hz,

corresponding to one replica, and the center frequency 1/δTs (= 1 Hz),
corresponding to the second replica. If we now pass this FT through an
LPF as in Figure 12-26 to eliminate all the high frequencies, we will
recover our original pair of frequency

spikes, which is the FT of the original cosine wave.

 Figure 12-26. The previous FT passed through an LPF gives the FT
of the original cosine function.

Visually, if we had the four sampling points in Figure 12-25A, we could

still see quite easily what the original signal should look like (Fig. 12-
27) by connecting the dots. It would be easier if we had more samples
because the more samples we use, the easier it is to see the shape of
the original signal (Fig. 12-27B).
 Figure 12-27. A, B: If enough samples are taken, the original signal
can be visually reconstructed by connecting the dots. The more
samples taken, the easier it is to visualize the original signal.

Let's now take another example in which we have fewer than four
samples. Let's try two samples per cycle (Fig. 12-28A). In this case,
1. The period is still 4 sec.
2. δTs = sampling interval is 2 sec.
3. 1/δTs = 1/2 = 0.5 Hz.

Let's see the FT of this example (Fig. 12-28B). Frequency is still 1/4.
The sampling interval is
 now 2 sec, so 1/δTs = 1/2 Hz. We show two spikes on either side of 1/2
Hz, which is the frequency range around the center frequency of 1/2
Hz.

Figure 12-28. A, B: If exactly two samples are taken per cycle of a

cosine function, the spikes will overlap exactly and the original
signal is still decipherable.

We see now that the pair of spikes centered at 1/2 Hz come right next
to the original pair of spikes of 1/4 Hz. They practically overlap. If we
pass this FT through an LPF to eliminate the higher frequencies, we still
recover the two frequency spikes, which are the FTs of the original
cosine signal; however, they will be increased in amplitude because of
the overlap (Fig. 12-29).
Now let's see what happens if we take even fewer samples (Fig. 12-
30A). Here,
1. The period is still 4 sec.
2. But δTs = 3 sec.
3. 1/δTs = 1/3 Hz.
4. The frequency is still 1/4 cycles/sec.
So, let's look at the FT (Fig. 12-30B). If we put this FT through an LPF,
the spikes associated with the center frequency (1/δTs) will interfere
with the original signal (Fig. 12-31). We will not only have the original
spikes but also have two extra spikes that are closer to the center. Now
we have the sum of two cosine waves rather than the spikes of the
single original cosine wave. We have the spikes of another cosine wave,
which are closer together, meaning that its transform is a cosine wave
of lower frequency. This is called aliasing.
What we wanted to approximate by sampling was the original signal,
having a frequency of 1/4 Hz. But as a result of undersampling, we also
obtain an undesired signal with a much lower frequency (Fig. 12-32).
Aliasing: An Analogy. As an analogy, let's consider a stage-coach wheel
in a Western movie. Sometimes, it appears as if the wheel is turning in
the reverse direction. Let's see how this happens. When we take a
motion picture, we are actually taking samples in time. Let's take one
point on the wheel and take a sample at time t = 0 (Fig. 12-33). At a
later time (t1), we take another frame. The point rotates a certain
distance forward as the wheel turns. When we watch this in a motion
picture we see this point rotating in a clockwise direction.
Figure 12-29. An LPF applied to the previous FT allows recreation
of the original signal (although the height of the spikes is doubled).

Figure 12-30. When too few samples are taken (A), the spikes will
get mixed up (B).
Figure 12-31. When a (low-pass filter) LPF is passed through the
previous example, two sets, instead of one set, of spikes are
produced, which would yield a different signal than the original
cosine function. This is called aliasing.
Figure 12-32. When too few samples are taken, the perceived
(aliased) frequency is different from the actual (original)
frequency.

Figure 12-33. Analogy of aliasing. If the motion picture frames of a

stage-coach wheel are taken fast enough, the wheel appears to
rotate correctly in the clockwise direction.
 Figure 12-34. If the frames are not taken fast enough, the wheel
will deceptively appear to be rotating in the counterclockwise
direction.

Now let's see what happens when we undersample (Fig. 12-34). Let's
say the first frame starts at the same point as before at t = 0. But on
the next frame, we wait longer (t2) to sample and don't sample until
the point almost comes around to its original spot. We wait a similar
time for the next sample and the point again comes around almost to
the spot on the second frame. If we take a motion picture of this, the
wheel will appear as if it's turning counterclockwise. This is because we
are undersampling. This is an example of aliasing. The wheel is actually
turning in a forward direction, but because of undersampling, it
appears to be doing something else.

Undersampling causes aliasing.

Aliasing comes from the term alias: a fake name. We have a real
frequency, but because of undersampling, we appear to have a fake
frequency. In the previous example in which the cosine wave was
undersampled, the real signal was a cosine wave with a frequency of
1/4 Hz,

but the “alias” cosine wave had a lower frequency (Fig. 12-31). In the
example of the wagon wheel, the real rotation was in a clockwise
direction, but the alias was in a counterclockwise direction, probably
with a slower rotation.
 Figure 12-35. Nyquist law: To avoid aliasing, the maximum
frequency in the signal (fmax) should be less than half the sampling
frequency (1/δTs). In other words, the sampling interval (δTs)
should be at least twice the minimum period (1/fmax). That is, at
least two samples per cycle (corresponding to the highest
frequency in the signal) are required to avoid aliasing.

Sampling Theorem (Nyquist Law)

Consider Figure 12-35. The sampling theorem states:

Nyquist law: If ωmax is the maximum frequency in the

signal, then the sampling rate must be at least twice
the maximum signal frequency to avoid aliasing, that
is, ωsampling = 1/δTs ≥ 2ωmax.

This is easy to see in the diagram (Fig. 12-35). We want the sampling
rate to be at least equal to the sum of the maximum frequencies of
adjacent boxes (to keep the boxes from overlapping).
 Figure 12-36. The original signal can be reconstructed with a
minimum of two samples per cycle.

In other words, the sampling rate should be at least twice ωmax. In

terms of sampling interval (δTs), it should be less than half of the
period of the signal (remember that δTs = 1/ωmax):
δTs <1/2 (period)
This is the Nyquist theorem. Basically, it means that if we want to
recover a signal from its samples, we need to take at least two samples
per cycle. We can take as many samples as we like, but sampling takes
time, so we want to take the minimum number of samples necessary to
recover the signal accurately from its FT. We saw diagrammatically
how, with a minimum of two samples per cycle, we could accurately
approximate the original cosine signal (Fig. 12-36).
Nyquist Theorem The maximum frequency we can recover is one-half
of the sampling rate.
Max frequency = 1/2(δTs)-1 or 1/δTs = 2(Nyquist frequency)
Question: What is the difference between sampling interval and
sampling time?
Answer: Sampling interval (δTs) is the time between sampling points
(Fig. 12-37). When we sample a signal, we can't possibly take an infinite
number of samples. So we take (N) samples and then we stop. This
gives the sampling time (Ts) the time it takes to sample the entire
signal.
Ts = N × (δTs)
The sampling time (Ts) is the sampling interval (δTs) multiplied by the
number of samples taken (N). The MR machine samples at a certain
interval

(say, about δTs = 50 µsec). Let's say we have 256 frequency-encoding

steps (N = 256). Then, sampling time = 256 × 50 µsec ≅ 13 msec.

Figure 12-37. The sampling interval δTs is the time interval

between two successive samples. The sampling time is the product
of δTs and the number of samples Nx, that is, Ts = Nx · δTs.

Let's prove the following:

Bandwidth = 1/sampling interval
Bandwidth =1/δTs
If the samples are close together, we get a higher bandwidth. If the
samples are further apart, we get a lower bandwidth. Earlier, we were
dealing with the bandwidth of an RF pulse. Now we are talking about
the bandwidth of the received signal, that is, the bandwidth of the
echo. The Nyquist theorem says that we should do a minimum of two
samples per cycle to reconstruct the original signal. Is there any benefit
in doing more samples per cycle?
The RF bandwidth is the “transmission” bandwidth. However, the
bandwidth associated with the sampled signal is the “receiver”
bandwidth. To prove the above relationship between bandwidth and
δTs, recall that frequency (which is number of cycles per second) is the
reciprocal of time (1/time).
δTs is the sampling interval (in the time domain), 1/δTs = frequency,
and if we're operating at the Nyquist frequency,
1/δTs = receiver bandwidth (Fig. 12.38)

Figure 12-38. At Nyquist frequency, 1/δTs = ω;max or δTs =

1/(2ωmax).
From the diagram, we see that if we operate at the Nyquist frequency,
then the bandwidth is equal to 2 × (maximum frequency ωmax). Thus,
Bandwidth = 2(ωmax) = 1/δTs
We generally do not want to take more than the minimum number of
samples to recreate the signal because more samples require more
time. We want to operate as efficiently as possible, that is, take as few
samples as possible, without causing aliasing.
We can, under certain circumstances, take more samples than two
samples per cycle. For example, when we activate the feature on the
scanner to avoid wraparound along the frequency-encode direction, the
scanner automatically performs oversampling to prevent aliasing. This
is done to play it safe by doubling the number of samples taken.

Digitized k-Space
We've already shown how we fill lines of the data space with our signal,
with each line performed using a different phase-encoding step. What
we actually put into each line of the data space are the sampled data
from each signal. Once we put these samples into each line of data
space, we complete the data space.
All MRI machines operate under the same principle when it comes to
sampling the signal. A minimum of two samples per cycle is taken and
put into the data space (Fig. 12-39A). If more than two samples per
cycle are taken, then the bandwidth is wider, but it won't give us a
better approximation of the signal (Fig. 12-39B). If we take less than
two samples per cycle, the adjacent bandwidths may overlap and cause
aliasing (Fig. 12-39C).
 Figure 12-39. A: Operating at Nyquist frequency. B: When more
samples are taken, the BW will be increased. C: When too few
samples are taken, aliasing may occur (Ts3 > Ts1 > Ts2).

Signal-to-Noise Ratio
When the bandwidth is narrowed, the signal-tonoise ratio (SNR or S/N)
is increased. (This topic is discussed at length in a later chapter.) SNR is
inversely proportional to the square root of the bandwidth. SNR is also
proportional to the volume of the pixel and to the square root of the
number of phase-encoding steps (Ny), the number of frequency-
encoding steps (Nx), and the number of excitations (NEX):

Therefore, when we activate the feature on the scanner to lower the

bandwidth (BW) on the second echo of a dual-echo T2-weighted spin-
echo image, we increase the SNR on the second echo. Remember that
when we decrease the bandwidth, we are increasing the sampling
interval (δTs). By increasing the sampling interval, we necessarily
increase the sampling time (Ts) because Ts =N(δTs). We do this only on
the second echo because the magnitude of the signal is always weaker
on the second echo, so we want to increase the SNR on the second
echo. However, as we decrease the bandwidth to improve SNR, we are
necessarily decreasing the number of samples per cycle; thus, we are
increasing the possibility of aliasing. This is the reason why there is a
limit on how much the bandwidth can be decreased. If we sample too
fast, the bandwidth is wider and SNR goes down. If we sample too
slowly, the bandwidth is narrow and we may get aliasing.
If we increase the sampling time (Ts), we then decrease the number of
slices we can take per TR because of the following formula:

↓ BW → ↑ δTs →Ts since Ts = N(δTs). Also ↓BW → ↓

number of slices since ↑Ts →↓number of slices.

Sampling of Composite Signals

Let's discuss the sampling theorem in the case of a more complicated
signal such as in Figure 12-40. When we have a composite signal (i.e., a
signal composed of two or more frequencies), we need

to take a minimum of two samples per cycle of the highest frequency

present in the signal.

Figure 12-40. An example of a composite signal.

Presume that the complex signal above is a combination of three

different cosine waves, each of a different frequency (Fig. 12-41). We
have three different signals that are added up to give us a certain
signal. The sampling theorem refers to the component of the signal that
has the highest frequency. Therefore, we want to take two samples per
cycle of the highest frequency component of the signal; this obviously
means more than two samples per cycle for the other lower frequencies
of the signal.
We are taking samples of the composite signal. Therefore, even though
we take a minimum of two samples per cycle of the highest frequency
component of the signal, we will necessarily be taking more than two
samples per cycle of all the lower frequency components of the signal.
 Figure 12-41. In a composite signal, two samples per cycle of the
highest frequency component are required to avoid aliasing.

Nyquist sampling theorem: A minimum of two

samples per cycle (corresponding to the highest
frequency present in the signal) are required to
reconstruct the original signal accurately from its
samples.

Example 1
a. At 1.5 Tesla, it typically takes 8 msec to perform one readout: Ts = 8
msec.
b. We have a matrix of, say, 256 × 256 pixels.
 By convention, the first number refers to the number of frequency-
encoding steps. The second number refers to the number of
phaseencoding steps. What is the bandwidth? We know that
BW = 1/δTs

What is δTs?
δTs = sampling time/number of samples we test = 8 msec/256 samples
BW = 1/δTs = 1/(8 msec/256) = 256/8 msec = 256/0.008 sec = 32,000 Hz
= 32 kHz = ± 16 kHz
This is a typical bandwidth for a typical readout with 256 frequency-
encoding steps and a sampling time of 8 msec. Therefore, a frequency
bandwidth of 32 kHz (±16 kHz) is a fairly typical frequency bandwidth
that we deal with in routine imaging. This means that the bandwidth
extends to +16 kHz on the right and -16 kHz on the left side of the
center frequency.

Example 2
What would happen if we go to a 512 × 512 matrix, that is, with 512
frequency-encoding steps?
a. We could either have a larger frequency bandwidth:

b. Or, we could double the sampling time and keep the sampling
interval (and the bandwidth) the same.
(We'll get into the relationship between the field of view and
bandwidth later.)
Key Points
In this chapter, we have presented the basic concepts
of signal processing, of which image processing is a
subset. As mentioned in the introduction, the
understanding of these concepts is crucial in
understanding the intricacies of image optimization,
which is one of the goals of every imager. Again,
memorizing the formulas is not as important as
understanding the concepts behind them.
Let's summarize:
1 ADC (analog-to-digital conversion) is the process in
which an analog (time-varying) signal is encoded to a
digital signal (containing a series of binary numbers
0s and 1s) represented as bytes (8 bits) in the
computer.
2 This is done by sampling the signal.
3 To be able to reconstruct the original signal from its
discrete samples, the Nyquist law must be satisfied.
Otherwise, aliasing will occur.
4 The Nyquist theorem states that the sampling
frequency must be at least twice the highest
frequency present in the signal.
5 Stated differently, if you take the waveform of the
component signal with highest frequency (remember
that each signal is a composite of many different
signals with varying frequencies), then at least two
samples per cycle are required to avoid aliasing.
6 Therefore, aliasing occurs because of
undersampling. To ensure that aliasing will not
happen, MR scanners may automatically perform
oversampling.
 7 The bandwidth (BW) is defined as the range of
frequencies in the signal.
8 BW = 1/δTs, where δTs is the sampling interval
(interval between two samples).
9 At Nyquist frequency, BW = 2ωmax, where ωmax is
the highest frequency in the signal, so that δTs =
1/BW = 1/(2ωmax).
10 Sampling time Ts = Nx · δTs, where Nx is the
number of frequency encodes.
11 SNR is given by

12 So, SNR is ↑ if BW ↓
13 A narrower BW is used when a higher SNR is
desired (e.g., on the second echo of a dualecho SE
image).
14 BW is ↓ if δ Ts is ↑(i.e., less samples are taken),
which may cause aliasing!
15 Now, if ↑ Ts is ↑ then Ts = Nx · δTs is ↑, which
causes TE to ↑. Because
Number of slices ≅ TR/TE then a narrower BW will
reduce the coverage.

Questions
12-1 According to the Nyquist theorem, to avoid aliasing:
(a) At most, two samples per cycle corresponding to the highest
frequency are required.
(b) At least two samples per cycle corresponding to the highest
frequency are required.
 (c) At most, two samples per cycle corresponding to the lowest
frequency are required.
(d) At least two samples per cycle corresponding to the lowest
frequency are required.

12-2 According to the Nyquist theorem, to avoid aliasing:

(a) The sampling frequency must be at least half the highest
frequency in the signal.
(b) The sampling frequency must be at least twice the lowest
frequency in the signal.
(c) The sampling frequency must be at least twice the highest
frequency in the signal.
(d) The sampling frequency must be at least half the lowest
frequency in the signal.

12-3 T/F The bandwidth (BW) is the inverse of the sampling interval
(δTs).

12-4 T/F SNR is directly proportional to 1/BW.

12-5 A narrower BW (all other things remaining unchanged) will

result in
(a) more SNR
(b) less coverage
(c) longer sampling time
(d) all of the above
(e) only (a) and (b)

12-6 T/F Aliasing occurs because of oversampling. Questions

13
Data Space

Introduction
Before we can understand k-space, we need to discuss the data space,
which is a matrix of the processed image data. To most radiologists,
kspace is in the twilight zone! We have already seen some of the basic
concepts of the data space and k-space in the previous chapters. In this
chapter, we will learn some of the properties of k-space in greater
detail. The understanding of k-space is crucial to the understanding of
some of the newer MRI fast scanning techniques such as fast spinecho
(FSE) and echo planar imaging (EPI).
 Figure 13-1. The data space (with both axes as time variables) is
an “analog” version of k-space.

Where Does k-Space Come From?

k-Space is derived from the data space, so it's really not as intimidating
as it initially appears. Figure 13-1 demonstrates a typical
representation of the data space with a 256 × 256 matrix.
1. Figure 13-1 is an “analog” version of kspace. The true k-space, as
we'll see later, is a digitized version of this figure, with axes referred
to as spatial frequencies.
2. In Figure 13-1, we have 256 phase-encoding steps. We keep the zero-
 step (i.e., no

phase encoding) in the middle of k-space, so we go from −127 phase

encode to +128 phase encode (bottom to top).
3. We also have 256 frequencies.
4. The y-axis, then, is the phase-encoding direction.
5. In the center, we put the signal acquired with no phase-encoding
gradient.
6. As one advances on the y-axis, each set has signal acquired with an
increasing phaseencoding gradient with maximum gradient at +128
phase-encoding step. Likewise, as one goes down from zero gradient
in the y-axis, each step has signal acquired with an increasing phase-
encoding gradient in the opposite direction with maximum gradient
at −127 phase-encoding step.
Let's now go back and review the spin-echo pulse sequence (Fig. 13-2).
1. We apply the 90° pulse using an appropriate slice-select gradient, Gz.
2. Next, we apply a 180° pulse and, after a time, TE receives an echo.
3. During this time, we apply the readout gradient, Gx.
4. Then we place a sampled version of this echo in one of the rows in k-
space. Let's say that this echo was obtained without using a phase-
encoding gradient in the y direction. We sample the signal and then
put it into the zero line in the data space.
5. With, say, a 256 × 256 matrix, we take 256 samples. Each of the 256
points in a row of the data space is a sample of the echo. (It's hard to
draw discrete samples, so we'll draw continuous signals in the data
space rows, realizing that each point in a row is a digitized sample of
the signal.)
 Figure 13-2. A spin-echo pulse sequence diagram.

6. For the second row in the data space, we do the exact same thing,
except in this step the signal is obtained using a largar phaseencoding
gradient in the y-axis.
Remember that the phase-encoding gradient causes dephasing of the
signal. Therefore, the signal for the second line of the data space will
be similar in shape to the first signal (because both are signals from the
same slice of tissue, just obtained at a different time) but smaller in
magnitude than the first signal (because it undergoes additional
dephasing due to the phase-encoding gradient). Thus, when we draw
this signal into the second line in the data space, we see that it is
similar in shape to the first signal, but slightly weaker—because it's
been dephased.
The signal that goes into the last line of the data space (+128) will be
almost flat because it has undergone maximum dephasing; likewise, as
we alternate to the signals placed below the zero line (i.e., −1, −2, …,
−127), a certain symmetry results. For instance, line (−1) is similar in
strength to line (+1) in that, whereas line (+1) experiences mild
dephasing due to a slight increase in magnetic field strength, line (−1)
experiences similar mild dephasing due to a slight decrease in magnetic
field strength. Likewise, the signal that goes into the first line in the
data space (−127) will be almost flat due to maximum dephasing in the
opposite direction of line (+128).

Remember that each line in the data space contains the signal obtained
from the entire image slice during a single TR. Each TR is obtained
using a different phase-encoding step in the y-axis.
Question 1: How long does it take to go from one row in the data space
to another?
Answer: It takes the time of one TR.
Question 2: How long does it take to go from one point (sample) in a
row to the next point (sample) in the same row of the data space?
Answer: It takes the time spent between samples, that is, the sampling
interval (ΔTs).
Question 3: How long does it take to fill one row of the data space?
Answer: Let's say that ΔTs [all equal to] 31 µsec and that there are 256
(N) samples along the readout axis. The sampling time is Ts = (ΔTs)(N) =
(31µsec)(256) = 7.9 msec
Therefore, it takes about 8 msec to fill one line of the data space. In
general, it takes Ts = Nx · ΔTs
to fill one line of the data space.
Question 4: How long does it take to fill one column of the data space?
Answer: It is the acquisition time Np × TR, where Np is the number of
phase-encoding steps.
Let's say TR = 3000 msec and Np = 256.
Acquisition time = (3000 msec) (256)
= 12.8 min
 If TR = 500 msec, then
Acquisition time = (500) (256) [all equal to] 2 min
It takes several milliseconds to fill one row of the data space. But it
takes several minutes to fill the columns of the data space.

Motion Artifacts
The preceding concept is one of the reasons why motion artifacts
manifest themselves mainly in the phase-encoding direction. In other
words, it takes much longer to gather the signal in the phase-encoding
direction than in the frequencyencoding direction, leaving more time
for motion to affect the image in the phase direction. Another reason,
as we shall see later, is that motion in any direction results in a phase
change; thus, motion artifact propagates along the phase-encoding
direction.

Properties of k-Space
Center of k-Space. The center of the data space contains maximum
signal. This finding is caused by two factors:
1. Each of the signals has its maximum signal amplitude in the center
column (Fig. 13-3). Recall that when we apply the 180° refocusing
pulse, the dephased signal begins to rephase and reaches maximum
amplitude when the protons are completely rephased. It then
decreases in amplitude as the protons dephase once more.
The middle column in the data space corresponds to the center of
each individual echo, and the more peripheral columns refer to the
more peripheral segments of the echoes: columns to the left of
center depict rephasing of the echoes toward maximal amplitude in
the data space; columns to the right of center depict dephasing of
the echoes away from maximal amplitude in the data space.
Therefore, as we go further out to the more peripheral columns, the
signal weakens. The most peripheral points in the signal to the left
are the weakest point of the signal as the signal just begins to
rephase (Fig. 13-3). Likewise, the most peripheral point in the signal
to the right is the weakest point after the signal has been refocused
 and then has regained maximum dephasing.
2. The maximum amplitude occurs in the center row because this line is
obtained without additional dephasing due to phase-encoding
gradients; subsequent rows with progressively larger phase-encoding
gradients have weaker signal amplitude.
Therefore, because the middle row has the strongest of all echoes and
the middle column contains all the peaks of the echoes, the center

point of the data space contains maximum amplitude, that is, maximum
signal-to-noise ratio (SNR) (Fig. 13-4).

Figure 13-3. The peripheral points in the signal have the weakest
amplitude; the center point has the maximal amplitude.

As we go farther out to the periphery in both directions, the signal

 weakens:
1. In the y direction because of progressively larger phase-encoding
steps
2. In the x direction because the echo signal has either not yet reached
maximum amplitude or is losing maximum amplitude due to
dephasing

Figure 13-4. The center of k-space always contains maximum

signal.

Image of k-Space. Because of the oscillating nature of the signals, the

image of the data space (and, thus, k-space) will appear as a series of
concentric rings of signal intensity with alternating bands of high and
low intensity as the signal oscillates from maximum to minimum, but an

overall decrease in intensity as one goes from the center to periphery

(Fig. 13-5A). So, the white and dark rings in k-space correspond to the
peaks and valleys of the echoes, respectively. The original raw data (k-
space) and the original image are shown in Figure 13-5.

Figure 13-5. A: The original raw data (k-space) of (B) the original
image (midline sagittal T1-weighted image of the brain).

Edges of k-Space. You might be thinking that if the center of k-space

contains the maximum signal, why not eliminate the periphery of the
signal and just make an image from the central high signal intensity
data (Figs. 13-4 and 13-6A)? We can actually make an image with this
data, but the edges of the structures imaged will be very coarse (Fig.
13-6B). The periphery of k-space contributes to the fine detail of the
image. Let's see how this happens.
 Figure 13-6. A: k-Space. B: The image constructed from only the
center of k-space. The details are reduced due to exclusion of
peripheral points in k-space.

Question: What type of information is available in the periphery of k-

space?
Answer: The periphery of k-space provides information regarding
“fineness” of the image and clarity at sharp interfaces.
Recall the following Fourier transforms of a sinc wave and its truncated
version from an earlier chapter (Fig. 13-7). As you can see, by
truncating

the signal (echo), ring artifacts are introduced in the Fourier transform.
Therefore, by eliminating the samples in the periphery of the data
space, the sharp interfaces in the image are degraded and the image
gets coarser. In other words, the fine detail of the image is
compromised when the edges of k-space are excluded. Figure 13-8B is
the image corresponding to the periphery of k-space (Fig. 13-8A).
 Figure 13-7. A: The FT of an ideal sinc function is a rectangle. B:
The FT of a truncated sinc function has ring down effects.

Image Construction. We can take a single line in k-space and make a

whole image. It wouldn't be a very pretty image, but it would still
contain all the information necessary to construct an image of the
slice.

Figure 13-8. A: k-Space. B: The image constructed from only the

 periphery of k-space. This image has minimal signal but contains
details about the interfaces in the original image.

There is absolutely no direct relationship between the center of k-

space and the center of the image. Likewise, there is no direct
relationship between the edges of k-space and the edges of the
image.
A point at the very edge of k-space contributes to the entire image. It
doesn't contribute as much to the image in terms of the SNR as does a
point

in the center of k-space because the center of k-space has maximum

signal, but the peripheral points still contribute to the clarity and
fineness of the image.

Figure 13-9. There is a one-to-one relationship between frequency

and position along the x-axis and between phase-encoding gradient
increment and position along the y direction.

Once we have all the data in k-space, we take the Fourier transform of
k-space to get the image.
Question 1: Why is the Fourier transform of k-space the desired image?
Answer: Because there is a one-to-one relationship between frequency
and position in the x direction and between phase-encoding gradient
strength1 and position in the y direction.
Question 2: Why is there a one-to-one relationship between frequency
and position?
Answer: Because, in our method of spatial encoding, we picked a linear
gradient in the x direction that correlated sequential frequency
increments with position; likewise, we picked a linear gradient in the y
direction that correlated sequential phase gradient increments with
position in the y direction (Fig. 13-9).
Thus, the center of the field of view of the image experiences no
frequency gradient and no phase gradient, and the points in the
periphery of the image experience the highest frequency and phase
gradients. In other words, there is a 1:1 relationship between frequency
and position in the image.
In summary, the frequency- and phaseencoding gradients provide the
position of a signal in space. They tell us which pixels each component
of the signal goes into in the slice under study.
Question: How are the shades of gray determined?
Answer: The shades of gray are determined by the magnitude or
amplitude of the signal (actually its Fourier transform) at each pixel.
Recall that the image of k-space looks like a series of concentric circles
of alternating intensity on a two-dimensional surface. If we now
incorporate amplitude as a third dimension, we would have the areas of
greater amplitude coming off the surface of k-space toward us like a
“warped” image (Fig. 13-10).
k-Space Symmetry. One step needs to be completed after receiving the
signal and before placing it in k-space that we have so far ignored. This
step is called phase-sensitive detection. We want to take the echo
signal, which is on a carrier frequency, shift it to zero frequency, and
divide the signal into its real (cosine) and imaginary (sine) components.
First, we start off with the signal that is being frequency- and phase-
shifted around a carrier frequency of 64 MHz for a 1.5-T magnet.
 However, it's hard to tell whether a signal has been frequency- or
phase-shifted unless we “ground” the signal back to “zero.”

Figure 13-10. A three-dimensional line drawing of k-space.

Therefore, we first subtract the carrier frequency of 64 MHz from the

signal (Fig. 13-11). We first take the signal and subtract a cosine wave
of center frequency ω0 from the signal. Then we take the signal and, in
a separate computation, subtract the sine wave of center frequency ω0
from the signal.
If we subtract ω0 from the signal, we center the signal at zero (in the
frequency domain). We then have a resultant signal whose center
frequency is 0. Perform this step twice to separate the signal into its
real component (cosine) and its imaginary component (sine).
Each data space has two components:
1. The data space with “real” (cosine) data: the signal that has (cos
ω0t) subtracted from it and brought back to 0 frequency.
 Figure 13-11. The process of image construction includes a
preliminary decomposition of signal into its real and imaginary
components. This in turn yields a real and an imaginary k-space,
that is, a real and an imaginary image.

2. The data space with “imaginary” (sine) data: the signal that has (sin
ω0t) subtracted from it and brought back to 0 frequency.

We now have two data spaces (Fig. 13-12): one with cosine data (real);
one with sine data (imaginary). Both have data centered at 0
frequency. In the data space with cosine data, we know that a great
deal of symmetry exists. A cosine function is an example of an even
function. If we look at a cosine function, we see that there is symmetry
to the right and left of zero. In addition, there is symmetry above and
below zero. Thus, if we put a pixel in a line of the data space to the
right of the 0 column, and above the 0 line (point a), the symmetry of
the cosine function would make us unable to discriminate between the
other (a) positions. The computer

couldn't tell the difference between any of the four pixel positions. This
is why we use the sine version of the data space.
 Figure 13-12. A, B: Spatial direction in k-space. The imaginary k-
space provides a series of left-right or up-down directions.

Now look at the data space with the sine data (Fig. 13-12B). Again, the
pixel is in the same place as the cosine data space; it is in a line of the
data space above the 0 line and to the right of the 0 column. However
(unlike the cosine data space), here we can distinguish it from the pixel
below the 0 line (−b). We can also differentiate it from the pixel to the
left of the 0 column (−b).
Why are these pixels different in the sine data space? Let's review the
sine function (sin ω0t) of the two pixels to the right of the 0 frequency
(Fig. 13-13A). The sine function is an example of an odd function
because of its inherent antisymmetry. The sine function changes
polarity above and below the 0 line. This allows us to differentiate the
two pixels.
Now, let's examine the sine function (sin ω0t) above the zero line of the
data space (Fig. 13-13B). Again, because of inherent antisymmetry of
the sine function on either side of the 0 frequency, the pixels will have
opposite polarity.

Complex Numbers
We said in Chapter 1 that a complex number can be divided into its real
(cosine) and imaginary (sine) parts. If we consider the cosine function
as the real component and the sine function as the imaginary
component, then we can add the sine and cosine together to get the
magnitude of the signal as well as its direction.
So now let's add up the data of the 4 pixels (Fig. 13-14). Because of the
changing polarity of the sine function, when we add the sine function
to the cosine function, we can distinguish the direction of the four
pixels (whereas with cosine function alone, we couldn't tell the
direction).
In the lines above 0 phase encoding:
Pixel a − ib is to the left of the 0 frequency.
Pixel a + ib is to the right of the 0 frequency.
In the lines below 0 phase encoding:
Pixel a + ib is to the left of the 0 frequency.
Pixel a − ib is to the right of the 0 frequency.
Conjugate (Hermitian) Symmetry. The conjugate of a complex number
a + ib is the complex number a − ib (i.e., with the same real component
but a negative imaginary component). From this and from Figure 13-14,
it is clear that k-space possesses conjugate symmetry, also known as
Hermitian symmetry.
Half NEX (½ NEX). In a “½ NEX”2 technique (half-Fourier in phase), we
acquire the data from the upper half of k-space and construct the
lower part mathematically (Fig. 13-15), thus reducing the scan time.
The trade-off is a reduced SNR by a factor of , to be exact (see
Chapter 17). Due to the presence of phase errors in the data, the
symmetry previously discussed may not be perfect. This is why when
employing such techniques, a few extra rows in the center of k-space—
which contains maximum signal—are always added to

allow for such phase corrections. That is, slightly more than 50% of k-
space must be sampled to maintain phase information.
Figure 13-13. A: Because the phase gradients corresponding to the
top and bottom half of k-space generally have opposite polarities,
the values in the corresponding imaginary k-space also will have
opposite polarities. B: Because the sine function is an odd function,
the left half of the signal is the reverse of the right half; thus the
corresponding points in the imaginary k-space will also have
opposite signs.

Figure 13-14. k-Space conjugate (Hermitian) symmetry can be

seen by adding the real and imaginary components of four
corresponding data points. Notice the conjugate symmetry
between left and right and between top and bottom.
 Figure 13-15. In half (or fractional) NEX, only half (or a fraction
of) of the rows in k-space (plus a few extra central rows) are used,
and the rest is constructed by symmetry.

Fractional Echo. In fractional echo, only the right half of the echo is
sampled, and the left half is constructed based on the right half (Fig.
13-16). (This allows TE to be shorter for fast scanning techniques like
turbo FLASH and Fast SPGR—see Chapter 21.)
¼ NEX. Because of the conjugate symmetry discussed previously,
theoretically you should be able to create an image using only one
quadrant of the combined real and imaginary data spaces (Fig. 13-17).
That is, you should be able to construct the entire k-space data from
only one quadrant. In reality, however, due to the presence of data
acquisition errors, perfect symmetry does not exist and doing so may
lead to phase errors and image distortion, which is probably why this
technique is not being used.

Figure 13-16. In fractional echo, only a fraction of the echo is

sampled.
 Figure 13-17. Due to conjugate symmetry of k-space, theoretically
you should be able to reconstruct the entire k-space from just one
of its quadrants. In reality, however, this may create excessive
phase errors due to actual imperfections in data symmetry.

Real and Imaginary Images. We discussed two components of the data

space, namely, the real and imaginary components. Their respective
Fourier transforms provide the real and imaginary components of the
image (Fig. 13-18).
Magnitude (Modulus) and Phase Image. Recall that given a complex
number c = a + ib, with a being the real and b the imaginary
component, the phase (angle) is given by tan θ = b/a and the

magnitude by .
This concept can be applied to the real and imaginary components of
the image (Fig. 13-18) to generate the magnitude and phase images
(Fig. 13-18).
The magnitude image (modulus) is what we deal with most of the time
in MRI. The phase image is used in cases in which the direction is
important. An example is phase contrast MR angiography, in which the
phase indicates the direction of flow, i.e., up versus down, anterior
versus posterior, or left versus right. In summary, tangent (phase angle)
= (imaginary/real), or phase angle = arctan (imaginary/real), and
 Figure 13-18. The FTs of the real and imaginary k-spaces provide
the real and imaginary images, respectively. The real and
imaginary images are used to create magnitude and phase images.

In actuality, when we do ½ NEX, we sample half of the phase-encoding

steps plus a few lines above or below the 0 line. We can then
compensate for phase errors and determine the actual phase. This is
referred to as overscanning.
Ideally, we want to have a real image with the imaginary part being
zero, and thus a zero phase. In reality, however, we have all sorts of
motion artifacts and gradient errors that create phase artifacts.
Therefore, in reality, phase is never zero. Sometimes, when service
engineers try to debug a system, they will sometimes look at the phase
image to figure out the problem.
 Figure 13-19. Magnitude image (A) and phase image (B) from a CSF
flow study in a patient with normal pressure hydrocephalus. Notice
the bright signal in the aqueduct of Sylvius (arrows) with image B
reflective of flow direction in this instance superior to inferior
flow.

We too can look at the phase image. In flow imaging, the phase image
is a velocity image and indicates magnitude and direction. For
example, in imaging the cerebrospinal fluid (CSF) flow through the
aqueduct, flow in the antegrade direction could be black, and flow in
the retrograde direction could be white on the phase images. Thus,
phase images in phase-contrast studies display the direction of flow
(Fig. 13-19).
 Figure 13-20. An example of a 2 × 4 matrix exposed to a frequency
gradient and no phase gradient.

Because the phase is never zero, we can combine the “real” image and
the “imaginary” image to get a composite image, which is the image
that we look at when we read an MR study. The modulus is the image
we look at; it combines the data corresponding to the Fourier transform
of the real and imaginary data spaces:
image = modulus
k-Space: An Example. The following is an example of a 2 × 4 matrix:

The number of frequency-encoding steps = Nx = 4.

The number of phase-encoding steps = Ny = 2.

Figure 13-21. The same example exposed to phase encoding. The

first row has no phase shift. The second row has a 180° phase shift
(thus changing the sign of the pixel values).

We will give magnitudes of either 1 or 0 to each pixel, so that:

A pixel with a magnitude = 1 will be white.
A pixel with a magnitude = 0 will be black.
In the first phase-encoding step, with no gradient in the y direction, we
apply a frequencyencoding gradient in the x direction, which allows us
to distinguish the columns. What we get then is a sum of pixels in each
column, without knowing from which row the components of the sum
originated (Fig. 13-20).
Remember that the Fourier transform can differentiate between
different columns because each column has a different frequency. The
Fourier transform of the signal will have two frequency spikes, with
column 1 having an amplitude of 1 + 0 = 1, and column 4 having an
amplitude of 1 + 1 = 2. Thus, applying the readout gradient allows us to
differentiate between different columns. However, we still haven't
differentiated between different rows. For instance, the amplitude = 2
in column 4 could be
1 + 1 or 0 + 2 or 2 + 0
With the single phase-encoding step, we have no idea how the sum of
the amplitudes is decomposed to provide the amplitude of individual
elements in each column. Remember that this first set of data was
obtained with no gradient in the phase-encoding (y) direction.
The first phase-encoding step was at zero value of the phase-encoding
gradient. For the next phase-encoding step, let's apply a gradient. The
gradient will be 360° divided by 2, or 180°. This means that the first
row will experience no gradient, and the second row will experience a
gradient such that the spins will be 180° out of phase with

the first row. This will result in no change in the numbers of the first
row. But the numbers in the second row will be 180° phase shifted
(i.e., they will be the negative of the original numbers).
 Figure 13-22. An example of a 4 × 4 matrix. The phase increments
here are 0°, 90°, 180°, and 270°. (In general, the phase increment
is 360°/Ny, where Ny is the number of phase-encoding steps.)

If we use the clock analogy to evaluate phase shift, the spins in the top
row, experiencing no phase shift, will all point upward. The spins in the
second row (which is experiencing a 180° phase shift) will all be
pointing downward (Fig. 13-21).
Thus, whereas the values in the first row will remain unchanged, the
values in the second row will be 180° reversed from what they were
with no phase shift.

Figure 13-23. The effect of phase shifts on pixel values for two
and four rows.

Row 1
Row 2 before 180° phase shift
 Row 2 after 180° phase shift

ASIDE: If we were to have four rows with four phase-encoding steps,

the steps would be: 0, 90°, 180°, and 270° phase difference between
rows, experiencing a steeper gradient with every successive TR (Fig.
13-22). In our study, with only two rows, we can only have two phase-
encoding steps:
1. Zero phase difference (no gradient) between rows.
2. With 180° phase difference, where one row experiences no phase
difference and the second row experiences a 180° phase shift from
the first row.
This division of phase-encoding steps into equal divisions of 360° all
relates to the cosine wave (Fig. 13-23). Therefore, for a phase
difference

of 180°, we get the negative value of the original number because cos
180° = −1. For a different phase angle, we would get a fraction of the
original value (from 0 to 1, or from 0 to −1, whatever cos θ is).
 Figure 13-24. The previous 2 × 4 example now exposed to a
frequency gradient and a phase gradient (i.e., 180° phase shift).

With this phase-encoding step, let's see what the Fourier transform
would be in the frequency-encoding direction (Fig. 13-24). The
amplitude in columns 1 to 3 remains unchanged from the 0 phase
readings. However, there is a change in the amplitude of column 4:
With zero phase, column 4 adds up to +2 (because 1 + 1 = 2).
With 180° phase, column 4 adds up to 0 (because 1 − 1 = 0).
With the change in phase, the second column is the sum of (+1) and
(−1), which is equal to 0. We still don't know what the original value of
each pixel was, but we do have a different total value in the Fourier
transform of this second line in k-space when we compare it with the
first line.
Now, let's do a little mathematics. The following is like solving two
 equations with two unknowns; solve for a, b, c, and d below:

Let's look back at the pixel values in the phase = 0 line and
1. Designate the pixels in the first column (a and b).
2. Designate the pixels in the fourth column (c and d).
Now look at the pixel values in the 180° phase line, and
1. Designate the pixels in the first column (a and − b).
Note that pixel (a) remains the same as 0 phase line because neither
experiences any phase change, but pixel (−b) is negative because it
experiences a 180° phase shift compared with pixel (b) in 0 phase
line.
2. Designate the pixels in column four as (c and −d).
Again note that pixel (c) remains the same as 0 phase line because
neither experiences any phase change. However, pixel (−d) is
negative because it experiences a 180° phase shift compared with
pixel (d) in phase = 0 line.

First equations: Second equations:

a + b = 1 c + d = 2

a − b = 1 c − d = 0
 Add 2a = 2 Add 2c = 2

a = 1 c = 1

b = 0 d = 1

By using the Fourier transforms of the two lines in k-space, we can

determine the amplitude values of each pixel in each column. This is
the concept of the (digital) Fourier Transform (DFT).
What is k-space in this example? Let's go back to Figure 13-20. The first
line in the data space corresponds to the sum of all the signals obtained
with 0 phase. With the gradient in the x direction (Gx) causing different
phase angles between the columns, the signal will consist of
1. (cos of column 1 frequency) with (magnitude = 1), for example, 1 cos
t
2. (cos of column 4 frequency) with (magnitude = 2), for example, 2 cos
4t
The sum of these signals will be the signal in the first line of the data
space (in this case, cos t + 2 cos 4t) in time domain. Then the signal is
sampled (four times in our example).
The second time around, the second line in the data space
corresponding to the 180° phase shift will give us:
1. (cos of column 1 frequency) with (magnitude = 1), for example, 1 cos
t
2. (cos of column 4 frequency) with (magnitude = 0), for example, 0 cos
4t, which is 0
Thus, we get a different signal in the time domain for the second line in
the data space. Then this signal is sampled. Remember that no direct
relationship exists between a point in the data space and the same
point on the image.
The Fourier transform of the data space contains the four frequencies
corresponding to the four samples taken during signal readout using the
Gx gradient for frequency encoding. The magnitude (amplitude) of the
frequencies correlates with the brightness on the image. In the x
direction, a 1:1 relationship exists between frequency and position on
the image. The amplitude at a certain frequency corresponds to the
brightness at the corresponding pixel position. In the y direction, a 1:1
relationship exists between the position y and the phase increment Δφ
(which is related to the gradient strength Gy).
To create the image, we perform a second Fourier transform on the
data space. This step is just an additional mathematical step.
Question: How many calculations are necessary to solve the set of
equations derived from k-space to create the image (i.e., the number
of calculations to solve the DFT)?
Answer: In our example of two rows of kspace with four samples in
each row, we had two equations per sample and four samples. So:
2 × 4 = number of calculations
In general, with an N × N matrix, the number of calculations = N × N =
N2.
Example
In a 256 × 256 matrix:
2562 = 216 calculations needed for DFT
Fast Fourier Transform (FFT)
Fast Fourier transform (FFT) is a signal processing transformation,
similar to Fourier transform that solves a DFT in a faster way. The
number of calculations for FFT is
Number of calculations = (N)(log2N)
Example
For a 256 × 256 matrix:
256 × log2 256 = (256)(8)
Because 8 is 1/32 of 256, we have cut down the number of calculations
by a factor of 32. For (log2N) to be a whole number, the number of
frequency-encoding steps has to be a power of 2. This is why
frequency-encoding steps are always a power of 2 (i.e., 2N such as 2, 4,
8, 16, 32, 64, 128, 256, and 512), usually 128, 256, or 512, whenever
FFT is used.

Key Points
We have introduced the often intimidating topic of k-
space. k-Space can initially be thought of as the
“data space” (which can be thought of as an “analog”
k-space), with each line in it representing a sampled
version of the received signal (the echo). In the data
space, the coordinates are in time. (Horizontal scale
is on the order of the sampling interval and the
vertical scale is on the order of TR.) The Fourier
transform of k-space is the desired image.
There is, however, one more step that comes after
obtaining the data space and before construction of
the true k-space, having to do with the concept of
“spatial frequencies,” which we shall discuss in
Chapter 16.

Questions
13-1 T/F The number of rows in the data space equals the number of
phaseencoding steps.

13-2 T/F Each row of data space corresponds to one frequency-

encoding gradient strength.

13-3 T/F The center of data space contains maximum signal.

13-4 T/F Each row of data space contains one of the received signals
(echoes).
13-5 T/F The axes of the data space are in the time domain.

13-6 T/F There is a direct relationship between the center of the k-

space and the center of the image.

13-7 T/F The right half of the data (or k) space is the mirror image of
the left half.

13-8 T/F The center of the data space is directly related to the
center of the image.

1The one to oneness in the y direction is related to the phaseencoding

gradient strength—and not just the phase. This is because the rows in
data space are differentiated by different phase-encoding gradient
strengths Gy.
2½ NEX is a misnomer because we are really halving the number of

phase-encoding steps, not the NEX.

14
Pulse Sequence Diagram

“A pulse sequence diagram is to an MR scientist as sheet music is to a

musician.”

Introduction
A pulse sequence diagram (PSD) illustrates the sequence of events that
occur during magnetic resonance imaging (MRI). It is a timing diagram
showing the radio frequency (RF) pulses, gradients, and echoes. Having
a good knowledge of the PSD will help the reader follow complicated
pulse sequences with more ease and understand the interplay among
various scan parameters.

PSD of an SE Sequence
Having been exposed to the concept of gradients, we are now able to
illustrate a complete PSD for, say, a spin-echo (SE) sequence (Fig. 14-1).
Everything in the figure looks like what we have discussed before
except for a few adjustments in
 Figure 14-1. A spin-echo PSD.

1. Slice-select gradient (Gz)

2. Frequency-encoding gradient (Gx)
1a. Once the slice-select gradient (Gz) is applied, then a gradient in the
negative direction is introduced in order to refocus the spins (Fig. 14-
2). Basically, every time we apply a gradient, we dephase the spins. In
the case of the Gz gradient, we dephase the spins in order to select a
slice. But after the slice is selected, we need to reverse the effect. The
purpose of the refocusing gradient is to rephase the spins in the slice-
select direction. (Alternatively, we can defocus the spins prior to slice
selection so that

they come back into phase with the second gradient pulse.)
 Figure 14-2. The slice-select gradient Gz is followed by a negative
lobe to refocus the spins.

1b. When the 180° pulse is applied, a sliceselect gradient may or may
not be applied. This is optional and depends on whether a single slice
or multiple slices are being acquired. But before and after the 180°
pulse, we apply a socalled crusher gradient so that the 180° pulse has
a tri-lobed shape (Fig. 14-3). This is just a triviality. When the 180°
pulse is applied, it may have elements that are not exactly 180° leading
to unwanted additional transverse magnetization. This may result in
the echo not focusing at time TE, as we would expect. So these
“crusher” gradients are applied to offset that error. (The first lobe is
used to balance the third; the second is slice selective; the third
destroys the free induction decay [FID] from the unwanted transverse
magnetization.)
2. There is an important adjustment in the readout gradient (Gx). If we
just apply a gradient while we're reading out the echo, we end up
dephasing everything (Fig. 14-4). By the time we get to the middle of
the signal, the signal intensity will be decreased because of the
dephasing caused by the gradient, and by the time we get to the end of
the signal, there will be maximum dephasing and so much signal loss
that there may not be any signal to read!

Figure 14-3. Crusher gradients are applied at each side of the

slice-selective gradient (applied during the 180° pulse) to achieve
more accurate refocusing at time TE.

So we apply a gradient in the negative direction that has an area equal

to ½ of that of the readout gradient (Fig. 14-5). The length of the
readout gradient is the sampling time (Ts). For stationary spins,
application of a gradient will make the spins go faster and faster and,
as they go faster, they'll get out of phase. With the negative gradient,
the stationary spins will have a maximum phase difference at the end
of the negative gradient. As the gradient is reversed and a gradient in
the positive direction is applied, the spins will rephase once again. This
occurs right at the midpoint of the readout, that is, at time TE.
Subsequently, they'll go out of phase again. We can see that at time TE
everything is refocused (Fig. 14-5).
If we didn't have the negative gradient lobe, the spins would begin to
dephase when the gradient is turned on, and at time TE there would be
an undesirable phase difference. Phase difference means a smaller
signal (Fig. 14-6).
Sometimes, we'll see the notation for the extra Gx gradient illustrated
differently (Fig. 14-7). It will be shown as a positive gradient rather
than a negative gradient (as we just discussed). You

might wonder that if the pre-readout Gx gradient is positive, aren't we

going to create more phase difference with an additional positive lobe?

Figure 14-4. If only a constant gradient Gx is applied during

readout, we end up dephasing all the spins.
Figure 14-5. Prior to the application of the readout gradient, a
negative gradient is applied, resulting in a tri-lobed gradient. The
negative lobe causes the spins to get out of phase. Then the spins
get back in phase in the center of the echo.
 Figure 14-6. In the absence of a tri-lobed Gx, the spins will
accumulate a phase in the center of the echo, thus yielding less
signal. The dashed line is our desired signal waveform.

The answer lies in the fact that in Figure 14-7 the positive pre-readout
Gx gradient comes before the 180° refocusing pulse. After the Gx pre-
readout gradient, we'll have a positive phase difference. This phase
difference will stay constant until the 180° pulse is applied (Fig. 14-8).
After the 180° RF pulse is applied, the phase difference will be
reversed. Then it will remain constant until the Gx gradient is applied.
Then, the spins will begin going back in phase, reaching a zero phase
difference at time TE, and going out of phase afterwards. So, we get
the same thing with both the positive and negative pre-readout Gx
gradients, depending on where in the pulse sequence the gradient is
applied.
 Figure 14-7. The first lobe of Gx can be applied either after the
180° as a negative lobe (as in previous figures) or as a positive lobe
prior to the 180° pulse.

Acquisition Time
In previous chapters, we talked about multislice imaging and we said
that there was a lot of “dead” time between the end of the echo and
the next 90° RF pulse (Fig. 14-9). We can use this “dead” time to our
advantage to process other slices.
 Figure 14-8. This diagram demonstrates how the spins get back in
phase at the center of the echo when an additional refocusing
positive gradient is applied prior to the 180° pulse.

Question 1: If we manage to fit in all the slices we want within the

dead time, what will the acquisition time for the study be?
Answer: It takes TR seconds to fill one line in the data space. Thus, the
acquisition time, which is the time it takes to fill the entire data space,
is TR times the number of lines in the data space.
Question 2: How many lines (rows) do we have in the data (or k) space?
Answer: The number of lines in k-space equals the number of phase-
encoding steps Np or Ny.
Question 3: Does it help to repeat the sequence all over again?
Answer: We can repeat the whole sequence over again (or repeat each
phase-encoding step over again) to average out the noise and increase
 the signal-to-noise ratio (SNR).

Figure 14-9. There always is a “dead” time between the

completion of readout and the next 90° pulse. This dead time can
be taken advantage of to acquire other slices.

Each cycle is called an excitation. The term NEX stands for the Number
of Excitations (also known as NSA—Number of Signal Averages). So the
acquisition time depends on
1. TR (the time to do one line of the data space)
2. Ny (the number of phase-encoding steps)
3. NEX (the number of times we repeat the whole sequence)
Acquisition time = (TR) (Ny) (NEX)
This formula is for a conventional SE sequence. Notice that the number
of slices doesn't even enter the equation. This is somewhat
counterintuitive if one is used to the principles of scan time in CT
because, in CT, the more slices we obtain, the longer the sequence will
be. This is not necessarily true in MR because of the fact that we can do
multiple slices within the time of

one TR. Obviously, though, if we decrease the TR, we decrease the

number of slices we're able to obtain. So the number of slices is
indirectly determined by the TR parameter.
Let's say that we want to do a T1-weighted study with a fairly wide
coverage. Increasing the TR to allow more slices per TR might be
counterproductive because T1 weighting will be reduced. Thus, there is
a trade-off between increasing coverage and achieving more T1
weighting.
Example 1
TR = 1000 msec, Ny = 256, NEX = 1
Acquisition time = (TR)(Ny(NEX) = (1000 msec)(256)(1) = sec ≅ 4.27 min
Let's say that we fit 10 slices in the TR. Then we could obtain the 10
slices in 4.27 min.
Example 2
If, on the other hand, we were only to obtain one slice per TR at a
time, we would have to repeat the sequence 10 times to obtain 10
slices, and then the scan time would instead be
(10)(1000 msec)(256)(1 NEX) ≅ 10 × 4.27 min = 42.7 min
which is, obviously, impractical.

Key Points
We have discussed the topic of pulse sequence
diagram (PSD) and illustrated one example for SE
imaging. In the chapters to come, we will see
examples of more complicated PSDs. Of course, the
PSD does not tell us all the parameters used in MR
imaging, such as the field of view (discussed in the
next chapter), but it offers an algorithm or
prescription for performing the study.

Questions
14-1 The acquisition time depends on which of the following? (one
or more)
(a) TR
(b) TE
(c) Nx
(d) Ny
(e) NEX

14-2
(a) Calculate the acquisition time for a multiacquisition SE
sequence with TR = 2000, NEX = 2, Ny = 128.
(b) Repeat (a) for single slice acquisition of 10 slices. Is this
practical?

14-3 The number of rows in the k-space equals

(a) Nx
(b) Ny
(c) NEX
(d) Nz
(e) TR
 15
Field of View

Introduction
A pulse sequence diagram (PSD) provides a timing algorithm for the
sequence of events that is carried out during an MR study. However, the
operator must specify the dimensions of the desired part of the body
under study. This is the subject of this chapter, namely, the field of
view (FOV). As we shall see shortly, there is a limitation as to how small
we can make the FOV, depending on the maximum strength of the
gradients and the bandwidth of the received signals.

Field of View (FOV)

We're going to discuss the relationship between the following entities:
1. FOV
2. Bandwidth
3. Gradients
Understanding these concepts is important because these features have
definite clinical applications.
Let's take an image with its x and y axes (Fig. 15-1). There is an FOV
along the x-axis. Normally, we apply a gradient that increases as we
move in the x direction (Gx). This means that we create magnetic
inhomogeneities along the x-axis in a linear fashion. Consequently,
1. At the center point of the FOV, the magnetic field will be B0.
2. On the right side of the FOV, the magnetic field will be greater than
B0.
3. On the left side of the FOV, the magnetic field will be less than B0.

The magnetic field along the x-axis is Bx. The value of Bx is given by the
linear equation: Bx = (Gx)x This equation shows that the value of the
magnetic field at any point along the gradient (Gx) is the slope of the
gradient (Gx) times the distance x along the x-axis (Fig. 15-2). Let's
multiply both sides of the equation by the gyromagnetic ratio γ: γ · Bx =
γ(Gx)x

Recall that (γBx) is the Larmor equation, which relates the magnetic
field strength to the frequency: frequencyx = γBx This equation states
that the frequency of oscillation at any point along the x-axis is
proportional to the magnetic field strength at that point, that is, fx =
γBx or fx = γ(Gx)x In other words, the frequency at any point along the
x-axis is proportional to the slope of the gradient (Gx) multiplied by the
position along the x-axis.
Figure 15-1. An image with axes x and y. The frequencyencode
gradient Gx causes the center of the field of view (FOV) to have
magnetic field strength B0 and the right and left ends to have
strengths greater and less than B0, respectively.
 Figure 15-2. The gradient Gx describes a linear equation Bx = Gx ·
x. Therefore, at x = 0, no net magnetic field is added to the
system, whereas at a positive value of x, a positive value of
magnetic field is added to the main field.

Let's see what happens at each end of the FOV (Fig. 15-3). At the right-
side end of the FOV (i.e., at x = FOV/2), the frequency is maximum
(call it fmax) because the Gx gradient, and therefore the magnetic field
strength, is maximum.
The formula for frequency is fx = γ(Gx)x Now, for fmax, the distance
along the x-axis is ½ FOV. So, fmax = γ(Gx)FOV/2 Remember this is after
we subtract the center frequency. Therefore, these measurements are
centered around the zero frequency. At the opposite end of the
gradient, we have −fmax: −fmax = −γ(Gx)FOV/2 What is the range of
frequencies? The frequency range is from −fmax to +fmax, that is,
Frequency range = −fmax → +fmax = ±fmax = 2fmax Another term for the
range of frequencies is bandwidth (BW). Thus, BW = ±fmax = 2fmax If we
take the frequency at the right-most side of the image and at the left-
most side of the image, we get the range of frequencies, or the
bandwidth. We already know that, for maximum frequency, fmax =
γ(Gx)FOV/2 Because BW = 2fmax we can conclude that BW = γ · Gx · FOV
We, therefore, see a dependent relationship among the field of view,
bandwidth, and gradient strength. Let's now solve the equation for the
FOV in the x direction:

This equation shows that the FOV is directly proportional to the

bandwidth and that the FOV is inversely proportional to the gradient.
Hence, if one wishes to decrease the FOV, one can use either (i) a
stronger gradient or (ii) a lower bandwidth.
 Figure 15-3. At each end of the FOV, the frequency fx (which is
proportional to gradient strength Gx) is maximal. This relationship
is given by fx = γ · Gx · x, where x = FOV/2 for fmax.

Top ↓ FOV:
1. ↑ Gradient
2. ↓ Bandwidth
There are limits as to how strong you can make the gradient and there
are also limits as to how low you can make the bandwidth.
 Question: What is the minimum FOV possible?
Answer: It is the minimum bandwidth divided by the maximum
gradient:

Gmax and BWmin are specific for each machine. For example, for a GE
Echospeed Plus 1.5-T scanner, maximum gradient strength = 23 mT/m
minimum bandwidth = ±4 kHz = 8 kHz Thus, the minimum FOV is
approximately: 8 kHz/(42.6 MHz/T × 23 mT/m) ≅ 0.8 cm Conversely, to
increase the FOV, we can
1. Increase the BW or
2. Decrease the gradient
To ↑ FOV:
1. ↓ Gradient
2. ↑ Bandwidth
(Remember that by increasing bandwidth, we get decreased signal-to-
noise ratio.)

Key Points
We have discussed the interesting relationship
among the FOV, BW, and gradients: FOV = BW/(γ ·
G) As we saw, there is a limit as to how small one
can make the FOV, depending on the minimum
allowable BW and the maximum possible gradient
strength:

Selecting a smaller FOV may cause an aliasing (or

 wraparound) artifact, depending on the size of the
structure being imaged. More on this is in Chapter
18.

Questions
15-1 If the minimum FOV = 30 cm for a frequency-encoding gradient
Gx = 5 mT/m, what would the minimum FOV be for a stronger Gx =
10 mT/m? (i.e., does a stronger Gx reduce or increase the minimum
FOV?)

15-2 The min FOV is inversely proportional to

(a) BW
(b) gradient strength
(c) TR
(d) TE

15-3 If the amplitude of a phase-encoding gradient Gy is 0.1 mT/m

and its duration is 2 msec, what is the phase shift of the transverse
magnetization from a tissue that is 20 mm = 2 cm from the center of
the FOV? Hint: δø = 360° × γ × Gy × duration × position, where γ =
42.6 MHz/T.

15-4 The minimum FOV can be reduced by

(a) increasing the gradient strength
(b) decreasing the bandwidth
(c) increasing the sampling interval
(d) all of the above
(e) only (a) and (b)
15-5 T/F Reducing the FOV minimizes wraparound artifacts
(aliasing).

15-6 What is the minimum FOV for a maximum sampling interval of

δTs = 10 µsec (i.e., without encountering aliasing) and a maximum
frequency gradient of 10 mT/m? Hint: BW = 1/δTs
(a) 47 cm
(b) 23.5 cm
(c) 47 mm
(d) 23.5 mm

15-7 The min FOV is directly proportional to

(a) BW
(b) gradient strength
(c) TR
(d) TE
16
k-Space: The Final Frontier!

Introduction
This chapter will summarize some concepts that we've already
discussed and clarify some of the fine points of k-space. Remember that
we have, up to this point, referred to the data space as an “analog” k-
space, and we said that the Fourier transform of the data space is the
image (Fig. 16-1).
This is, in fact, correct, but there is a problem with this concept: the
matrix in the data space is very asymmetric. In the frequency-encoding
direction, the interval between two samples (i.e., the sampling interval
ΔTs) is on the order of microseconds, so that the total time to take all
the samples (i.e., the sampling time Ts) is on the order of milliseconds.
The time intervals in the phaseencoding direction, however, are each
on the order of one TR (i.e., on the order of seconds). The total time to
obtain all the data in the phaseencoding direction is the scan time for
one acquisition (on the order of minutes).

Figure 16-1. The data space (with axes ΔTs and TR) is in the time
 domain. k-Space (with axes Δky) is in the spatial frequency domain
and is derived from the data space. The Fourier transform of k-
space is the image (with axes x and y), which is in the frequency
domain.

Thus, in the data space we have a matrix whose x-axis is on the order
of milliseconds and whose y-axis is on the order of minutes. This would
give us a very asymmetric matrix. The true kspace is the same matrix
as the data space, but with a different scale. Recall that

In the last chapter, we talked about the field of view (FOV). We derived
Equation 16-1, showing the relationship among the FOV, the bandwidth
(BW), and the gradient strength. According to this formula, the FOV is
equal to the BW divided by the product of γ and G. We also know from
a

previous chapter that the bandwidth is inversely related to the

sampling internal (ΔTs), that is, BW = 1/ΔTs
From the following two formulas:

and
we can derive a new formula for the FOV:

To calculate this in terms of distance and time, we need to invert both

sides of the previous formula to obtain:

Now consider the FOV in the x and y directions. If we consider the FOV
in the x direction, the formula tells us that we need to take the
gradient strength and the sampling interval in the x direction.

The term (γ · GX · ΔtX) is denoted ΔkX. If we now look at Figure 16-1,

we see that (ΔkX) is the unit interval in k-space in the x direction. Thus,
Δkx = γGxΔtx Let's discuss the units of measurement in the above: γ =
gyromagnetic ratio = MHz/T Gx = gradient strength = mT/m Δtx =
sampling interval = msec so Δkx = (MHz/T)(mT/m)(msec) = (cycles/sec ·
T) × (T/m) × sec = cycles/m Thus, ΔkX has units of cycles/m or
cycles/cm.

Figure 16-2. The FT of k-space is the image. There is a

relationship between the axes in k-space and the image: ΔkX =
1/FOVX, Δx = 1/kX, where FOVX = NX · Δx and kX = NX • ΔkX.

The main thing to remember is that ΔkX is 1/FOV in the x direction.

Δkx = 1/FOVx The above formula tells us that the interval in k-space is
equal to 1/FOV in the x direction (where the FOV, according to
Equation 16-1, depends on the bandwidth and gradient strength in the x
direction). This fact is shown diagrammatically in Figure 16-2.
From this, we can see a direct relationship between k-space and the
image in that the interval in k-space has an inverse relationship to the
FOV of the image. For example, if the FOV of the image is 10 cm, then
Δkx = 1/FOV = 1/10 cm = 1/0.1 m = 0.1 cm-1(cycles/cm) = 10m-
1(cycles/m)

Therefore, with an FOV of 10 cm in the image, the pixel size of k-space

is 0.1 cm−1 or 10 m−1. (Remember that the unit of the axes in k-space is
1/distance or cycles/distance.) The inverse of this is also true: Δx =
pixel size in the image kX = sum of the pixels in k-space
In summary: Δx = 1/kx, Δkx = 1/FOVx Δkx = γ · Gx · Δtx, kx = γ · Gx · tx
where Δx is the pixel size in the x direction. In essence, this is how we
figure out the pixel size in the image. Δx = pixel (x direction) = FOVX/NX
This formula tells us that the pixel size in the x direction is equal to the
FOV divided by the number of pixels in the x direction.
Example
1. Calculate the pixel size for the following situation: FOVx = 128 mm
Nx = number of sampling points in the x direction = 128 then the pixel
size in the x direction = Δx = 128 mm/128 = 1.0 mm
2. Calculate the dimensions in k-space for the above example: Δkx =
pixel size in k-space = 1/FOVx Δkx = 1/128 mm = 1/12.8 cm ≅ 0.08 cm-1
= 8 m-1 kx = 1/Δx = 1/1 mm = 1 mm-1 = 10 cm-1 = 1000 m-1 = 128Δkx The
unit in k-space is ΔkX = cycles/distance kX = cycles/distance

The measurement cycles/distance is called the spatial

frequency.

Therefore,

k-Space is in the spatial frequency domain.

This is different from the other frequency that we have discussed thus
far. Before, we discussed the Fourier transform of a signal that varied
with time, with the transform in the frequency domain. Spatial
frequency is a different kind of frequency.

The usual frequency = cycles/time

Spatial frequency = cycles/distance

Therefore, when k-space is said to be in the “frequency domain,” we

are referring to the “spatial frequency domain,” which, as we just saw,
is a mathematical manipulation of the data space (which is in the time
domain).
Remember that the units of the axes in data space are time:
milliseconds and minutes. In k-space, we have converted these axes
into “spatial frequencies” that have units of cycles/distance, measured
in cm-1 (cycles/cm) and m-1 (cycles/m). If we Fourier transform k-
space, we get the desired image.
Mathematically, we could go straight from the data space, via a Fourier
transform, directly to the image. We are simply renaming the variables
(i.e., γ GX ΔtX = ΔkX). This renaming is known as an algebraic
manipulation, but to go through this intermediate mathematical step in
k-space allows us to work with a space that is more symmetric; now,
the distance in the x direction of k-space and the distance in the y
direction of k-space are roughly similar (as opposed to the difference
between milliseconds and minutes in the x direction and y direction in
the data space, respectively).
This same concept in the y direction is somewhat harder to understand,
but the same principles hold, namely,

The interval kY in k-space is inversely proportional to the FOV in the y

direction.
The distance in k-space in the y direction is inversely proportional to
the pixel size of the image in the y direction.
One more mathematical principle concerns the relationship between
phase and frequency: θ = ∫ωdt
In other words, the phase θ is the integral of the frequency with
respect to time, where (angular frequency) is given by the Larmor
equation1: ω = γ B = γ · G · x
In other words, the frequency ω is proportional to the magnetic field
strength which is, in turn, proportional to the gradient strength
multiplied by the distance. Thus, θy = ωyty = γ · By · ty = γ · Gy · y · ty
or θy = (γGyty) · y Remember that: Δky = γGyΔty and (ky = Δky · Ny) and
(ty = Δty · Ny) so ky = Δky · Ny = γGyΔty · Ny Therefore, ky = γ · Gy · ty so
θy = (ky) · (y) We thus have a very simple relationship between phase
and position along the y direction: θy = (ky) · (y) = (γGy) (ty) (y) In the y
direction, the gradient at y depends on the position of y. In contrast,
we always apply the same gradient in the x direction regardless of what
position the x direction is going (i.e., Gx is independent of x). However,
in the y direction, we apply different gradients at different points along
the y-axis (Gy varies with y: it is 0 at y = 0 and gets progressively larger
with increasing y).

Key Points
The true k-space is a mathematically manipulated
variant of the data space, with axes referred to as
“spatial frequencies.” Therefore, k-space is in a
“spatial” frequency domain. Spatial frequencies kX
and kY are inversely proportional to distance (with
units of cycles/cm). The Fourier transform of k-space
is the desired image.
 The spatial frequencies kX and kY are expressed as:
kx = γ · Gx · tx ky = γ · Gy · ty with units in cycles/cm.

Questions
16-1 T/F Spatial frequencies have units 1/distance (cycles/cm).

16-2 T/F
(a) The axes in k-space are designated kx and ky.
(b) The axes in k-space are in the frequency domain (with units
1/time or cycle/sec).

16-3 T/F The Fourier transform of the k-space produces the desired
image.

16-4 ΔkX is equal to

(a) 1/FOVx
(b) γ GXΔtX
(c) kX/NX
(d) all of the above
(e) only (a) and (b)

16-5 T/F
(a) The center of k-space contributes to maximum image
contrast.
(b) The periphery of k-space contributes to image details.

16-6 T/F k-Space can be thought of as a digital (in the spatial

frequency domain) version of the data space (which is in the time
domain).
1You may have wondered why we have been using ω and f somewhat

interchangeably in our equations, although these are two separate

entities. This practice is all right as long as you keep the right units for
the gyromagnetic ratio γ (i.e., MHz/T when dealing with f and 2π
MHz/T when dealing with ω).
17
Scan Parameters and Image Optimization

Introduction
In this chapter, we will discuss all the important parameters in MR
imaging that the operator can control and adjust. We will then see how
these changes influence the image quality. Every radiologist is
comfortable with a particular set of techniques; therefore, “custom-
made” techniques can be achieved only if the radiologist is aware of
the parameters and trade-offs involved.

Primary and Secondary Parameters

Primary parameters are those that are set directly:
 From the primary parameters above, we can get the secondary
parameters (which are also used to describe the image):
1. S/N ratio (SNR)
2. Scan time
3. Coverage
4. Resolution
5. Image contrast
Unfortunately, optimization of these parameters may involve some
trade-offs. To gain some advantage with one parameter, we might have
to sacrifice another parameter. Let's start out with the concept of
signal-to-noise ratio (SNR).
 Signal-to-Noise Ratio. What we want is signal. What we don't want is
noise. Although we can't completely eliminate noise, there are ways to
maximize the SNR. SNR is given by

which makes sense because Ny × NEX × Ts is the total time the machine
is “listening” to the spin echoes.
Since Ts = Nx · ΔTs and ΔTs = 1/BW, then Ts = Nx/BW.

Therefore, SNR depends on

1. Voxel volume = Δx · Δy · Δz
2. Number of excitations (NEX)
3. Number of phase-encoding steps (Ny)
4. Number of frequency-encoding steps (Nx)
5. Full bandwidth (BW)
Let's go through each parameter and see how SNR is affected.

Voxel Volume
If we increase the voxel size, we increase the number of proton spins in
the voxel and, therefore, increase the signal coming out of the voxel
(Fig. 17-1). The voxel volume is given by Voxel volume = Δx · Δy · Δz
where Δx = pixel size in the x direction, Δy = pixel size in the y
direction, and Δz = slice thickness.
NEX (Number of Excitations or Acquisitions). NEX stands for the
number of times the scan is repeated. Let's say we have two signals (S1
and S2), corresponding to the same slice (with the same Gy). There is
constant noise (N) associated with each signal (N1 = N2 = N). If we add
up the signals (assuming S1 = S2 = S), we get S1 + S2 = 2S However, if we
add up the noise, we get

This formula does not make sense at first glance. Why do we get N
and not 2N? The answer has to do with a somewhat complicated
statistical concept and the so-called random Brownian motion theory,
which deals with the spectral density of the noise.

Figure 17-1. A voxel is a three-dimensional volume element with

dimensions Δx, Δy, and Δz. The more spins in a voxel, the more
signal. Therefore, increasing voxel size increases SNR.

In a simplistic approach, think of the noise as the variance (σ2) of a

Gaussian distribution (σ = standard deviation). Then, for the sum of
the two noise distributions, the variance is additive and given by σ12 +
σ22 = σ2 + σ2 = 2σ2 from which the standard deviation is calculated to
be

This is where the factor comes from. However, you don't need to
know the underlying math—you just need to understand the concept. In
summary:

The resulting signal will be twice the original signal. The resulting
noise, however, will be less—it will be the square root of 2 multiplied by
the noise N.
In other words, if we increase the number of acquisitions by a factor of
2, the signal doubles and noise increases by , for a net 2/ = ;
thus, SNR increases by a factor of .
Therefore, ↑ NEX by a factor of 2 → ↑ SNR by a factor of .
Think of the NEX as an averaging operation that causes “smoothing”
and improvement in the image quality by increasing the signal to a
greater degree (e.g., factor 2) relative to the increase in the noise
(e.g., factor ). As another example, increasing NEX by a factor of 4
results in an increase of signal by 4 and an increase of noise by or 2.
 Thus, SNR increases by 4/2 or twofold.
Ny (Number of Phase-Encoding Steps). The same concept holds for Ny.
That is, similar to NEX, there is a 41% ( ) increase in SNR when Ny is
doubled. As with NEX, when the number of phase-encode steps
doubles, signal doubles and noise increases (randomly) by (for a net
increase in SNR).
Bandwidth. An inverse relationship exists between BW and SNR. If we
go to a wider bandwidth, we include more noise, and the SNR
decreases. If we decrease the bandwidth, we

allow less noise to come through, and the SNR increases.

↓ BW → ↑ SNR To be exact, decreasing the BW by a factor of 2 causes
the SNR to improve by a factor of .
In general, decreased bandwidth causes the following:
1. Increased SNR
2. Increased chemical shift artifact (more on this later)
3. Longer minimum TE (which means less signal due to more T2 decay).
Remember that Bandwidth = 1/ΔTs = Nx/Ts
Therefore, a longer sampling time (Ts), which is necessary for a
decreased bandwidth, results in a longer minimum TE. With a long
TE, increased T2 dephasing results in decreased signal. However, the
contribution from reduced noise due to a lower bandwidth outweighs
the deleterious effect of reduced signal due to greater T2 decay from
increased TE.
4. Decreased number of slices. This decrease is caused by the longer TE.
Remember, number of slices = TR/(TE + Ts/2 + To) where Ts is the
total sampling (readout) time and To is the “overhead” time. A
narrower bandwidth is usually used on the second echo of a T2-
weighted dual-echo image because, with the second echo, we have a
longer TE and we are able to afford the longer sampling time. On the
first echo, however, we can't afford to use a narrower bandwidth
because we can't afford to lengthen the TE. However, we probably
 don't need the smaller bandwidth anyway because we already have
enough SNR on the proton density-weighted first echo of a long T2,
double-echo image. A typical Ts for a 1.5-T scanner is 8 msec,
resulting in a BW (for a 256 matrix) of BW = Nx/Ts = 256/8 = 32 kHz =
± 16 kHz = 125 Hz/pixel
Note that BW can be described as “full bandwidth” (32 kHz in example
above), ± the Nyquist frequency (which is ± 16 kHz above and defines
the FOV) or “bandwidth per pixel” (which is unambiguous if you forget
the ±).
A typical “variable bandwidth” option includes
1. A wide bandwidth (±16 kHz) on the first echo, and
2. A narrow bandwidth (±4 kHz) on the second echo, thus increasing SNR
and counteracting T2 decay effects.
Question: How does the gradient affect the BW?
Answer: Recall from Chapter 15 that the field of view (FOV) is given by
FOV = BW/γGx or Gx = BW/γFOV
For a given FOV, increasing the gradient Gx causes increased BW and,
therefore, decreased SNR.

SNR in 3D Imaging
In 3D imaging, we have the same factors contributing to SNR, plus an
additional phaseencoding step in the z direction (Nz):

From this equation, you can see why SNR in 3D imaging is higher than
that in 2D imaging. Specifically,

Another way to look at SNR is to say that SNR depends on only two
factors:
1. Voxel size
2. Total sampling time
Sampling time (Ts) is the time that we sample the signal. Therefore, it
makes sense that the more time we spend sampling the signal, the
higher the SNR will be. Let's look again at the formula for SNR (in 2D
imaging):

Recall that Ts = Nx/BW or 1/BW = Ts/Nx

so

We know that Ny is the number of phase-encoding steps, which is the

number of times we sample the echo corresponding to a particular
phaseencoding gradient Gy, and that NEX is the number of times we
repeat each phase-encoding step. In essence, the factor T = Ts · Ny ·
NEX is the total sampling time of all the echoes received for a
particular slice. Thus,

In summary, SNR can be increased by doing the following:

1. Increasing TR
2. Decreasing TE
3. Using a lower BW
4. Using volume (i.e., 3D) imaging
5. Increasing NEX
6. Increasing Ny
7. Increasing Nx
8. Increasing the voxel size
 Resolution. Spatial resolution (or pixel size) is the minimum distance
that we can distinguish between two points on an image. It is
determined by Pixel size = FOV/number of pixels ↑ Ny → better
resolution If we increase the number of phase-encoding steps, what
happens to SNR? Obviously, better resolution usually means poorer SNR.
However, if we look at Equation 17-2, it appears that by increasing Ny,
the SNR should increase! What's the catch? The catch is, we are keeping
the FOV constant while increasing Ny. Take, for example, Pixel size
along y-axis = Δy = FOVy/Ny By increasing Ny, we are making the pixel
size smaller. Now, recall that Voxel volume = Δx · Δy · Δz = FOVx · FOVy
· Δz/(Nx · Ny) Incorporating this information into Equation 17-2 gives us
another way of expressing SNR:

This formula allows us to better separate the factors affecting SNR.

From this, we can conclude the following:
1. If we keep FOV constant and increase Ny, we will decrease SNR.
↑ Ny, FOV constant → ↓ SNR
2. If we increase Ny and increase FOV, thus keeping pixel size constant,
then we will increase the SNR. Yet the resolution doesn't change.
What is the trade-off here? The answer is the acquisition time, which
is proportional to Ny.
↑ FOV, pixels fixed → ↑ SNR, ↑ time
3. If we increase slice thickness Δz, we get not only more SNR, but also
more partial volume artifact.
4. If we increase NEX, we get more SNR at the expense of longer
acquisition time. For 3D imaging, Equation 17-4 is modified to SNR
 (3D) = (FOVx)(FOVy)(FOVz)

Basically, if we want better spatial resolution in a given acquisition

time, we have to sacrifice SNR. Let's consider a few examples.
1. What happens if we increase the number of pixels with the FOV
constant?
a. Increase resolution.
b. Decrease SNR (refer to Equation 17-4). Therefore, as we decrease
the pixel size, we increase the resolution and decrease the SNR.
c. Increase scan time (number of pixels increases in phase-encode
direction).
2. What happens if we decrease the FOV and keep the number of pixels
constant?
a. Increase the resolution.
b. Decrease SNR.
c. Potentially increase aliasing artifact.

3. How do we determine the pixel size (resolution)?

It is determined by dividing the FOV by the number of encoding steps.

Example
For FOV = 250 mm and a 256 × 256 matrix Nx = Ny = 256 Pixel size
(x) = FOVx/Nx = 250/256 ≅ 1 mm in x direction. Pixel size (y) =
FOVy/Ny = 250/256 ≅ 1 mm in y direction. In the x direction, there
are two ways of increasing resolution (for a given FOV):
1. Increase Nx by reducing the sampling interval ΔTs (i.e., by increasing
the BW) and keeping the sampling time Ts fixed (recall that Ts = Nx ·
ΔTs). The advantage here is no increase in TE; the trade-off is a
reduction in SNR (due to increased BW).
2. Increase Nx by lengthening Ts and keeping ΔTs (and thus BW) fixed.
Here, the SNR does not change, but the trade-off is an increased TE
(due to a longer Ts) and less T1 weighting (this is only a concern in
short echo delay time imaging).

Acquisition Time
The acquisition time or scan time, as we have seen previously, is given
by Scan time = TR · Ny · NEX where Ny is the number of phase-encoding
steps (in the y direction).
For fast spin-echo (FSE) imaging (discussed in detail in Chapter 19), the
above is modified to FSE time = TR · Ny · NEX/ETL where ETL = echo
train length (4, 8, 16, 32).
For 3D imaging, the scan time is given by Time (3D) = TR · Ny · Nz · NEX
where Nz is the number of phase-encoding steps (partitions) in the z
direction. In other words, Time (3D) = Nz · time(2D) Multiplication by
such a large number (e.g., Nz = 32 to 64 or 128) might at first seem to
result in an excessively long scan time for 3D imaging, but the TR used
in 3D gradient-echo imaging is approximately 100 times smaller (order
of 30 msec) compared with the TR used in conventional spin-echo
imaging; we can perform a 3D scan in a reasonable time. Recently, 3D
FSE imaging (discussed in Chapter 19) has also become feasible.

Example
1. Calculate the acquisition time of an SE sequence with TR = 3000
msec, Ny = 256, and NEX = 1.
Solution: Scan time = 3000 × 256 msec = 768 sec = 12.8 min
2. Calculate the acquisition time of an FSE sequence with the above
 parameters and an ETL of 8

3.
a. Calculate the acquisition time of a 3D gradient-echo sequence with
TR = 30 msec, Ny = 256, NEX = 1, and Nz = 60.
Solution: Scan time = 30 × 256 × 1 × 60 msec = 460.8 sec = 7.68 min
b. If TR = 300 in the previous example, then the scan time = 76.8 min
= 1 hr and 16.8 min, which is, obviously, impractical. Hence, 3D
techniques use gradient-echo sequences employing a very short TR.

TR. What happens if we increase or decrease TR?

1. Increasing TR:
a. increases SNR (according to the T1 recovery curve)
b. increases coverage (more slices)
c. decreases T1 weighting
d. increases proton density and T2 weighting
e. increases scan time
2. Decreasing TR:
a. decreases SNR
b. decreases coverage
c. increases T1 weighting

d. decreases proton density and T2 weighting

e. decreases scan time
Sometimes an MR technologist will find that, for a certain TR, the
required coverage cannot be achieved. Therefore, to increase the
coverage, he or she might increase the TR. However, in so doing, T1
weighting is decreased, which may be an undesirable effect.
 Coverage
Coverage is the distance covered by a multislice acquisition. It depends
on the number of slices and on the slice thickness and the interslice gap
(Fig. 17-2). Because number of slices = TR/(TE + Ts/2 + To) then
Coverage = TR/(TE + Ts/2 + To) × (Slice thickness + gap) where Ts is the
sampling time and To is the “overhead” time, as we've discussed in
previous chapters.
Let's summarize:
1. Coverage is increased if we:
a. increase slice thickness
b. increase interslice gap
c. increase TR or decrease the last TE (i.e., increase TR/TE ratio)
d. decrease sampling time Ts (resulting in a lower TE), that is,
increase the bandwidth
2. Coverage is decreased if we:
a. increase TE
b. increase Ts
c. increase ETL in FSE imaging (due to longer final TE)
 Figure 17-2. Coverage is determined by slice thickness Δz and by
the interslice gap. Coverage = number of slices × (Δz + gap).

3. Increasing interslice gap causes

a. increased coverage
b. decreased “cross-talk” artifact
c. increased SNR (due to increasing effective TR by reducing cross-
talk)
d. decreased detection of small lesions (which may lie within the gap)

TE (Time to Echo)
Question: What happens if we increase or decrease TE?
Answer:
1. By increasing TE, we:
(a) increase T2 weighting
(b) increase dephasing and thus decrease SNR (according to the T2
decay curve)
(c) decrease number of possible slices (decrease coverage), because
number of slices ≈ TR/TE
(d) no change in scan time (unless, of course, the coverage is not
adequate and either longer TR or extra acquisitions are required)
2. The reverse is true for decreasing TE:
(a) decrease T2 weighting and increase T1 or proton density weighting
(b) increase SNR (less dephasing). However, if TE is reduced by reducing
Ts (i.e., increasing BW), SNR may be reduced!
(c) increase coverage
(d) no change in scan time
Question: What causes lengthening of the minimum TE?
Answer:
1. TE should be long enough so that the side lobes of the 180° pulse do
not interfere with the side lobes of the FID or the echo (Fig. 17-3).
Remember that we need a Fourier transform of the RF pulse with a
square shape to be able to get ideal contiguous slices. To do this, the
RF must be a sinc wave (sinc t = sin t/t) with as many side lobes as
possible. This, in turn, will lengthen the 90° and 180° RF pulse.
2. If TE is so short that it allows interference between the 180° RF
pulse and the FID, an

FID artifact (or zipper artifact) will appear along the zero frequency
line.

Figure 17-3. To avoid overlapping of the FID and the side lobes of
the 180° pulse, you need to increase TE. This increase is one cause
of lengthening the minimum TE.

Question: How can TE be shortened?

Answer:
 1. One way is to decrease the sampling time Ts. However, this results in
a higher BW and therefore a lower SNR (Equation 17-2).
2. There is a limit as to how short TE could be. The factors limiting
minimum TE include
(a) duration of RF pulse (especially the 180° pulse)
(b) duration of FID
(c) Ts or BW (which influence the SNR)
3. TE can also be shortened by switching to a gradient-echo sequence
because a 180° refocusing pulse is no longer used.
Contrast on a spin-echo technique can be summarized (Table 17-1):

TI (Inversion Time)
As we saw in Chapter 7, inversion recovery sequences employ an
additional 180° pulse before the 90° pulse.

Table 17-1

TR TE

T1W Short Short

PDW Long Short

T2W Long Long

Advantages
1. Can suppress various tissues by selecting the appropriate TI. More
specifically, as we saw in Chapter 7, if TI = 0.693 × T1 (tissue x) then
 tissue x is “nulled” or “suppressed.”
2. STIR (short TI inversion recovery) sequences suppress fat by selecting
TI = 0.693 × T1 (fat) Since at 1.5 Tesla, T1 of fat is approximately 200
msec, then to null fat, we must select TI = 0.693 × 200 ≅ 140 msec
3. FLAIR (fluid-attenuated inversion recovery) sequences suppress fluid
by selecting TI = 0.693 × T1 (fluid)
This sequence is used, for example, in the brain to suppress
cerebrospinal fluid (CSF) to increase the conspicuity of periventricular
hyperintense lesions such as multiple sclerosis plaques. Since at 1.5 T,
T1 of CSF is approximately 3600 msec, then to null CSF, we have to
select TI = 0.693 × 3600 ≅ 2500 msec

Disadvantages
1. Decreased SNR
2. Decreased coverage (by a factor of about 2 due to the presence of
the extra 180° pulse)

Key Points
In this chapter, we discussed the important and
practical factors that influence the quality of MR
imaging. To improve the quality of the images, it is
crucial to have a firm grasp of the parameters that,
directly or indirectly, affect the scan. We introduced
the primary and secondary parameters that are used
to determine MR images (refer to the Introduction in
this chapter). In a nutshell, the name of the game is
“trade-offs.” Often, one cannot gain advantage in one
area without sacrificing another.

Questions
17-1 For a TR = 1500 msec, 2 NEX, and a 128 × 128 matrix, calculate
the scan time for
(a) a single slice
(b) 10 slices (performed one at a time)
(c) 10 slices performed using a multislice (multiplanar)
acquisition

17-2 Calculate the maximum number of achievable slices for a TR =

1000 msec, TE = 80 msec, sampling time Ts = 20 msec, and
“overhead time” To = 10.

17-3 The concept of variable BW: in order to improve SNR, the

lowest possible BW is selected. Suppose that the BW is halved:
(a) How is the SNR affected?
(b) What happens to chemical shift artifacts?
(c) How does this affect the maximum number of slices?

17-4 The acquisition time of a single acquisition gradient-echo

sequence with TR = 30 msec, TE = 10 msec, NEX = 2, Ny = 256 for
acquiring 15 slices is about
(a) 15.36 sec
(b) 153.6 sec
(c) 230.4 sec
(d) 15,360 sec
(e) 230,400 sec

17-5 The SNR in 3D imaging is equal to the SNR in 2D imaging times

the factor:
(a) Nz

(b)
 (c) Ny
(d)

17-6 Increasing TE leads to a decrease in all of the following except

(a) T2W
(b) signal
(c) coverage
(d) SNR

17-7 SNR can be increased by

(a) increasing NEX
(b) decreasing BW
(c) increasing Ny
(d) increasing Nx
(e) increasing voxel volume
(f) increasing TR
(g) decreasing TE
(h) all of the above
(i) only (a)-(e)

17-8 Increasing Ny leads to

(a) better resolution
(b) increased SNR (fixed FOV)
(c) increased SNR (fixed pixels)
(d) increased scan time
(e) all of the above
(f) only (a), (c), (d)
 (g) only (a), (b), (d)

17-9 For a 128 square matrix and an FOV of 25 cm, the pixel size is
about
(a) 0.5 mm
(b) 1 mm
(c) 1.5 mm
(d) 2 mm

17-10 The SNR is proportional to the square root of

(a) BW/Nx · NEX
(b) BW/Ny · NEX
(c) Nx · Ny · NEX/BW
(d) Ny · BW/NEX

17-11 Increasing TR leads to an increase in all the following except

(a) scan time
(b) SNR
(c) T1W
(d) T2W
(e) coverage

17-12 Minimum TE can be reduced by

(a) reducing the duration of the RF pulses
(b) reducing the sampling time Ts
(c) increasing the bandwidth
(d) using a sequence that doesn't use 180° pulses (as in gradient
echo)
(e) all of the above
17-13 The acquisition time in 3D imaging is equal to that in 2D
imaging times the factor:
(a) Nz

(b)
(c) Ny
(d)

17-14 Coverage is increased by increasing all of the following except

(a) slice thickness
(b) interslice gap
(c) TR
(d) BW
(e) TE

17-15 In STIR, T1 should be set to

(a) 1.44 T1 (fat)
(b) (1/√2) T1 (fat)
(c) 2 T1 (fat)
(d) 0.693 T1 (fat)
(e) (1/0.693) T1 (fat)
(f) choices (b) or (d)

17-16 In FLAIR, T1 should be set to

(a) 0.693 T1 (fluid)
(b) (ln 2) T1 (fluid)
(c) (−1ln 0.5) T1 (fluid)
 (d) all of the above

17-17 Match (i) STIR; (ii) FLAIR with

(a) dark fluid
(b) dark fat
 18
Artifacts in MRI

Introduction
MRI, as with any other imaging modality, has its share of artifacts.
It is important to recognize these artifacts and to have the tools to
eliminate or at least minimize them. There are many sources of
artifacts in MRI. These are summarized as follows:
1. Image processing artifact
a. Aliasing
b. Chemical shift
c. Truncation
d. Partial volume
2. Patient-related artifacts
a. Motion artifacts
b. Magic angle
3. Radio frequency (RF) related artifacts
a. Cross-talk
b. Zipper artifacts
c. RF feedthrough
d. RF noise
4. External magnetic field artifacts
a. Magnetic inhomogeneity
5. Magnetic susceptibility artifacts
 a. Diamagnetic, paramagnetic, and ferromagnetic
b. Metal
6. Gradient-related artifacts:
a. Eddy currents
b. Nonlinearity
c. Geometric distortion
7. Errors in the data
8. Flow-related artifacts
9. Dielectric effects
Let's discuss this list in more detail.

Image Processing Artifact

Aliasing (Wraparound). Refer to the discussion on undersampling in
Chapter 12.
Spin-Echo Imaging. Let's say we're studying the abdomen (Fig. 18-1). If
the field of view (FOV) only covers part of the body, we know that we
may get aliasing (wraparound), but what causes the aliasing?
We have a gradient in the x direction (Gx), with a maximum frequency
(fmax) at one end of the FOV, and a minimum frequency (−fmax) at the
other end of the FOV. These are the Nyquist frequencies (discussed in
Chapter 12). Any frequency higher than the maximum frequency
allowed by the gradient cannot be detected correctly.
The gradient doesn't stop at the end of the FOV. The gradient is going
to keep going because we still have magnetic fields outside the space
designated by the FOV. The parts of the body outside the FOV (in this
case, the arms) will be exposed to certain magnetic field gradients.
One arm will receive a magnetic field that will generate a frequency
higher than fmax for the FOV. It may be twice the frequency of fmax—
twice the intended Nyquist frequency. The computer cannot recognize
these frequencies above (fmax) or below (−fmax). They will be
recognized

as a frequency within the bandwidth. The higher frequency will be

recognized as a lower frequency within the accepted bandwidth.

Figure 18-1. For a given FOV and gradient strength, the maximum
frequency fmax corresponds to the edges of the FOV. Any part
outside the FOV will experience a higher frequency. The higher
frequencies outside the FOV may be aliased to a lower frequency
inside the FOV. This will cause a wraparound artifact.

For example, if the higher frequency were 2 kHz higher than (fmax), it
would be recognized as 2 kHz higher than (−fmax), and therefore its
information would be “aliased” to the opposite side of the image—the
side of the FOV that corresponds to the lowest frequencies (Fig. 18-1).
The part of the body and arm on the left side of the patient that is
outside the FOV and is exposed to a higher magnetic field will have
spins oscillating at a frequency higher than (fmax). Thus, it will be
identified as a structure on the right side of the patient—that side of
the image associated with lower frequencies.
 Likewise, the arm and body outside the FOV on the right side of the
patient will experience spins oscillating at frequencies lower than
(−fmax) and will also be incorrectly recognized by the computer. For
example, if the lower frequency were 2 kHz lower than (−fmax), it
would be recognized as 2 kHz lower than (fmax), and its information
would be “aliased” to the opposite side of the image—the side of the
FOV that corresponds to the higher frequencies. This process is also
called wraparound—the patient's arm gets “wrapped around” to the
opposite side.
The computer can't recognize frequencies outside the bandwidth (which
determines the FOV). Any frequency outside of this frequency range is
going to get “aliased” to a frequency that exists within the bandwidth.
The “perceived” frequency will be the actual frequency minus twice
the Nyquist frequency.
f (perceived) = f (true) −2f (Nyquist)
Why then do we usually see wraparound in the phase-encoding
direction? Remember that the number of phase-encoding steps is
directly related to the length of the scan time. The phase-encoding
steps can be lowered by shortening the FOV in the phase-encoding
direction versus the frequencyencoding direction also known as
rectangular FOV (see Chapter 23). If the FOV is shortened too much in
this direction versus the actual extent of the body then wraparound will
occur. Figure 18-2 contains an example of wraparound.
3D Imaging. Wraparound artifact can also be seen in 3D imaging in all
three directions.
1. It can be seen along the x and y directions, as with spin-echo
imaging.
2. It can also be seen along the slice-select (phase-encoded) direction
at each end of the slab (e.g., the last slice is overlapped on the first
slice, as in Figs. 18-3 and 18-4).

Example
Suppose the frequency bandwidth is 32 kHz (±16 kHz). This means
that if we're centered at zero frequency, the maximum frequency
fmax = +16 kHz and minimum frequency (−fmax) = −16 kHz (Fig. 18-
1). If we have a frequency in the arm (outside

the FOV) of + 17 kHz, the perceived frequency will be

f (perceived) = +17 kHz −2(16 kHz) = −15 kHz
Now, the arm, which is perceived as having a frequency of −15 kHz
(rather than +17 kHz), will be recognized as a structure with a very low
frequency—only 1 kHz faster than the negative end frequency of the
bandwidth—and will be identified on the opposite side of the image,
the low-frequency side.

Figure 18-2. Axial T1 (A) and PD (B) images of the lumbar spine
demonstrate aliasing of the arms in Figure B (arrows). Figure B was
obtained with a smaller FOV resulting in aliasing artifact. Also note
that the patient has a filum terminale lipoma (black arrows in A
and B).
 Figure 18-3. 3D gradient-echo T1 image without gadolinium of the
abdomen shows slice direction aliasing with the kidneys appearing
to be in the lungs (arrows). Also note that there is aliasing in the
phase-encoding direction (anteroposterior) from the inferior
abdominal image's anterior subcutaneous tissue “wrapping around”
posteriorly (arrowheads).

Remedy. How do we solve this problem?

1. Surface coil: The simplest way is to devise a method by which we
don't get any signal from outside the FOV. With the patient in a large
transmit/receive coil that covers the whole body, we will receive
signal from all the body parts in that coil, and those parts outside the
FOV will result in aliasing. But

if we use a coil that only covers the area within the FOV, we will only
get signal from those body parts within the maximum frequency
range, and no aliasing will result. This type of coil is called a surface
coil. We also use a surface coil to increase the signal-to-noise ratio
(SNR).
Figure 18-4. 3D coronal gradient-echo T1 image of the brain
shows slice direction aliasing of the anterior skull on to the brain.
 Figure 18-5. To avoid aliasing, increase the FOV.

2. Increase FOV: If we double the FOV to include the entire area of

study, we can eliminate aliasing. To do so, we have to use a weaker
gradient. The maximum and minimum frequency range will cover a
larger area, and all the body parts in the FOV will be included within
the frequency bandwidth; therefore, no aliasing will result (Fig. 18-
5). To maintain the resolution, double the matrix with a weaker
gradient (Gx). The maximum and minimum frequency range will still
be the same as the stronger gradient. They will just be spread out
over a wider distance. Remember, to increase the FOV, we have to
 use a weaker gradient.
3. Oversampling: Two types are discussed:
a. Frequency oversampling (no frequency wrap [NFW])
b. Phase oversampling (no phase wrap [NPW])
a. Frequency oversampling (NFW): Frequency oversampling
eliminates aliasing caused by undersampling in the frequency-
encoding direction (refer to the sampling theorem in Chapter
12). Oversampling can also be performed in the phaseencoding
direction by increasing the number of phase-encoding gradients.
b. Phase oversampling (NPW): We can double the FOV to avoid
aliasing and, at the end, discard the unwanted parts when the
image is displayed (Fig. 18-6). This is called no phase wrap
(NPW) by some manufacturers. It is also called phase
oversampling by other manufacturers. Because Ny is doubled,
NEX is halved to maintain the same scan time. Thus, the SNR is
unchanged. (The scan time might be increased slightly because
overscanning performs with slightly more than ½NEX.) An
example of this is seen in Figure 18-7.

4. Saturation pulses: If we saturate the signals coming from outside the

FOV, we can reduce aliasing.
5. 3D imaging: In 3D imaging, if we see this artifact along the slice-
select axis, we can simply discard the first and last few slices.
Chemical Shift Artifact. The principle behind the chemical shift
artifact is that the protons from different molecules precess at slightly
different frequencies. For example, look at fat and H2O. A slight
difference exists between the precessional frequencies of the hydrogen
protons in fat and

H2O. Actually, the protons in H2O precess slightly faster than those in
fat. This difference is only 3.5 ppm. Let's see what this means by an
example.
Figure 18-6. In no phase wrap, aliasing is avoided by doubling the
FOV in the y direction and, at the end, discarding the unwanted
part of the image.
 Figure 18-7. Sagittal STIR image (A) of the cervical spine with
craniocaudal phase-encoding direction demonstrates aliasing of the
brain onto the upper thoracic spine. (B) The same image after no
phase wrap was applied. Truncation artifact is also seen (arrows).

Example
Consider a 1.5-T magnet. The precessional frequency is as follows:

1. Frequency = ω0 = γB0 = (42.6 MHz/T) (1.5 T) ≈ 64 MHz = 64 × 106 Hz

2. 3.5 ppm = 3.5 × 10−6
3. (3.5 × 10−6)(64 × 106 Hz) ≅ 220 Hz
In other words, at 1.5 T, the difference in precessional frequency of the
hydrogen protons in fat and in H2O is 220 Hz.

Example
We now have a 0.5-T magnet. The precessional frequency of protons
in a 0.5-T magnet is 1/3 of a 1.5-T magnet. The frequency difference
is then
1/3(220 Hz) = 73 Hz
Therefore, at 0.5 T, the difference in precessional frequency of the
hydrogen protons in fat and in H2O is only 73 Hz. In other words, if
we use a weaker magnet, we will get less chemical shift.
How does this affect the image? Chemical shift artifacts are seen in the
orbits, along vertebral endplates, in the abdomen (at organ/fat
interfaces), and anywhere else fatty structures abut watery structures.
In a 1.5-T magnet, the sampling time (Ts) is usually about 8 msec. Let's
take 256 frequency points in the frequencyencoding direction.
BW = N/TS = 256/8 msec BW = 32 kHz
These formulas show that the entire frequency range (i.e., bandwidth)
of 32 kHz covers the whole length of the image in the x direction.
Because we have the FOV of the image in the x direction divided into
256 pixels, each pixel is going to have a frequency range of its own,
that is, each pixel has its own BW:
BW/pixel = 32 kHz/256 BW/pixel = 125 Hz
(This representation of BW on a “per pixel” basis is used by Siemens
and Philips. It has the advantage of less ambiguity than the ±16 kHz
designation, should the “±” be deleted.) Thus, each pixel contains 125
Hz of information (Fig. 18-8). Stated differently, the pixel bin contains
125 Hz of frequencies. Now, because fat and H2O differ in the
precessional frequency of hydrogen by 220 Hz at 1.5 T, how many pixels
does this difference correspond to?
Pixel difference = 220 Hz/125 Hz/pixel ≈ 2 pixels
Figure 18-8. At 1.5 T with a BW of 32 kHz and 256 pixels, there
 will be about 125 Hz/pixel (32 kHz/256 = 125 Hz), that is, there is
125 Hz of information in each pixel. This may be a better way of
describing the BW of a scanner because there is no ± confusion.

This means that fat and H2O protons are going to be misregistered from
one another by about 2 pixels (in a 1.5-T magnet using a standard ±16
kHz bandwidth). (Actually, it is fat that is misregistered because
position is determined by assuming the resonance property of water.) If
pixel size δx = 1 mm, this then translates into 2 mm misregistration of
fat.
MATH: For the mathematically interested reader, it can be shown that
Chemical shift

where γ = 42.6 MHz/T, B is the field strength (in T), BW is the

bandwidth (in Hz), and FOV is the field of view (in cm).
Let's now consider chemical shift artifact visually (Fig. 18-9).
Remember that H2O protons resonate at a higher frequency compared
with the hydrogen protons in fat. With the polarity of the frequency-
encoding gradient in the x direction set such that higher frequencies
are toward the right, H2O protons are relatively shifted to the right
(toward the higher frequencies), and fat protons are relatively shifted
to the left (toward the lower frequencies). This shifting will result in
overlap at lower frequency and signal void at higher frequency. This in
turn leads to a bright band toward the lower frequencies and a dark
band toward the higher frequencies on a T1-weighted (T1W) or proton
density (PD)-weighted conventional spin-echo (CSE) image. (On a T2W
CSE, fat is dark, so the chemical shift artifact is reduced.
Unfortunately, on a T2W FSE [fast spin echo—see Chapter 19], fat is
bright, and the chemical shift artifact is seen.) We will see this
misregistration artifact anywhere that we have a fat/H2O interface.
Also remember that this fat/H2O chemical shift artifact only occurs in
the frequency-encoding direction (in a conventional spin-echo image or
in gradient-echo [GRE] imaging).
 Figure 18-9. Chemical shift effect between fat and water causes a
bright band toward the lower frequencies (due to overlap of fat
and water at lower frequencies) and a dark band toward the higher
frequencies (due to lack of fat and water signals).

Example—Vertebral Bodies
With frequency-encoding direction—in this case, going up and down
(and “up” having higher frequency) —the fat in the vertebral body
would be misregistered down, making the lower endplate

bright due to overlap of water and fat and the top endplate dark due
to water alone (Fig. 18-10). If we increase the pixel size, the
misregistration artifact will increase.

Figure 18-10. Chemical shift artifact in the vertebral endplates

produces a dark band in the inferior endplates and a bright band in
the superior endplates (assuming the frequencyencoding direction
to be upward).
 Figure 18-11. Axial T2 FSE image shows alternating bright and dark
signal around the kidneys along the frequency-encoding
(transverse) direction (white arrows). Patient also has bilateral
pheochromocytomas (white arrowheads), a nonfunctioning islet
cell tumor in the tail of pancreas (wide black arrow) and a simple
cyst in the left kidney (black arrow). Patient had von Hippel-Lindau
syndrome.

Figures 18-11 through 18-15 are examples of chemical shift artifact.

 Figure 18-12. Axial T2* gradient-echo image of the knee shows
first-order chemical shift (arrows) along the frequency direction
(anteroposterior). Note that phase-encoding (transverse) direction
“ghosting” artifact is also seen (arrowhead). A knee effusion is also
demonstrated.

Question: What factors increase chemical shift artifact?

Answer:
1. A stronger magnetic field strength
2. A lower BW:
Figure 18-13. Coronal spoiled gradientecho T1 (TR 93/TE 1.8
msec) image of the abdomen shows the typical alternating bands of
dark and bright signal at the fat/water interfaces from chemical
shift artifact along the frequency-encoding direction
(craniocaudal).
Figure 18-14. PD (A) and T2 (B) images of the posterior fossa show
alternating bright and dark bands along the frequency-encoding
direction (anteroposterior; arrows). Notice that the thickness of
the artifact is wider in the T2 image secondary to a bandwidth
(BW) of ± 4 kHz versus the PD image's BW of ± 16 kHz. Also note
that the T2 image shows only the dark band well since the fat has
low signal in this CSE T2 sequence, which minimizes the amount of
bright signal. Patient had mature teratoma.
Figure 18-15. Coronal T2 (A) of the knee demonstrates chemical
shift artifact in the frequency-encoding direction (craniocaudal)
secondary to the fatty marrow juxtaposed to an enchondroma.
Additional sagittal T2 (B) image again shows the artifact; however,
now the bright and dark bands have swapped directions due to the
frequency-encoding direction swapping (e.g., lower frequencies
 are superior in the coronal and inferior in the sagittal image).
Finally, sagittal T2 image with chemical (spectral) fat saturation
(C) shows no chemical shift effect.

If we decrease the BW, we have a lower BW per pixel and fewer

frequencies/pixel. As an example, if instead of 32 kHz, we have a 16
kHz BW, then
BW/pixel = 160 kHz/256 = 62.5 Hz/pixel
Now, each pixel covers 62.5 Hz, but the chemical shift is still 220 Hz.
Consequently, 220 Hz/62.5 Hz/pixel ≅ 4 pixel misregistration
Decreasing bandwidth results in increased chemical shift artifact.
This is one of the side effects of selecting a lower BW on your scanner.
Unfortunately, the chemical shift due to field strength and that

due to BW are independent and additive; thus, higher field/low BW

techniques have the worst chemical shift artifact (Fig. 18-14).
3. Smaller pixels:
If we keep the BW of 32 kHz and the FOV the same but increase the
number of frequency-encoding steps to 512 (instead of 256), the pixel
bin will have half as many frequencies:
Pixel bin = 32 kHz/512 = 62.5 Hz/pixel
again leading to a greater chemical shift, as above (i.e., 4 pixels
instead of 2).
Solution. How can you fix chemical shift artifacts?
1. Get rid of fat using fat suppression. If there is no signal from fat,
there can be no chemical shift. This can be done with a spectroscopic
“fat sat” pulse or a STIR sequence (Fig. 18-15).
2. Increase pixel size by keeping FOV the same and decreasing Nx
(trade-off: deteriorates resolution).
3. Lower the magnet's field strength (not practical!).
 Figure 18-16. Chemical shift effect of the second kind. Fat and
water protons get in and out of phase at various values of TE.
Specifically, they are in phase at TE of 0, 4.5, and 9 msec and out
of phase at TE of 2.25, 6.75 msec, and so on. This effect can be
represented graphically by a sine function.

4. Increase bandwidth (trade-off: lowers SNR).

5. Switch phase and frequency directions. This will just change the
direction of the chemical shift.
6. Use a long TE (causes more dephasing and less signal from fat).
Chemical Shift of the “Second Kind.” This phenomenon applies to GRE
techniques (see Chapters 20 and 21). As discussed previously, fat and
water protons precess at slightly different frequencies in the transverse
plane (220 Hz at 1.5 T). Because water precesses faster, it gets 360°
ahead of fat after a short period of time. Thus, there will be times (TE)
when fat and water spins will be totally in phase and times when they
will be 180° out of phase. At 1.5 T, fat and water are in phase every
4.5 msec. This number is derived by the following:
Frequency difference between fat and water = 220 Hz
Period = 1 frequency = 1/(220 Hz) = 0.0045 sec = 4.5 msec
In Figure 18-16, fat and water are in phase initially at TE = 0 msec, go
out of phase at TE = 2.25 msec, and are back in phase at TE = 4.5 msec.
In general, at 1.5 T, fat and H2O go in and

out of phase every 2.25 msec. This is called a chemical shift effect of
the second kind.
Boundary Effect. If the selected TE is 2.25, 6.75, 11.25, 15.75 msec,
and so on, fat and water protons will be out of phase and a dark
boundary will be seen around organs that are surrounded by fat (such as
the kidneys and muscles). This result is called the boundary effect,
bounce point artifact, or India Ink Etching, which is caused by
chemical shift of the second kind. This type of imaging is referred to as
out-of-phase scanning, referring to the fact that at these TEs, fat and
water spins will be 180° out of phase. This phenomenon does not just
occur along the frequency-encoding axis (as with the chemical shift
artifact of the first kind) because it is a result of fat and water protons
phase cancellation in all directions (Fig. 18-17). (Boundary effect does
not occur in conventional SE techniques because of the presence of the
180° refocusing pulse, which is absent in GRE techniques.)
Figure 18-17. In-phase (A) and out-of-phase (B) spoiled gradient
T1 images show the “boundary effect” on the out-of-phase images
circumferentially at every fat/water interface (arrows in B). Also
 note that a left adrenal adenoma loses signal substantially on the
out-of-phase image (arrowhead).

Remedy
1. Make fat and H2O in phase by picking appropriate TE.
2. Increase the BW (trade-off: decreases SNR).
3. Use fat suppression.
Truncation Artifact (Gibbs Phenomenon). This artifact occurs at high
contrast interfaces (e.g., skull/brain, cord/cerebrospinal fluid (CSF),
meniscus/fluid in the knee) and causes alternating bright and dark
bands that may be mistaken for lesions (e.g., pseudo syrinx of the
spinal cord or pseudo tear of the knee meniscus).
The cause is inability to approximate exactly a steplike change in the
signal intensity due to a limited number of samples or sampling time.
The ripples in Figure 18-18 are responsible for

the parallel bands seen at such sharp interfaces. This artifact is seen
mostly in the phase direction (because we typically have few pixels and
lower resolution in phase compared with frequency). Incidentally, the
correct term is “truncation artifact.” “Gibbs phenomenon” refers to
the infinitely thin discontinuity that still persists with an infinite
number of pixel elements. Figures 18-7, 18-19, and 18-20 contain
examples of truncation artifact.
 Figure 18-18. Truncation artifact causes a ring-down effect
because the FT of a truncated sinc function has ripples at its
edges.

Remedy
1. Increase sampling time (↓ BW) to reduce the ripples. (Remember, a
wider signal in time domain means a narrower one in frequency
domain.)

Figure 18-19. Sagittal fat-saturated T2 image (A) shows minimal

truncation artifact (white arrow); a sagittal STIR image (B) shows
wider truncation artifact (white arrow). The T2 image was
obtained with 224 phase-encoding steps, and the STIR used 192
steps. Patient also had a small nonhemorrhagic cord contusion at
C1/2 (black arrows in A and B).
2. Decrease pixel size:
a. increasing the number of phase encodes, and
b. decreasing the FOV

Partial Volume Artifact. This artifact has the same concept as

computerized tomography (CT). To reduce it, decrease the slice
thickness (δz). Figure 18-21 contains an example of partial volume
artifact.

Patient-Related Artifact
This artifact is caused by voluntary or involuntary patient motion and
by the patient's anatomy. Pulsating motion in vessels is also an

interesting source of motion-related artifacts. (More on this in later

chapters.)
 Figure 18-20. Proton density sagittal image of the knee shows
truncation artifact mimicking posterior medial meniscus tear
(white arrow). Note extension of high signal beyond the meniscus
(black arrow).

Motion Artifact. Motion artifact is caused by the patient's (voluntary or

involuntary) movements (random) or by pulsating flow in vessels
(periodic). We only get motion artifacts in the phase-encoding
direction.
 Figure 18-21. Axial FLAIR image (A) shows some signal within a
right convexity lesion that would not be consistent with a simple
arachnoid cyst. Additional coronal FLAIR image (B) shows the signal
to be around, but not in, the lesion (arrows). The high signal on
the axial image was due to partial volume averaging. The high
signal was flow-related enhancement of CSF flowing around the
cyst.

Question: Why is motion artifact only seen in the phase-encoding

direction?
Answer: The reason is twofold:
1. First of all, motion along any magnetic field gradient results in
abnormal phase accumulation, which mismaps the signal along the
phase-encoding gradient.
 Figure 18-22. Ghost artifacts are equidistant replica of a pulsating
structure, such as the aorta, along the phase direction.

2. Also, there is a significant asymmetry in the data space (see Chapter

13) so that it takes much less time to sample the signal via frequency
encoding (on the order of milliseconds) than to do a single phase-
encoding step (on the order of seconds). Thus, most motions
experienced during clinical MRI are much slower than the rapid
sampling process along the frequency-encoding axis. This disparity
between frequency- and phase-encoding periods allows motion
artifacts to be propagated mainly along the phase-encoding axis.
Motion artifacts along the frequencyencoding axis may occur, but they
are insignificant (at best, they may cause minimal blurring).
Periodic Motion. Periodic motion is caused by pulsating or periodic
motion of vessels, heart, or CSF. In the example in Figure 18-22 (also
Fig. 18-23), with a cross section of the body through the aorta, and
with the phase-encoding in the AP direction, we will get “ghost”
artifacts of the aorta equally separated. The artifacts become fainter
with increasing distance from the original structure. The separation
(SEP) between the “ghosts” is given by

Another way of expressing this is

SEP = acquisition time/T(motion)
where T(motion) is the period of motion of the object (in this case, the
aorta).

Example
The aorta pulsates according to the heart rate. If the heart rate is
HR = 60 beats/min = 60 bpm = 1 beat/sec
then the period of motion = T(motion) = 1 sec.
 Figure 18-23. Axial STIR image of the neck showed marked
ghosting from both arteries and veins in the phase-encoding
direction (anteroposterior).

This means that we have a pulsation every 1 sec. For example, if we

have a TR = 500 msec = 0.5 sec, NEX = 1, Nx = 256, then
SEP = 0.5 × 256/1 = 128/1 = 128 pixels
Therefore, we get two ghosts in the image. If the heart rate is 120
bpm, then we get
SEP = 128/0.5 = 256 pixels
 and only one ghost.

If we multiply this by pixel size, we get the distance between the

“ghosts.” Therefore, if we increase TR, number of phase-encoding
steps, or NEX, we can increase the separation of ghosts so that they
won't be so numerous within the body part we're studying. More rapidly
pulsating flow (i.e., shorter period) also causes more separation. If the
FOV is too small, the “ghost” images outside the FOV might get
“aliased” into the FOV. The ghosts may be dark or bright depending on
the phase of the pulsating structure with respect to the phase of the
background. If they are in phase, they'll be bright, and if they are out
of phase, they'll be dark—Figures 18-24 and 18-25.

Remedy
1. Use spatial presaturation pulses to saturate inflowing protons and
reduce the artifacts.
Figure 18-24. Axial spoiled gradient image of the abdomen shows
both positive (black arrow) and negative (white arrow) “ghost”
artifacts from the aorta. The negative “ghost” mimics a vertebral
body lesion.
 Figure 18-25. Axial T2 image of the thoracic spine shows positive
“ghost” artifacts from the CSF that simulate lung nodules.

2. Increase separation between ghosts by increasing TR, Ny, or NEX

(which is tantamount to increasing scan time).
3. Swap phase and frequency: although this only changes the direction
of the artifacts, it does allow differentiation between a true lesion
and an artifact.
4. Use cardiac gating.
5. Use flow compensation.
Random Motion. Random motion is caused by the patient's voluntary or
involuntary movements (e.g., breathing, changing position, swallowing,
tremors, and coughing). It causes blurring of the image. We may get
parallel bands in the phaseencoding direction as well (Fig. 18-26).
Although

this may simulate truncation artifacts, it is different in that truncation

causes fading parallel bands.
Figure 18-26. Axial FLAIR image shows marked motion artifact in
the phase-encoding direction (transverse) from eye movement.
 Figure 18-27. Axial CSE T1 image with fat saturation post
gadolinium (A) shows both periodic and random phase direction
(anteroposterior) artifact (arrows). A breath-hold spoiled gradient
T1 image with fat saturation post gadolinium (B) shows near
complete resolution of motion artifact. Note that there is
increased magnetic susceptibility artifact (arrow in B) in the
gradient-echo image. Patient had a liver hemangioma.

Remedy
1. Patient instruction: Don't move! (probably the most useful remedy)
2. Respiratory compensation (RC) (uses chest wall motion pattern to
reorder scan and minimize motion)
3. Use of glucagon in the abdomen to reduce artifacts due to bowel
peristalsis
4. Sedation
5. Pain killers
6. Faster scanning (FSE, GRE, EPI, etc.); sequential 2D rather than 3D
scanning (see Fig. 18-27 for an example).
CSF Flow Effects. Dephasing of protons due to CSF motion may
sometimes simulate a lesion. Flow compensation techniques can reduce
this effect. Examples include the following:
1. Pseudo aneurysm of basilar artery due to pulsatile radial motion of
CSF around it (Fig. 18-28).
2. Pseudo MS plaques in the brainstem due to CSF flow in the basal
cisterns.
3. Pseudo disc herniation, again secondary to CSF flow.

Remedy
1. Be certain that “lesions” are seen on all pulse sequences (artifacts
tend to only be seen on one image).
2. Use cardiac gating.
3. Use flow compensation.
Figure 18-28. Axial T2 image shows marked signal void around the
basilar artery mimicking an aneurysm in this 3-year-old patient.
 Figure 18-29. Sagittal PD (A) and T2 (B) fat-saturated images of
the knee show magic angle artifact as seen by increased signal on
the short TE PD image (arrow in A), whereas the tendon itself is
not thickened and has dark signal on the T2 image (B). Joint
effusion is also seen. (Courtesy of D. Beall, MD, San Antonio,
Texas.)

Magic Angle Artifacts. In imaging the joints, if a tendon is oriented at a

certain angle (55°) relative to the main magnetic field, then the tendon
appears brighter on T1-and PD-weighted images, but normal on T2-
weighted images. This artifactual increased intensity might potentially
be confused with pathology.
Collagen, which is responsible for the majority of tendon composition,
has an anisotropic structure. This anisotropic structure has properties
that vary with the direction of measurement and is responsible for
dependence of T2 of tendons on their orientation. (Isotropic structures,
however, have properties independent of their orientation.)
At the magic angle, the T2 of the tendon is slightly increased. This
increase is negligible when TE is long. However, when TE is short (as in
T1- or PD-weighted images), the result is increased signal intensity. The
mathematics behind this T2 prolongation has to do with some of the
mathematical terms in the Hamiltonian going to zero at θ = 55° (see
Figs. 18-29 and 18-30 for examples).
MATH: This magic angle effect is the solution to the equation
3(cos θ)2 − 1 = 0 → (cos θ)2 = 1/3
or

which is calculated to be θ ≈ 55°. The above equation comes from a

complicated mathematical theory dealing with the so-called dipolar
Hamiltonian.

RF-Related Artifacts
Cross-talk. We have already discussed this issue in previous chapters.
The problem arises from the fact that the Fourier transform (FT) of the
RF pulse is not a perfect rectangle but rather has side lobes (Fig. 18-
31). We shall use a simpler version of the RF profile, as in Figure 18-32.
If we consider two adjacent slices, there will be an overlap in the FT of
their RF pulses (Fig. 18-32). Cross-talk causes the effective TR per slice
to decrease (due to saturation of protons by the RF signals for adjacent
slices). Thus, more

T1 weighting will result (this is particularly problematic for PD- and T2-
weighted images). Also, due to reduced effective TR, the SNR will
decrease.
Figure 18-30. Angled sagittal PD (A) and T2 (B) fat-saturated
images of the shoulder demonstrate magic angle artifact of the
intra-articular biceps tendon. There is increased signal on the PD
image (A), whereas the tendon has dark signal and overall normal
appearance on the T2 image. Acromioclavicular joint high signal is
from osteoarthritis changes. (Courtesy of D. Beall, MD, San
Antonio, Texas.)
Figure 18-31. The actual RF has a finite time span, yielding side
lobes or rings. A Gaussian RF pulse has a Gaussian FT.

Figure 18-32. Side lobes of the FT of RF pulses (such as in the case

of Gaussian curves) may overlap, causing cross-talk.
 Figure 18-33. To reduce cross-talk, gaps are introduced between
slices.

Figure 18-34. The larger the interslice gap is, the less cross-talk is
observed.

In short, cross-talk causes increased T1 weighting and decreased SNR.

Remedy
1. Gaps can be introduced between adjacent slices (Fig. 18-33).
2. Two acquisitions with 100% gaps can be interleaved.
3. The RF pulse can be lengthened to achieve a more rectangular pulse
profile.
Let's discuss these in more detail:
1. If we increase the gap between slices, we reduce cross-talk (Fig.
18-34). The tradeoff is an increase in the unsampled volume and
the increased potential for missing a small lesion located within
the gap.
2. It doesn't matter which way we order the slices (we can do slice 1,
then slice 3, then slice 2, etc.). Adjacent slices still will be sharing
a certain frequency range and cause cross-talk. The only way to
eliminate cross-talk is to do two separate sequences each with a
100% gap, such as:

Figure 18-35. The closer the profile of the RF pulse (actually its
FT) is to a rectangle, the better we can achieve contiguous slices
without encountering cross-talk.

First sequence: odd slices 1, 3, 5, 7, …

Next sequence: even slices 2, 4, 6, 8, …
This is the technique of “true” interleaving. Interleaving within a single
sequence will not totally eliminate cross-talk, although it might reduce
it somewhat. The interslice gap in this case is usually 25% to 50% of the
slice thickness and a simple sequence is performed. Interleaving in the
true sense, however, will double the scan time because it employs two
separate sequences.
Contiguous Slices. The RF pulse on newer scanners more closely
approximates a rectangular wave (Fig. 18-35). With this feature, we
may have a 10% to 20% interslice gap without significant cross-talk.
However, with reduced interslice gap, we reduce coverage and need
more slices. Remember, we are talking tradeoffs again.
RF Zipper Artifact. This artifact is one form of central artifacts (the
other form is RF feedthrough,

discussed later). They are referred to as zippers due to the formation

of a central stripe of alternating bright and dark spots along the
frequency-encoding axis (at zero phase), as in Figure 18-36. Two
sources of zipper artifacts are discussed here.
 Figure 18-36. Zipper artifact at zero phase.

FID Artifacts. Free induction decay (FID) artifacts occur due to

overlapping of side lobes of the 180° pulse with the FID, before it has
had a chance to completely decay (Fig. 18-37). This overlapping causes
a “zipper” artifact along the frequency-encoding direction.

Remedy
1. Increase the TE (increases the separation between the FID and the
180° RF pulse).
2. Increase slice thickness (δz). This in effect results from selecting a
wide RF BW, which narrows the RF signal in the time domain, thus
lowering chances for overlap.
Stimulated Echo. This artifact also appears as a narrow- or wide-band
noise in the center along the frequency-encoding axis. The mechanism
is similar to FID artifacts. In this case, imperfect RF pulses of adjacent
slices or imperfect 90° −180° −180° pulses of a dual-echo sequence
form a stimulated echo that may not be phase-encoded, thus appearing
in the central line along the frequency-encoding axis.
Figure 18-37. FID artifact. The side lobes of the 180° and the FID
may overlap, causing a zipper artifact at zero frequency along the
phase direction.
Figure 18-38. RF feedthrough causes a zipper artifact at zero
 frequency along the phase direction.

Remedy
1. Use spoiler gradients.
2. Adjust the transmitter.
3. Call the service engineer.
RF Feedthrough Zipper Artifact. This artifact occurs when the
excitation RF pulse is not completely gated off during data acquisition
and “feeds” through the receiver coil. It appears as a “zipper” stripe
along the phase-encoding axis at zero frequency (Fig. 18-38).
Remedy. Alternate the phase of the excitation RF pulses by 180° on
successive acquisitions; the averaged phase-alternated excitations will
essentially eliminate RF feedthrough.
RF Noise. RF noise is caused by unwanted external RF noise (e.g., TV
channel, a radio station, a flickering fluorescent light, patient
electronic monitoring equipment). It is similar to RF feedthrough
except that it occurs at the specific frequency (or frequencies) of the
unwanted RF pulse(s) rather than at zero frequency (Fig. 18-39).

Remedy
1. Improve RF shielding.
2. Remove monitoring devices if possible.
3. Shut the door of the magnet room!

External Magnetic Field Artifacts

Artifacts related to B0 are usually caused by magnetic inhomogeneities.
These nonuniformities are usually due to improper shimming,
environmental factors, or the far extremes of newer short-bore
magnets. This can lead to
image distortion (Fig. 18-40). They can be reduced in SE and FSE
imaging by using 180° refocusing pulses. They can be a source of image
inhomogeneity when a fat suppression technique is used.
 Figure 18-39. Axial T2 image shows RF noise (arrows) from
monitoring devices in this recent postoperative patient. There is
also an epidural hematoma (arrowhead).

In GRE imaging, small spatial nonuniformities cause moiré fringes

(zebra pattern) due to the overlay of the primary image and aliased
overlay (Fig. 18-41).
Remedy. Appropriate shimming coils (auto shimming) can minimize
the problem.

Magnetic Susceptibility Artifacts

As discussed in Chapter 2, all substances get magnetized to a degree
when placed in a magnetic field, and their magnetic susceptibility
(denoted by the Greek symbol χ) is a measure of how magnetized they
get.
There are three types of substances—each with a different magnetic
susceptibility—commonly dealt with in MRI: paramagnetic, diamagnetic,
and ferromagnetic. These substances were described in Chapter 2 and
are briefly reviewed here:
1. Diamagnetic substances with no unpaired electrons have negative
magnetic susceptibility χ (i.e., χ < 0 and µ = 1 + χ < 1). They are
basically nonmagnetic. The vast majority of tissues in the body have
this property.
2. Paramagnetic substances contain unpaired electrons, have a small
positive χ (i.e., χ > 0 and µ > 1), and are weakly attracted by the
external magnetic field. The rareearth element gadolinium (Gd) with
seven unpaired electrons is a strong paramagnetic substance. Gd is a
member of the lanthanide group in the periodic table. The rare-
earth element dysprosium (Dy) is another strong paramagnetic
substance that belongs to this group. Certain breakdown products of
hemoglobin are paramagnetic: deoxyhemoglobin has four unpaired
electrons and methemoglobin has five. Hemosiderin, the end stage of
hemorrhage, contains, in comparison, more than 10,000 unpaired
 electrons. It is in a group of substances referred to as
superparamagnetic, with magnetic susceptibilities 100 to 1000 times
stronger than paramagnetic substances.
3. Ferromagnetic substances are strongly attracted by a magnetic field
and have a large positive χ, even larger than that of
superparamagnetic substances. Three types of ferromagnets are
known: iron (Fe), cobalt (Co), and nickel (Ni). Susceptibility artifacts
in MRI occur at interfaces of differing magnetic susceptibilities, such
as at tissue-air and tissue-fat interfaces (examples include paranasal
sinuses, skull base, and sella). These differences in susceptibilities
lead to a distortion in the local magnetic environment, causing
dephasing of spins, with signal loss, mismapping (artifacts), and poor
chemical fat saturation

(Figs. 18-42 through 18-44). Ferromagnetic substances (such as

metallic clips and foreign bodies), with their large susceptibilities,
lead to substantial field distortion and artifacts (Figs. 18-45 through
18-47).
 Figure 18-40. Axial T2 fat-suppressed image (A) shows distortion
of the upper abdomen and lack of effective fat saturation
secondary to magnetic field inhomogeneities at the fringe of a
short-bore magnet. Comparative slice more inferiorly (B) has
expected fat saturation and normal appearance without distortion.
In another patient (C), a sagittal T1 image of the spine shows
distortion of the extreme cranial and caudal features in a short-
bore magnet. Patient also had two lower thoracic spine
compression fractures (arrows).

Question: Which MR technique is least sensitive to magnetic

susceptibility effects?
Answer: In decreasing order, echo planar imaging (EPI), gradient-echo
(GRE) imaging, conventional spin echo (CSE), and fast spinecho (FSE).
FSE is least sensitive to magnetic susceptibility effects due to the
presence of multiple refocusing 180° gradients.

Gradient-Related Artifacts
Eddy Currents. Eddy currents are small electric currents that are
generated when the gradients are rapidly switched on and off (i.e., the
resulting sudden rises and falls in the magnetic field produce electric
currents). These currents will result in a distortion in the gradient
profile (Fig. 18-48) and in turn cause artifacts in the image.
Nonlinearities. Ideal gradients are linear. However, as in other aspects
of life, there is no such thing as an ideal gradient. These nonlinearities
cause local magnetic distortions and image

artifacts. The effect is similar to artifacts related to B0

inhomogeneities.
Figure 18-41. Coronal postgadolinium spoiled gradient T1 image
with chemical (spectral) fat saturation (A) shows moiré fringes
(black arrows). Comparison with single-shot FSE (SSFSE) T2 (B)
image demonstrates decreased artifact. Also note ghosting on
image A from the heart and aorta (white arrow) and increased
magnetic susceptibility from an inferior vena cava (IVC) filter
(white arrowhead).

Figure 18-42. Axial T2 FSE with inhomogeneous fat saturation

shows “ghosting” artifact from the unsaturated anterior abdominal
subcutaneous fat.
Figure 18-43. Axial postgadolinium fat-saturated gradient-echo T1
image of the abdomen (A) shows “blooming” artifact from the
interface of the diamagnetic gas and the adjacent soft tissues (best
at the splenic flexure). This effect is minimized on the T2 FSE (B)
image. Note aliasing in the phase-encoding (anteroposterior)
direction on both images. Image A also has inhomogeneous fat
saturation at the diamagnetic interface.
Figure 18-44. Coronal postgadolinium gradient-echo T1 image with
fat saturation shows magnetic susceptibility from the densely
concentrated paramagnetic substance gadolinium, resulting in dark
renal collecting systems with a fringe of bright signal. Also note
mild moiré fringes artifact.
Figure 18-45. Axial T2 FSE image shows metallic susceptibility
artifact from an MRI compatible aneurysm clip in the area of the
left internal carotid terminus.
Figure 18-46. Coronal T1 image shows metallic susceptibility from
metallic foreign body at the base of the fifth digit.
Figure 18-47. Axial EPI B0 (A), CSE T2 (B), CSE PD (C), and FSE T2
(D) images show the varying effects of different pulse sequences on
metallic susceptibility in a patient with dental braces. The EPI is
the worst. The CSE T2 is worse than the PD due to a lower BW (±4
kHz) for the T2 versus the higher BW (±16 kHz) for the PD. Finally,
the T2 FSE is the best (BW still ±16 kHz), secondary to multiple
180° refocusing pulses.
Figure 18-47. (Continued.)

Figure 18-48. Eddy currents result from rapid on-and-off switching

of the gradient and cause distortion in the gradient profile and thus
the image.
 Figure 18-49. Nonlinearities in the gradient cause distortion in the
image. For instance, a circle may appear elliptical.

Geometric Distortion. Geometric distortion is a consequence of

gradient nonlinearities or gradient power drop-off. Figure 18-49
illustrates this concept. The real gradient has dampened peaks, causing
image distortion (e.g., a circle may appear elliptical). (Figure 18-50 is
an example due to gradient nonlinearities in the more demanding echo
planar sequence.) If you find this is a problem, then you need to call
your service engineer to fix it.

Figure 18-50. Axial T2-weighted EPI image (A) of the abdomen

shows geometric distortion of the normal ovoid shape of the
 abdomen. An accompanying SSFSE T2 image (B) shows this patient's
true shape.

Errors in the Data

Errors in the data are caused by a single calculation error in processing
the data related to the kspace of a single slice. The result is a crisscross
striation artifact that is present across a single image and not present
on any other image. (See Figure 18-51.)

Remedy
1. Delete the discrete error and average out the neighboring data.
2. Simply repeating the sequence solves the problem.
 Figure 18-51. Axial T2-weighted image of the cervical spine on a
0.23 T magnet shows diagonal lines coursing throughout the image
related to a single data point error in k-space.

Flow-Related Artifacts
Motion artifacts were discussed previously, including periodic flow
artifacts. Other flow-related phenomena are discussed in Chapters 26
and 27.

Dielectric Effects
As the wavelength of the radiowave approaches the dimensions of the
body part being imaged, there can be areas of brightening and
darkening due to standing waves. This is most pronounced at 3 T and
above. Since the body is a conducting medium, the artifact is often
called “dielectric effect” (Fig. 18-52). It seems to be worse in large
body parts, that is, the abdomen, and seems to be quite common when
ascites is present. The solution for dielectric effects is parallel
transmission or “transmit SENSE.”

Figure 18-52. 3.0 T proton density-weighted image of the

abdomen demonstrates cirrhosis and ascites with regional area of
decreased signal intensity due to dielectric effect or “standing
waves” (arrow).
 Key Points
In this chapter, we discussed the most common and
important causes of potential artifacts in MR imaging
that every MR radiologist should be aware of. For a
list of these artifacts, refer to the Introduction in this
chapter. There are a few other less significant
sources of artifacts that were not discussed in this
chapter.

Questions
18-1 Regarding chemical shift artifact:
(a) protons in fat resonate at 3.5 ppm higher than protons in
water
(b) at 1.5 T it is about 220 Hz
(c) at 1.5 T and for a 32 kHz BW and 256 × 256 matrix, it is about
2 pixels
(d) all of the above
(e) only (b) and (c)

18-2
(a) Determine the chemical shifts (in terms of numbers of pixels)
for the following situations (assume 256 frequency-encoding
steps):

B0 0.2 T 0.5 T 1.0 T 1.5 T

BW

50 kHz

10 kHz
 4 kHz

(b) Repeat this table in terms of millimeters, given an FOV = 24

cm = 240 mm.
(c) What is your conclusion?

18-3 Periodic motion causes “ghost” artifacts along the phase-

encoding direction. The number of pixels between two consecutive
ghosts is given by (SEP = separation)
SEP = TR • NEX • Ny/T = acquisition time/T
where T = period of the oscillating motion.
(a) Calculate SEP for the aortic ghosts (HR = 60 bpm, i.e., T = 1
sec) when TR = 200 msec = 0.2 sec, NEX = 1, Ny = 256.
(b) What is the maximum number of ghosts you could potentially
see along the phase-encoding axis in example (a)?
(c) What is the effect of increasing NEX?

18-4 Wraparound can be reduced by all of the following except:

(a) using a surface coil
(b) decreasing the FOV
(c) using presaturation pulses
(d) using a no phase wrap option
(e) using a no frequency wrap option

18-5 Truncation artifacts can be reduced by all of the following

except:
(a) decreasing pixel size
(b) increasing sampling time
(c) increasing Ny
(d) increasing FOV
18-6 T/F Chemical shift artifact causes a bright band toward the
higher frequency and a dark band toward the lower frequency at a
water/fat interface.

18-7 Chemical shift is decreased by all of the following except:

(a) lowering the bandwidth
(b) using a fat suppression technique
(c) using a lower field magnet
(d) using a longer TE

18-8 T/F Fat and water protons get out of phase at TE of odd
multiples of 2.25 msec.

18-9 Chemical shift in general can be represented by

(a) 3.5 × 10−6 γB · Nx/BW
(b) 3.5 × 10−6 γB · FOV/BW
(c) 3.5 × 10−6 γB/(BW · Nx)
(d) both (a) and (b)

18-10
(a) Calculate the separation (in pixels and mm) between aortic
ghosts for TR 500 msec, NEX 1, Ny 128, HR 80 bpm, and FOV 20
cm.
(b) What is the maximum number of ghosts you could potentially
see within the FOV?

18-11 Paramagnetic elements include all of the following except:

(a) gadolinium
(b) dysprosium
 (c) cobalt
(d) methemoglobin
(e) both (c) and (d)

18-12 Motion artifacts can be reduced by all of the following except:

(a) fast scanning
(b) sedation
(c) 3D imaging
(d) flow compensation

18-13 CSF flow can lead to all of the following artifacts except:
(a) pseudo MS plaques in the brainstem
(b) pseudo disc herniation
(c) pseudo basilar artery aneurysm
(d) pseudo syrinx

18-14 T/F Magic angle artifact demonstrates increased signal on

proton density images in a tendon that is positioned perpendicular to
the main magnetic field.

18-15 Cross-talk artifact can be reduced by all of the following

except:
(a) increasing the gradient strength
(b) increasing interslice gaps
(c) double acquisition with 100% gaps interleaved
(d) improving the RF profile

18-16 The number of ghost artifacts can be reduced by all of the

following except:
(a) flow compensation
 (b) presaturation pulses
(c) decreasing Ny
(d) increasing TR

18-17 Truncation artifacts include

(a) pseudo meniscal tear
(b) pseudo syrinx
(c) pseudo MS plaques
(d) all of the above
(e) only (a) and (b)
(f) only (a) and (c)

18-18 T/F Motion artifacts occur only along the phase-encoding

direction.
19
Fast Spin Echo

Introduction
In this chapter, we will discuss the elegant and cunning technique of
fast spin echo. This technique was first proposed by Hennig et al.1 and
was called RARE (rapid acquisition with relaxation enhancement).
However, it is commonly referred to as fast spin echo (FSE) or turbo
spin echo (TSE). Different manufacturers have different names for it
(Table 19-1).
Consider the pulse sequence diagram in Figure 19-1. This pulse
sequence can be used for either a CSE or an FSE study.

Conventional Spin Echo

Let's first talk about a conventional spin echo (CSE or SE) study and
again see how the lines in k-space are filled in. With a CSE,
immediately after the 90° RF pulse, an FID (free induction decay) is
formed. At a time TE after the first 90° pulse (time TE/2 after the first
180° refocusing pulse—17 msec in our example), we receive the first
spin echo. We have a whole train of 180° refocusing pulses, after each
of which we get another echo. Each echo is a multiple of 17 msec.
Notice that each successive echo has less amplitude as a result of T2
decay.
In a CSE sequence, we can get two echoes, that is, we apply two 180°
RF pulses and get back an echo from each pulse, each with a different
TE. However, we could have as many echoes as we want in a CSE
sequence. In our example, we have an eight-echo sequence, all easily
occurring within the time of one TR.

Table 19-1
 Manufacturer Name

GE , Hitachi, Toshiba Fast spin echo (FSE )

Siemens, Philips Turbo spin echo (TSE )

With each TR in a CSE, we have a single phase-encoding step. Each of

the echoes following each 180° pulse is obtained after a single
application of the phase-encoding gradient in CSE. Each echo has its
own k-space, and each time we get an echo, we fill in one line of k-
space (Fig. 19-1).
In CSE, each k-space will generate a different image, that is, we'll get a
first echo image, a second echo image, and so on. With eight 180°
pulses generating eight echoes, we'll have eight different k-spaces and
eight different images. If we had 256 different phase-encoding steps,
we would do this 256 different times. The scan time would be Scan
time = (TR) (number of phase-encoding steps) (number of excitations
[NEX]) Within the scan time, we'll get eight images, one of each TE
echo. If we were only interested in the last echo image, we wouldn't
have to bother with filling in the k-spaces for the first seven echoes.
However, the first seven echoes come “free.” We don't save any time
by not doing them because we have to wait out the time it takes to

get to the last echo anyway. In a dual-echo CSE sequence, the first echo
is always “free”—it doesn't cost any time. (However, as we shall see
later, that isn't true with FSE.)
 Figure 19-1. A spin-echo PSD with eight echoes. The echo spacing
(ESP) is 17 msec.

Thus, for each k-space in CSE, we repeat the TR 256 times (each at a
different phase-encoding gradient) and fill in the k-space for each echo
with 256 different lines. For an eight-echo train, we get eight different
images.
 Figure 19-2. In FSE, k-space is filled eight lines at a time in one
shot (within one TR).

Fast Spin Echo

By using the same example, we can see how FSE works. FSE is a very
elegant way of manipulating the CSE technique to save time. Again,
we'll start with a train of eight echoes (ETL = 8). However, now we will
only have one k-space. We'll fill this k-space eight lines at a time (Fig.
19-2). Instead of having eight separate

k-spaces, one for each echo, we will have one k-space using the data
from all eight echoes.
 Figure 19-3. After two shots, 16 lines in k-space will be filled.

Within the time of one TR (one “shot”), we will get eight lines, one
from each echo, in the single k-space. With the next TR, we'll
accumulate eight more lines, one from each echo, and we'll also put
them into the same k-space (Fig. 19-3).
For each shot/TR, we will fill in another eight lines into the single k-
space. Because we have a total of 256 lines in k-space, and because
during each TR we are filling in eight lines of k-space at a shot, we only
have to repeat the process 32 times (i.e., 256/8 = 32) to fill 256 lines of
k-space.
In CSE, it took one TR for each line of k-space. Therefore, in a CSE
study, we have to repeat the TR 256 times. In this manner, we have cut
the time of the study by a factor of eight.

Example
Acquisition time of CSE sequence with TR = 3000 msec, Ny = 256, NEX =
1 is
SE time = (TR)(number of phase-encoding steps)(NEX)
= (3000)(256)(1)msec
= 12.8 min
This time for the FSE sequence with TR = 3000 msec is
FSE time = (TR)(number of phase-encoding steps/ETL)(NEX)
= (3000)(256/8)(1)msec
= 1.6 min
In this example, we have shortened the time of the study from 12.8 min
to 1.6 min, a factor of eight times faster than the CSE study.

Echo Train Length

Echo train length (ETL) refers to the number of echoes used in FSE. The
ETL ranges typically from 3 to 32. The time interval between successive
echoes (or between 180° pulses) is called the echo spacing (ESP). A
typical ESP is on the order of 16 to 20 msec at a typical high field
bandwidth (BW) of 32 kHz (±16 kHz).
Figure 19-4 is an example. Let's say we want an image with contrast
reflecting a TE of approximately 100 msec. In FSE, the only TEs we can
choose are integral multiples of the ESP (ESP = 17 msec in our
example). This is called the effective TE (TEeff). We will see shortly
that this is not a true TE. Therefore, in our example, TEeff=102 msec (=
6 × 17 msec).
Remember that the center of k-space has maximum signal, and as we
go out to the edges of k-space, we get less and less signal. Therefore, if
we divide k-space into 8 slabs of 32 lines each (1 slab for each echo),
the center slab will be assigned to the sixth echo, that is, the echo
corresponding to the chosen TEeff = 102 msec (Fig. 19-5).
 Figure 19-4. An example of FSE with TEeff of 102 msec. The phase
gradient corresponding to this echo is minimum. The associated
echo is placed in the center row of k-space.

In FSE, before each 180° pulse, we place a different value of the phase-
encoding gradient. For the 180° pulse before the echo we choose as the
TEeff (in this case, 102 msec), we use a phaseencoding gradient with
the lowest strength. Each subsequent phase-encoding step will have a
gradient with more and more amplitude. This increase will result in the
most signal coming from the echo at 102 msec (because this signal is
obtained with a minimum phase gradient) and

decreasing signal from all the other echoes because their signals are
obtained with increasing phase-encoding gradients (Fig. 19-6).
Figure 19-5. The 32 lines in the center of k-space correspond to
the echoes associated with the weakest phase gradients.

Figure 19-6. The echo corresponding to TEeff of 102 msec has the
largest peak.
In the next TR, we again pick a phase-encoding step for the sixth echo
that will be close to zero gradient and all the other phase-encoding
steps will be close to their previous respective gradient values, so that
again the maximum signal during this TR will be obtained from echo 6
(TE = 102 msec) and progressively weaker signals will be obtained from
the other echoes.
The signals from the sixth echo (from the 1st to the 32nd TR) will all be
placed in the center of k-space (Fig. 19-5). The signals from the other
echoes will be placed in the other slabs. The echoes that experience
progressively greater phase-encoding gradients (and therefore less
signal) fall into slabs further away from the center slab, and those
echoes experiencing the weakest phase-encoding gradients (and
therefore having more signal) are placed closer to the center slab. k-
space is organized so that the greatest amount of signal comes from the
center of k-space and the least amount of signal comes from the
periphery of k-space.
Therefore, if we choose an ETL of 8 echoes, we will have 8 slabs in k-
space, with each slab containing 32 lines from 32 shots. Echo slab
corresponds to a different echo. Let's see what the echo looks like (Fig.
19-6).
Because the center slab belongs to the lowest phase-encoding gradient,
it will have the least amount of dephasing. The signal received at TE
=102 msec will have the greatest amplitude. As we go farther away
from 102 msec in either direction, the signal amplitude will get
progressively smaller because the phase-encoding gradients get
progressively larger.
By definition, the maximum signal comes at the TEeff time. But we still
get echoes from the other TEs that do not help our contrast. The
signals from these other echoes are all in the same k-space. Even
though the respective signal amplitudes from the other TEs are
progressively smaller, the further away in time they are from the TEeff;
they still contribute to the contrast from the TEeff. This is why it is
called an effective TE and not a true TE.
In a way, what we are doing is averaging the echoes, although it is a
weighted average. By appropriately picking the slabs, we put most of
the weight on the echo corresponding to 102 msec (the TEeff) and less
weight on the other echoes. As we go away from the center slab of k-
space, we are reducing the weighting, that is, we are reducing the
contribution of the data on that slab to the effective echo.
The previous example gives us a long TR/long TE, which is a T2-
weighted image. Now we want to do a proton density-weighted image
with a long TR/short TE and a TEeff of approximately 30 msec (Fig. 19-
7). In this case, we would assign the center slab of k-space to
correspond to the second echo, that is, TE =

34 msec. The echo with maximum magnitude will be at a TE of 34

msec, and the magnitude of the signal would fall off progressively with
subsequent echoes. This is because, now, the second echo will be
assigned the weakest phaseencoding gradients, and the subsequent
echoes will have progressively stronger phase-encoding gradients.
 Figure 19-7. Another example with TEeff of 34 msec. Here, the
largest echo corresponds to TE of 34 msec.

Remember that with the TEeff of 34 msec, we are still getting the
cumulative signal from the entire echo train of eight echoes. So even
information from an echo corresponding to a TE of 136 msec (8 × 17 =
136) is contributing to the signal of the TEeff of 34 msec, which we don't
want. Therefore, for a T1-weighted study, we usually pick a smaller ETL
such as 4. With an ETL of 4, we would only do four phase-encoding
steps. Thus, k-space would only have four slabs, and the longest echo
contributing to the contrast of the shortest TEeff (e.g., 17 msec) would
be the echo at 68 msec (4 × 17 = 68). This will eliminate the T2 effect
on a T1-weighted image caused by the signal contribution of the longer
echo times.
Question: What happens to the time it takes to do an FSE study when
we decrease the ETL from 8 to 4?
Answer: The time of the study will be increased by a factor of 2.
Question: If we have an ETL of 8, how many lines of k-space do we fill
in with each TR?
Answer: We fill in eight lines of k-space with each TR, each line going
to a separate slab within k-space.
Question: How many times do we have to repeat the TR to fill up k-
space with 256 lines?
Answer: In general, we have to repeat it by the following ratio:

Remember that scan time is calculated by the following formula:

The numerator of the formula is the same as for a CSE study:

Scan time (CSE) = (TR)(Ny)(NEX)
The difference in FSE is that the numerator is divided by the ETL. So,
in the case of an 8-ETL study, the number of times we would have to
repeat the TR (i.e., the number of shots) to complete 256 lines in k-
space would be
256/8 = 32
In the case of a 4-ETL study, we would fill four lines of k-space with
each TR. The number of times we would have to repeat the TR to
complete 256 lines in k-space is now
256/4 = 64 times
Example 1
1. Consider a T2-weighted study with TR = 3000 msec, Ny = 256 and
NEX = 1, and let's compare the scan time between CSE and FSE
(using a TEeff = 102 msec and ETL = 8).

Scan time (CSE) = (TR)(Ny)(NEX) = (3 sec)(256)(1) = 12.8 min

2. Repeat the above for a T1-weighted study with TR = 500 msec and
FSE using ETL = 4

Scan time (CSE) = (TR)(Ny)(NEX) = (0.5 sec)(256)(1) = 128 sec = 2 min, 8

sec
 Figure 19-8. Increasing the ETL causes a reduction in coverage
(number of slices).

Trade-offs
What are the trade-offs of FSE imaging?
Slice Coverage. As we increase the ETL to increase the speed of the
exam, we also decrease the number of slices we can do in a study (Fig.
19-8). In one TR, with eight ETL, we can fit in so many slices. If we use
the same TR, but now use an ETL of 16, it will take twice as long to
receive the echo (i.e., 16 × 17 msec = 272 msec) because now we have
to accumulate data from 16 echoes (each a multiple of 17 msec).
Because the time it takes to accumulate 16 lines of k-space is double
the time it takes to accumulate eight lines, we can fit in only half the
number of slices into one TR.
Therefore, the trade-off is that we decrease the number of slices as we
increase the ETL.
↑ETL→↑speed ↔↓coverage(↓number of slices)
One way to get around this is to increase TR. Although increasing TR
increases scan time, we save so much time by using a long ETL
compared with CSE that we can afford a longer TR. We don't need to be
limited any more to a TR of 3000 msec. We can go up to 4000 to 6000
msec, get more coverage, and still save time.
Let's say we need a coverage of fifteen 5-mm slices with a 2-mm gap to
cover an area of interest (like the brain).

Example 2
Let's choose TR = 3000 msec, ETL = 8, NEX = 1, Ny = 256.
The number of slices we can do in any TR depends on the length of the
longest echo (we'll leave out sampling time for now):
Number of slices ≤ TR/TE
In the case of an ETL of 8, the longest echo is 136 msec (17 × 8 = 136).
So,
Number of slices ≤ TR/TE = 3000 msec/136 msec [all equal to] 22
This formula is regardless of TEeff we pick. The scan time for this
example would be

= (3000 msec)(256)(1)/8
= 1.6 min
Therefore, we can do 22 slices in 1.6 min with an ETL of 8.

Example 3
Let's now choose a different ETL:
ETL = 16, NEX =1, Ny=256, TR = 3000 msec
With an ETL of 16 (since echoes are multiples of 17 msec), the longest
echo would be 272 msec (16 × 17 = 272). So,
Number of slices = TR/TE
= 3000 msec/272 msec [all equal to] 11

With an ETL of 16, we can do only 11 slices, but in 0.8 min. The scan is
faster, but we have limited coverage, and we have not accomplished
the minimum of 15 slices we needed for the exam.

Example 4
To increase the coverage with an ETL of 16, let's now increase the TR:
TR = 4500 msec, ETL = 16, Ny=256, NEX=1 Now,
Number of slices ≤ TR/TE
= 4500/272 msec [all equal to] 16

Even though we have limited coverage with an ETL of 16, we resorted

to increasing the TR to 4500 msec, which allowed us a coverage of 16
slices (enough to cover the 15 slices we needed for the exam), and we
were still able to keep the scan time below the scan time required for
an ETL of 8.
Sometimes you'll look at a scan and find that it took twice as long to do
the study as you thought it would. What can happen is that the
technologist will try to get a coverage that is too wide for the TR
chosen. If nothing is corrected, the machine will “default” to
performing two separate acquisitions to provide the coverage, thus
making the exam twice as long. With a simple calculation to allow
either increasing the TR slightly or increasing slice thickness slightly,
you could rectify the situation more efficiently.

Multi-echo FSE
Consider the case of eight echoes again. With CSE, we would have eight
k-spaces, and the first seven echoes would come “free.” That's not true
for FSE. Every single one of the echoes is used to fill in a line in a k-
space. If we want to do double-echo imaging, we have twice as many
lines in k-space to fill (in two k-spaces). Thus, we have to either (i) give
 up half of the echoes per ETL, in which case the scan time is going to
be doubled or (ii) repeat the scan twice, and again the scan time is
increased. Therefore, regardless of the way we do it, it's going to cost
us time.
In FSE, the first echo is no longer “free!”
There are three ways to get a double-echo image with FSE:
1. Full echo train
2. Split echo train
3. Shared echo
Full Echo Train. In a full echo train, all echoes in the train contribute
to the image. Thus, the full ETL is completed for effective TE1 (TE1eff)
before TE2eff is performed. In other words, two separate concatenated
sequences are acquired. For example, for an

ETL of 8 and a 256 × 256 matrix (Fig. 19-9), 32 echo trains would be
required to fill k-space (8 × 32 = 256).

Figure 19-9. In a full echo train, the entire echo train is completed
for TE1eff and TE2eff.
Split Echo Train. In a split echo train, the first half of the echo train
contributes to the image with TE1eff and the second half to TE2eff
(thus, two k-spaces are created). For example, for an ETL of 8 (Fig. 19-
10), only four echoes would be applied to each effective TE. Therefore,
64 trains would be required to fill k-space (4 × 64 = 256).
Shared Echo. In a shared echo approach, the first and last echoes in
the train are emphasized for TE1 and TE2, respectively, and the echoes
in between are shared for both images. This approach has the
advantage of shorter ETL compared with a full or split echo train
approach, allowing more slices to be acquired for a given TR (since the
number of slices is roughly determined by TR divided by the product of
ESP and ETL, i.e., number of slices [all equal to] TR/[ETL × ESP]).

Figure 19-10. In a split echo train, the first half of the echoes is
used for TE1eff and the second half for TE2eff.

In Figure 19-11, an ETL of 5 is used and four lines of k-space are filled
per TR, three of which are the same for the first and second echo
images. (Thus, there is some overlap of the “information” in the two
images, compared with the four split echo approach.) Filling four
echoes per pass provides the same efficiency as a split echo approach,
that is, 64 echo trains will
be required to fill k-space (4 × 64 = 256). However, the shorter ETL
allows 60% more slices to be acquired in the same time. Let's prove this
mathematically:

Figure 19-11. In a shared echo approach, the first and last echoes
are emphasized for TE1 and TE2, respectively, and the rest are
shared between the two echoes. In this example, the ETL is 5 and
the central three echoes are shared.

Number of slices (ETL = 5)/number of slices (ETL = 8)

= [TR/(5 × ESP)]/[TR/(8 × ESP)]
= 8/5 = 1.6 = 100% + 60%
 A variant of the shared echo approach is the socalled keyhole imaging.
In this technique, k-space is covered completely on the first image, but
only the central portion (e.g., 20%) of k-space is covered on subsequent
images, providing most of the contrast (recall that the center of k-
space contains most of the signal). This approach has a disadvantage in
that the high spatial frequency outer portion (e.g., 80%) of k-space is
shared information; however, it has the advantage of speeding up the
subsequent imaging by a factor of 5 (100%/20% = 5). Thus, keyhole
imaging is useful when fast repetitive imaging of the same slice is
required, for example, for perfusion imaging.
Each of the previous methods has its advantages and disadvantages.
The advantages of the full echo train approach are full flexibility in
selecting the TEeff and the ETL for both the first and the second
echoes; the disadvantage is more contrast averaging.
The disadvantage of the split echo approach is that the TE2eff is
constrained in the second half of the echo train. For example, for an
ETL of 16 and an ESP of 17 msec, the minimum TE2eff is 9 × 17 = 153
msec, which might be longer than desired. The advantage is less
contrast averaging and crisper images. In general, the split echo train is
used for ETLs of 8 or less, whereas the full echo train is used for ETLs
greater than 8.
The advantage of the shared echo approach is increased coverage
(number of slices). The disadvantage is an overlap of information
between the two echo images. (Fig. 19-12 gives an example.)

Advantages of FSE
1. The scan time is decreased (which allows faster scanning).
2. The signal-to-noise ratio (SNR) is maintained because we still have
256 phaseencoding steps.
3. The increased speed allows for highresolution imaging in a reasonable
amount of time. An example is 512 × 512 imaging through the
internal auditory canals with very long TR.
4. Motion artifacts will be less severe. Because the 180° pulses are
 evenly spaced, there is a natural even-echo rephasing effect. For
instance, cerebrospinal fluid (CSF) motion artifacts are much less
severe on FSE than on CSE images.
5. The rephasing from the multiple 180° pulses leads to less distortion
from metallic objects on FSE images (see Chapter 18; also see the
discussion below on magnetic susceptibility).

Figure 19-12. Sagittal proton density (A) and T2 (B), and axial
proton density (C) and T2 (D) images of the lumbosacral spine
 were acquired using a shared echo approach. Note that the CSF is
brighter than would be expected in a CSE proton density
acquisition. The fat is bright on the T2 images, which is also
typical of the FSE technique. This patient has minimal L5/S1
anterolisthesis that is secondary to bilateral dysplastic facets
(arrows in C and D).

6. Similarly, FSE images are much more tolerant of a poorly shimmed

magnet than are CSE images.

Disadvantages of FSE
1. Reduced coverage, that is, decreased number of slices.
2. Contrast averaging (k-space averaging) so that
a. CSF is brighter on proton density-weighted FSE images. This is
caused by the effect of averaging all the echoes into a single k-
space. We are still contributing data from very long TEs into

the proton density image, even though the weighting is toward the
lower TE. Therefore, we will have some T2 effect (i.e., bright CSF)
on a short TEeff. To alleviate this problem, either use a shorter ETL
(to exclude longer TEs) or a higher BW (to decrease ESP and the
minimum TEeff).
b. Pathology: MS plaques and other lesions at the brain/CSF
interface may be missed on FSE. CSF appears brighter on FSE
proton density images (as discussed above) and, therefore, the
distinction between CSF and periventricular high-intensity plaques
is more difficult. As in (a), to alleviate this problem, use a shorter
ETL, which helps exclude longer TEs in the echo (fluid-attenuated
inversion recovery [FLAIR] has generally solved this problem).
3. Magnetization transfer (MT or MTC) effect in FSE. MTC is
inadvertently present in FSE. This is caused by the presence of
multiple, rapid 180° pulses containing off-resonant frequencies.
When MT (discussed in more detail in Chapter 25) is intentionally
 produced, an RF pulse of 500 to 3000 Hz off the bulk water resonance
saturates protein-bound water in the broad peaks on either side of
the bulk phase water peak. Because the 180° pulses are rapid (in the
time domain), they have a broad BW (in the frequency domain), thus
containing frequencies off the bulk water resonance frequency. These
frequencies suppress proteinbound water like a fat saturation pulse
suppresses fat.
4. Normal intervertebral discs are not as bright on T2-weighted FSE
images compared with CSE. This is caused by the MT effects in FSE
previously discussed. They diminish the contrast between desiccated
(usually dark) and normal (usually bright) discs.
5. Magnetic susceptibility effects will be less than with CSE. This is
caused by decreased dephasing from closely spaced (refocusing) 180°
pulses that leave little time for spins to dephase as they diffuse
through regions of magnetic nonuniformity. In FSE, the signal loss is
minimized because of the rephasing effects of multiple 180° pulses.
Therefore, T2-weighted FSE images are less sensitive to magnetic
susceptibility effects such as metal or hemorrhage (e.g.,
deoxyhemoglobin and hemosiderin) than are T2-weighted CSE images
(Fig. 18-47).
6. Fat is bright on T2-weighted FSE images. This is due to suppression
of diffusionmediated susceptibility dephasing caused by the closely
spaced 180° pulses2 (Fig. 19-13). You could do a fat-saturated FSE to
decrease the intensity of fat.

Example 5
Performing a fat-saturated T2-weighted FSE sequence of the knee to
suppress fat in the marrow will allow bone marrow edema to stand out
against a dark marrow background. This technique increases the
sensitivity of detecting bone bruises (contusions) by about 30%3 (Fig.
19-14).

Other Features of FSE

More recent versions of FSE with high performance gradients allow
other options, such as higher BWs, flow compensation, and three-
dimensional (3D) FSE.
Higher BWs cause a reduction in the sampling time Ts and thus ESP,
allowing a lower minimum TEeff. Therefore, the previously hyperintense
CSF on supposedly proton density-weighted images can now be made
isointense to white matter. Use of a split (versus full) echo train also
minimizes unwanted T2 contributions by only averaging the first four
echoes of an eight-echo train. (This is the same principle that allows
T1-weighted FSEs to be performed, i.e., higher BWs and shorter echo
trains.)

Figure 19-13. Axial FSE T2 image of the brain (A) shows an

isointense to gray matter meningioma along the left parietal
convexity (arrow). Sagittal FSE T2 image of the pelvis (B) in a
different patient demonstrates marked focal thickening of the dark
junctional zone (arrows) along the fundus and midbody diagnostic
of adenomyosis. Note in both examples that the fat is bright.
 Figure 19-14. Sagittal FSE T2 (A) and fat-saturated FSE T2 (B)
images of the knee demonstrate marked increased conspicuity of
focal subchondral bone marrow edema of the distal femur in the
fat-saturated image.

Flow compensation is an added feature that is possible with high

performance gradients; such gradients allow a higher maximum
strength that can be applied over a shorter period. As a result, flow
compensation with high performance gradients does not place a large
time burden on the cycle.

3D FSE
With the advent of high performance gradients (see Chapter 30), 3D
FSE imaging in reasonable scan times is available. This imaging is
particularly useful for the brain, cervical spine, and lumbar spine where
bright CSF (T2- weighted) images are required in more than one plane.
The basic idea in 3D imaging (3D techniques in connection with
gradient-echo imaging is discussed in Chapter 20) is to have a
phaseencoding gradient not only in the y direction but also in the z
direction (Fig. 19-15). Therefore, the multiple slices in 2D FSE are
replaced by multiple slabs. Crusher gradients (see Chapter 14) are
applied before and after every 180° pulse (which is slice selective).
Each echo is first phase encoded, then sampled, and finally phase
unwound (i.e., a rewinder gradient is applied—see Chapter 21).
Consequently, the scan time will be T(3D FSE) = (TR × NEX × Ny ×
Nz)/ETL where Ny and Nz are the number of phaseencoding steps along
the y and z axes.

Figure 19-15. A PSD for a 3D FSE. Phase-encoding gradients are

applied along both y and z axes.

High performance gradients have higher maximum strengths and thus

allow a significant reduction in the gradient duration. This, in turn,
allows a larger ETL (decreasing scan time) and a reduction in minimum
TE (increasing the coverage). Therefore, despite a significant increase
in the acquisition time (compared with 2D FSE) caused by the addition
of the Nz factor into the previous formula, the scan can still be
performed in a reasonable time.

Example 6
 What is the scan time for a T2-weighted 3D FSE technique with TR =
3000 msec, ETL = 64, NEX = 1, Ny = 256, and Nz = 32?
Answer:
T = 3 sec × 1 × 256 × 32/64 = 384 sec = 6 min, 24 sec
which is very reasonable.

Advantages of 3D FSE
1. Higher SNR (compared with 2D FSE)
2. High (1-mm) isotropic resolution (Fig. 19-16)

Figure 19-16. Axial 3D 1.5-mm FSE T2 (A) of the internal

auditory canals shows a left intracanalicular mass consistent with
acoustic schwannoma (arrow). B: A CSE 5-mm T2 image is shown
for comparison. Also note that the subcutaneous fat is bright on
the FSE (A), whereas it is relatively dark on the CSE T2 (B).

3. Less partial volume averaging (due to thinner slices)

4. Capability to perform high-quality reformation in any plane (due to
ability to generate isotropic voxels)
5. Lower cross-talk (for slab-interleaved approaches)
6. Reduced magnetic susceptibility and field inhomogeneity artifacts
(compared with 3D gradient-echo techniques)

Fast IR
By adding a 180° inversion pulse prior to FSE, fast IR can be achieved.
For more details, refer to the discussion on fast FLAIR in Chapter 25.
Fast STIR is similar to fast FLAIR except that the TI is chosen to null fat
instead of fluid.

Key Points
FSE imaging provides almost all the advantages of
conventional spin-echo imaging at a faster speed.
The basic idea behind FSE is utilization of multiple
180° refocusing pulses, which allows filling multiple
lines in k-space for a single TR. The ETL is defined as
the number of 180° pulses and echoes in the train.
As a result, compared with CSE, the acquisition time
T in FSE is reduced by a factor of ETL:
T(FSE) = T(CSE)/ETL
The increased speed also allows for other features
not achievable by CSE techniques, such as high-
resolution imaging of, for example, the internal
auditory canals with very long TRs and, more
recently, 3D imaging using FSE.

Questions
19-1 Advantages of FSE include all of the following except
 (a) increased speed
(b) decreased ferromagnetic susceptibility artifacts
(c) decreased motion artifacts
(d) increased number of possible slices

19-2 Suppose TR = 4000 msec, TE = 100 msec, NEX = 2, Ny = 256.

(a) Calculate the scan time for CSE.
(b) Calculate the scan time for FSE with ETL = 8.

19-3 T/F Increasing ETL leads to increased speed and coverage.

19-4 The scan time for FSE is given by

(a) TR · NEX · Ny
(b) TR · Ny · ETL/NEX
(c) ETL/(TR · NEX · Ny)
(d) TR · NEX · Ny/ETL

19-5 Dual-echo imaging in FSE can be achieved by

(a) split echo train
(b) full echo train
(c) shared echo approach
(d) all of the above
(e) only (a) and (b)

19-6 Disadvantages of FSE compared with CSE include all of the

following except
(a) brighter CSF on proton density-weighted images
(b) fat is brighter
(c) increased magnetic susceptibility effects
 (d) normal intervertebral discs are not as bright

19-7 Calculate the scan time for a 3D FSE technique with TR = 4000
msec, TE = 100 msec, Nx = 128, Ny = 128, Nz = 32, NEX = 1, ETL =
64.

19-8 T/F Given comparable parameters, the scan time of an FSE

sequence is equal to that of a CSE divided by the ETL.

1Hennig J, Nauerth A, Friedburg H. RARE imaging: a fast imaging

method for clinical MR. Magn Reson Med. 1986;3:823-833.

2For more details, refer to Henkelman RM, Hardy PA, Bishop JE, et al.

Why fat is bright in RARE and fast spinecho imaging. J Magn Reson
Imaging. 1992;2:533-540.
3For more details, see Kapelov SR, Teresi LM, Bradley WG, et al. Bone

contusions of the knee: increased lesion detection with fast spin-echo

MR imaging with spectroscopic fat saturation. Radiology. 1993;189:901-
904.
20
Gradient Echo: Part I (Basic Principles)

Introduction
In this chapter, we introduce the gradient-echo (GRE) pulse sequence.
It is also called gradientrecalled echo (GRE), the reason for which will
become clear later in the chapter. The major purpose behind the GRE
technique is a significant reduction in the scan time. Toward this end,
small flip angles are employed, which, in turn, allow very short
repetition time (TR) values, thus decreasing the scan time.
Consequently, such techniques are also referred to as partial flip angle
techniques. One of the most important applications of GRE is the ability
to employ three-dimensional (3D) imaging, thanks to the higher speed
of GRE due to very short TRs. The major differences between GRE and
spin-echo (SE) sequences are explained in this chapter.

Gradient-Recalled Echo
As mentioned in the introduction, the purpose of the GRE technique is
to increase the speed of the scan. Recall that the scan time for
“conventional” techniques is given by

where TR is the repetition time, Ny the number of phase-encoding

steps, and NEX the number of excitations. Now, Ny is generally selected
to yield a certain resolution (too small an Ny degrades the resolution),
and NEX is chosen to yield a certain signal-to-noise ratio (SNR).
Consequently, the only parameter in Equation 20-1 that can be
controlled to reduce the scan time is TR.
In other words, we wish to select a TR that is as small as possible and
still be able to receive a reasonable echo to create an image. If we use
90° radio frequency (RF) pulses, then with very small TRs, the
longitudinal magnetization is not given sufficient time to recover to a
reasonable value, as depicted in Figure 20-1. This causes a significant
reduction in the longitudinal magnetization and the subsequent
transverse magnetization (Fig. 20-2). That is, the amplitude of the
received signal will be diminished significantly (and, thus, SNR is
deteriorated).
To rectify this problem, an RF pulse yielding a smaller flip angle α is
used instead of the usual 90° RF pulse. This change causes incomplete
flipping of the longitudinal magnetization into the x-y plane, producing
transverse magnetization, called, Mxy (Fig. 20-3). In addition, the major
component of magnetization remains along the z-axis after the RF pulse
(called Mz). Consequently, even with small TRs, there will be sufficient
longitudinal magnetization at the time of the next cycle.
MATH: For the mathematically oriented reader, the magnitude of the
transverse and longitudinal magnetizations are given by

right after the α RF pulse, where M0 is the initial magnetization.

After this point, the process of recovery along the z-axis and the
process of dephasing in the x-y plane are exactly the same as in its SE
counterpart. There is, however, one major difference between GRE and
SE sequences. In SE, a 180° refocusing pulse is used to eliminate
dephasing

caused by external magnetic field inhomogeneities. This pulse is not

used in GRE imaging.
Figure 20-1. After a 90° pulse, longitudinal recovery after a short
TR will be very small.

Figure 20-2. A: The initial longitudinal and transverse

magnetizations. B: After a short TR, the subsequent longitudinal
and transverse magnetizations will be smaller for a 90° RF pulse.
Figure 20-3. A-B: If a small flip angle α is used, only a portion of
longitudinal magnetization is flipped into the x-y plane, and thus a
portion of longitudinal magnetization will remain.

Figure 20-4. After a partial flip angle, longitudinal recovery will no

longer be small after a short TR (because partial flip retains a
portion of initial longitudinal magnetization). However, a 180° RF
refocusing pulse would be counterproductive.
Question: What is the reason for not using a 180° refocusing pulse in
GRE?
Answer: Because we use a small flip angle in GRE, there would be a
large component in the longitudinal axis at half the echo time (TE/2).
(In SE imaging, however, this longitudinal component is insignificant at
TE/2 because TE/2 is much smaller than T1 of the tissue.) Let's see
what would happen to this longitudinal magnetization if we were to
apply a 180° pulse. Although the application of a 180° pulse does yield
rephasing in the x-y plane, it also causes Mz to invert and point south
(Fig. 20-4). To recover this inverted vector back in the north direction
requires a long TR, which is clearly not the desired case in GRE. (The
above is not a problem in conventional SE imaging because the
longitudinal magnetization at TE/2 is very small and inverting it does
not pose any significant loss of signal.)

Figure 20-5. Instead of using a 180° pulse, a bilobed gradient is

used that has a negative lobe as well as a positive lobe of twice the
duration.

Question: In the absence of a 180° pulse, how does one form an echo?
Answer: One way is to measure the free induction decay (FID) instead.
However, it's impractical to do so because the FID comes on too early
and decays more rapidly than we can spatially encode the signal. We
need time to apply a phase-encoding gradient and prepare the signal
for frequency encoding. We also need time to let the RF pulses and
gradients die down before doing anything else. To this end, we will
intentionally dephase the FID and rephase (or recall) it at a more
convenient time, namely, at TE. This is accomplished via a refocusing
gradient in the x direction and is illustrated in Figure 20-5. This
gradient has an initial negative lobe that intentionally dephases the
spins in the transverse plane and thus eliminates the FID. It is then
followed by a positive lobe that rephases the spins, thus restoring the
FID in the form of a readable echo. The area under the negative lobe is
equal

to half the area under the positive lobe. The refocusing occurs at the
midpoint of the positive lobe. Figure 20-6 illustrates how the FID is
refocused at TE. In other words, the FID is first eliminated and then
recalled at time TE; hence the name gradient-recalled echo.

Figure 20-6. The bilobed gradient causes dephasing of the FID and
 its recalling at time TE (at the center of the positive lobe).
Because there is no 180° rephasing pulse, the rate of decay is given
by T2* (instead of T2).

Slice Excitation
The other difference between GRE and SE is that in GRE imaging, TR
may be too short to allow for processing of other slices. For this reason,
short TR GRE techniques may acquire only one slice at a time. This is
referred to as sequential scanning. (In the next chapter, we will
introduce variations of GRE that allow multiplanar imaging by
lengthening the TR.) In other words, in a sequential mode (single-slice)
GRE, the total scan time is given by

That is to say, acquisition of additional slices increases the scan time in

a sequential mode in GRE imaging.
In the 3D GRE technique (see below), the scan time is given by

where Ny and Nz refer to the number of phaseencoding steps in the y

and z directions, respectively.
Example
Suppose TR = 50 msec, TE = 15 msec, α = 15°, NEX = 1, and Ny =
128. Then TR × NEX × Ny × 50 × 128 × 1 = 6400 msec = 6.4 sec
Thus, it takes only 6.4 sec to obtain a single slice. If we were to
obtain 10 slices, the scan time would be Scan time (10 slices) = 6.4 ×
10 = 64 sec = 1 min, 4 sec
whereas acquisition of 20 slices would take Scan time (20 slices) = 6.4
× 20 = 128 sec = 2 min, 8 sec
which is twice as long.

Magnetic Susceptibility
The lack of a 180° refocusing pulse results in greater dephasing of spins
compared with conventional SE. This in turn results in greater
sensitivity to magnetic susceptibility effects. This increased sensitivity
can be both problematic (e.g., increased artifact at the air/tissue
interfaces) and advantageous (e.g., when searching for subtle
hemorrhage), depending on the application (Figs. 20-7 and 20-8).
Figure 20-7. Axial T1 gradient echo out of phase (TR 110/TE 2)
(A), and T1 gradient echo in phase (TR 110/TE 4) (B) images show
marked “blooming” artifact (arrows) from the patient's known
spinal fixation hardware. The out-of-phase image (A) has less
artifact than the in phase image (B) due to the shorter TE. Axial T2
FSE (C) and ½ NEX SSFSE T2
Figure 20-7. (continued) (D) images have less artifact due to
multiple 180° refocusing RF pulses with the SSFSE having the least
 due to the longest echo train length. Finally, an axial T2 FSE image
with fat saturation (E) shows ferromagnetic artifact (arrows) and
magnetic field changes resulting in paradoxical water saturation
(arrowheads) and very poor fat saturation.

Steady-State Transverse Magnetization

There is one more major difference between GRE and SE pulse
sequences. Whereas at the start of each cycle in SE imaging in which
there is negligible component of magnetization Mxy in the x-y plane,
this may not be the case in GRE imaging. In other words, GRE may have
residual transverse magnetization at the end of the cycle, which will
be affected by the next RF pulse. This is because in GRE imaging, TR
may be too short to allow for complete dephasing (i.e., T2* decay) of
the spins in the transverse plane. (In contrast, in SE imaging, TR is long
enough to allow complete dephasing of the spins in the x-y plane.)
After a few cycles, this residual transverse magnetization reaches a
steady state, referred to as Mss. This scheme is illustrated in Figure 20-
9 and is elaborated upon further in the next chapter.

Tissue Contrast
Figure 20-10 depicts a generic GRE pulse sequence diagram (PSD).
There are three operatorcontrolled parameters that affect the tissue
contrast: α, TR, and TE. Let's discuss how each of these plays a role in
this matter.
First, consider a small flip angle α (e.g., 5° to 30°). As can be seen in
Figure 20-11, a small flip angle results in a large amount of (persistent)
longitudinal magnetization after application of the RF pulse. This result
implies that complete recovery of the longitudinal magnetization to its
initial value will take much less time than following the 90° RF pulse of
an SE sequence. Consequently, given two tissues with different

T1 values, a large difference will not exist between their respective T1

curves, and thus T1 recovery plays a minor role. This is better
illustrated in Figure 20-12, in which tissue A's and tissue B's T1 curves
are drawn for two different flip angles, 10° and 90°. The T1 curves for
10° α start higher on the z-axis owing to the fact that they are only
partially flipped into the x-y plane. It can be inferred from this figure
that the T1 differences between the two tissues is much less apparent
for the smaller flip angle.
Figure 20-8. Coronal T2* gradient-echo image of the brain shows
multiple dark foci in the brainstem from the patient's known
multiple cavernous angiomas. This sequence is very sensitive for
subtle blood products.

Figure 20-9. A-D: Because TR is short, a fraction of transverse

magnetization remains at the end of the cycle, which eventually
reaches a steady state, Mss. This steady-state component is
affected by the next RF pulse.
 Figure 20-10. A gradient-echo (GRE) PSD.

A small flip angle reduces T1 weighting.

A small α also implies a small transverse magnetization and thus a small

steady-state component, reducing T2* weighting. As shown in Equation
20-2, Mxy is directly proportional to M0; thus, tissue contrast is
predominantly affected by proton density. This relationship is

illustrated in Figure 20-13. In this diagram, tissue A (e.g., H2O) has a

higher proton density than tissue B (e.g., fat), that is N(A) > N(B). At
time TE, then, the difference between A and B is predominantly
explained by their respective proton densities N(A) and N(B).
Figure 20-11. A small flip angle results in a large amount of
longitudinal magnetization.
 Figure 20-12. When the flip angle is small, it is difficult to
discriminate the T1 contrast between two tissues. Thus, small α
reduces T1 weighting.

A small flip angle yields proton density weighting.

Conversely, a large α (e.g., 75° to 90°) allows better differentiation of

the T1 characteristics of the two tissues. This should be fairly obvious
because in the extreme case when α is 90°, we are in a situation similar
to SE, given that TR is not very small. However, if TR is very small,
there won't be sufficient time for T1 recovery, thereby reducing T1
weighting (in this case, the large steadystate component that
accumulates due to the large α increases T2* weighting—see later text).
 Figure 20-13. Small flip angles also result in small transverse
magnetization, and thus poor T2* weighting and enhanced proton
density weighting.

A large flip angle (with a large TR) yields more T1

weighting.

For an intermediate α (e.g., 30° to 60°), the result is mixed contrast,

although the gain in T1 weighting caused by larger flip angles tends to
outweigh the gain in T2* weighting from increased steady state.
Next, consider the TR factor. If TR is very short (say a few
milliseconds), then there won't be enough time for complete decay of
transverse magnetization before the next α pulse. The residual
transverse magnetization (e-TR/T2*) will essentially contribute to the
next signal (at low flip angle). Thus, a short TR enhances T2* weighting.
More simplistically, for TR < 3T2* (say about 100 msec), contrast
depends on T2*.
 A short TR (with a small flip angle) increases T2*
weighting.

Now, consider a long TR. Remember that “long” is a relative term; in

GRE, a “long” TR is approximately several hundred milliseconds, which
is commensurate with a “short” TR in SE! A long TR allows more
recovery of the T1 curves and thus better differentiation of different T1
values. It also causes more T2* decay, reducing the steady-state
component and thus reducing T2* weighting.

A longer TR enhances T1 weighting.

Finally, let's discuss the TE factor. This parameter's role in GRE is similar
to the SE technique. That is, a short TE reduces T2* weighting and
enhances T1 or proton density weighting.

A short TE reduces T2* weighting and increases

proton density (PD) or T1 weighting. A long TE
enhances T2* weighting.

You might wonder what happens when these parameters are

conflicting. For example, what contrast would a large TR and a small
flip angle produce? The answer is that ultimately one parameter is
going to dominate. In this example, as described previously, a small α
predominates and produces a PD-weighted image. How about a short TR
and a large α? This combination results in a mixed contrast that is
proportional to the T2/T1 ratio of the tissues. (This is the contrast
produced by a technique called steady-state free precession (SSFP); see
Chapter 21.) Figures 20-14 through 20-16 contain some examples of GRE
imaging.
Figure 20-14. Axial T1 gradient-echo image of the abdomen shows
typical signal appearance of the abdomen similar to CSE T1
technique. There is a focus of bright T1 signal in the left kidney
that represented a hemorrhagic cyst in this patient with von
Hippel-Lindau syndrome.
Figure 20-15. Axial T2* gradient-echo (TR 684/TE 20 msec) image
of the knee shows high signal in the posterior horn of the medial
meniscus diagnostic of meniscal tear (arrow). Typical T2 signal is
seen. Incidental typical chemical shift artifact is seen in the
transverse (frequency-encoding) direction (arrowheads).
MATH: Earlier in this book, we introduced a mathematical formula
expressing the signal intensity (SI) with respect to TR, TE, T1, T2, and
N(H), namely

Now, with the addition of the flip angle α, this equation is modified as
follows:

Figure 20-16. Axial T2* 3D gradient-echo images from the same

patient on the same magnet 1 year apart show the second image to
have more T2* weighting. The flip angle was constant (5°);
however, image A had a TR 27/TE 6.9 msec, and image B had a TR
35/TE 16 msec. This demonstrates that the longer TE effect
predominated versus the effects of a longer TR. Incidental central
disc protrusion is seen in both.
Equations 20-4 and 20-5 differ in several ways. The most obvious
difference is the term in brackets containing the variable α. Also, note
that T2 is replaced by T2*, as is expected in GRE imaging (there is no
180° refocusing pulse).
Equation 20-5 is interesting in the extreme cases, namely, at α = 0° and
α = 90°. When α = 90° (which is the case in SE), then sin α = 1 and cos α
= 0, and the equation becomes

which is the same as Equation 20-4 with T2 replaced by T2*, as

expected.
When α ≅ 0 (i.e., very small) then cos α ≅ 1 and sin α ≅ α, and Equation
20-5 becomes

which is dependent on the proton density N(H) as well as T2*.

Signal-to-Noise Ratio
The SNR in a GRE technique is decreased per echo (compared with SE)
due to shorter TRs in GRE; however, more echoes are obtained in GRE
per unit time, somewhat compensating for the former effect.

Chemical Shift Artifact of the Second Kind (Dixon

Effect)
This phenomenon was discussed in Chapter 18 (Artifacts in MRI) and
applies only to GRE techniques. As discussed there, fat and water
protons precess at slightly different frequencies in the transverse plane
(220 Hz at 1.5 T). Immediately after the RF pulse, fat and water are
both tipped (partially) into the transverse plane and are in phase. At
various TEs later, fat and water spins will be in phase or out of phase.
At 1.5 T, fat and water get back into phase every 4.5 msec after the RF
pulse. This number is derived by the following formula:
Frequency difference
between fat and water = 220 Hz
Period = 1/frequency = 1/(220 Hz) = 0.0045 sec = 4.5 msec
Thus, fat and water are in phase initially at TE = 0, go out of phase at
TE = 2.25 and back in phase at TE = 4.5, etc. In general, at 1.5 T, fat
and H2O go in and out of phase every 2.25 msec after the RF pulse
(refer to Fig. 18-16). This is referred to as chemical shift of the second
kind.
 Figure 20-17. Axial postgadolinium T1 gradient-echo images
obtained in phase (TR 93/TE 4.5 msec) (A) and out of phase (TR
93/TE 2.2 msec) (B) demonstrate the “boundary effect” at the
interface of fat and water protons (arrows in B). Also note subtle
typical chemical shift in the in phase image (arrowheads) along the
transverse (frequency-encoding) direction.

If the selected TE is 2.25, 6.75, 11.25, 15.75 msec, etc., a dark

boundary is seen around organs that are surrounded by fat (such as the
kidneys and muscles). This is referred to as a boundary effect, which is
not observed in SE imaging because of the presence of refocusing 180°
pulses. This type of imaging is referred to as out-of-phase scanning,
referring to the fact that at these TEs, fat and water spins will be 180°
out of phase and cancel each other out. It should be remembered that
GRE especially at lower bandwidths will also demonstrate standard
chemical shift. Figure 20-17 is an example of both types of chemical
shift.
To overcome the boundary effect, we can do the following:
1. Make fat and H2O in phase by picking appropriate TE.
2. Suppress fat.
3. Decrease pixel size.
The Dixon method1 takes advantage of this phenomenon to create
solely “water excitation” or “fat excitation” images. To understand how
this is done, suppose that the signal intensities from fat and water are
designated F and W, respectively. Then, the signal intensities of the
image with fat and water in phase (Iip) and out of phase (Iop) are given
by Iip = W + F Iop = W − F

Thus, W = (Iip + Iop)/2 F = (Iip − Iop)/2 allowing only water (W) or fat
(F) imaging. This is the method of Dixon, which is particularly useful for
spectroscopic fat saturation at low field strengths.
3D GRE Volume Imaging
3D imaging with contiguous thin slices is feasible by using GRE
techniques. This type of imaging is accomplished by an addition of a
phase-encoding step (Nz) in the slice-select direction (z-axis).
Consequently, the total scan time is now

where Nz is the number of phase-encoding steps in the z direction. Nz is

usually a power of 2 (namely, 32, 64, and 128), but because the
addition of a phase-encoding step in the z direction may cause
wraparound artifacts in this direction as well, a few of the slices at
each end are discarded and the number of slices displayed is slightly
less (e.g., 28, 60, and 120). This provides a slab of slices.
You may wonder how it is possible to achieve 3D imaging in a
reasonable time given the large multiplicative factor Nz. The answer
lies in the extremely short TR.

Figure 20-18. A PSD for a 3D GRE. Phase-encoding gradients are

 applied along both y and z directions. The slice-select gradient is
also replaced by a slab-select gradient.

Example
Calculate the scan time for a 3D GRE technique in the C-spine with
the following parameters: TR 30/TE 13/α5°/NEX 2/256 × 192/Nz 64
Scan time = (TR)(NEX)(Ny)(Nz) = (30)(2)(192)(64) = 737280 msec =
737.28 sec = 12 min, 17 sec A typical PSD is shown in Figure 20-18. The
major difference here is the addition of an RF pulse for selection of a
slab (i.e., a slab-select gradient), as well as application of phase-
encoding steps in the z direction (i.e., slice encoding). This is depicted
in Figure 20-19.
Volume imaging can be performed using isotropic (cubic, with δx = δy =
δz) voxels or anisotropic (noncubic) voxels. The advantage of the
former is the ability to perform high-quality reformation in any plane of
choice. 3D imaging is now also possible using newer FSE techniques that
employ high performance gradients. More on this is in a later chapter.

Advantages of 3D GRE
1. Rapid volume imaging of thin contiguous slices without cross-talk
(Figs. 20-20 and 20-21)
2. Reformation capabilities (especially if isotropic)
 Figure 20-19. The three gradients in 3D GRE are illustrated here.

3. Increased SNR, because

Advantages of GRE
1. Increased speed
2. Increased sensitivity to magnetic susceptibility effects of hemorrhage
(allowing better detection compared with SE)

Figure 20-20. Axial T2* gradient-echo image of the cervical spine

acquired with 3D technique shows abnormal bright signal in the
right lateral aspect of the cord from multiple sclerosis (arrow).
 Figure 20-21. Axial T1 gradient-echo image of the abdomen
acquired with 3D technique shows normal enhancement of the
visualized organs. It was performed with chemical (spectral) fat
saturation (see Chapter 23 for details).

3. 3D imaging (e.g., in the cervical spine) in a reasonable time

4. Imaging of flowing blood (i.e., MR angiography)

Disadvantages of GRE
1. Decreased SNR caused by (a) small α, reducing the transverse
magnetization, and (b) very short TR, not allowing sufficient recovery
of the longitudinal magnetization.
2. Increased magnetic susceptibility artifacts (caused by lack of a 180°
 refocusing pulse), most noticeable at air-tissue interfaces such as in
the region of the paranasal sinuses or the abdomen (Fig. 18-43).
3. T2* decay because there are no 180° rephasing pulses. This results in
increased sensitivity to magnetic field inhomogeneities, intravoxel
dephasing, and magnetic susceptibility artifacts (Fig. 18-41).
4. Compared with FID imaging, the GRE technique (which is really an
FID-recalled technique) uses a longer TE, thus reducing the SNR
caused by increased T2* decay.
5. Introduction of chemical shift effects of the second kind, resulting in
a dark band around organs with water-fat interfaces, such as the
kidneys, liver, spleen, etc.

Key Points
1 The objective of GRE techniques is to reduce the
scan time.
2 Scan time is proportional to TR; a very small TR is
possible in GRE imaging.
3 Because a small TR does not allow reasonable
recovery of the longitudinal magnetization (thus
significantly diminishing SNR), a partial flip angle α
(<90°) should be used.
4 Because TR may be too short to allow complete
dephasing of the spins in the transverse plane, a
residual transverse magnetization may remain before
the next cycle.
5 A refocusing gradient (readout direction) is
employed that eliminates the original FID and later
recalls it at the echo time TE (hence the name
gradient-recalled echo, or GRE).
6 Because TR may be too short to acquire other
 slices within one TR period, GRE techniques often
perform one slice at a time. (In the next chapter, we
will discuss multiplanar GRE techniques.)
7 As a result, the scan time is also proportional to
the number of slices obtained, that is, scan time =
TR × Ny × NEX × (number of slices)
8 The tissue contrast is a function of the flip angle α,
the repetition time TR, and the echo time TE. A
simplified way of presenting the results is
summarized in Table 20-1.

Table 20-1

Small Large

α ↑ PDW ↑ TI W

TR ↑ T2 *W ↑ T1W

TE ↑ PDW ↑ T2 *W

Questions
20-1 T/F Regarding chemical shift of the second kind, at 1.5 T, fat
and water protons get out of phase at TE = 2.2, 6.7 msec, etc.

20-2 T/F A smaller flip angle increases PD weighting and reduces T1

weighting.

20-3 T/F GRE techniques often acquire one slice at a time.

20-4 T/F In general, GRE techniques use a partial flip angle because
very short TRs are used to reduce the scan time.

20-5 T/F In the absence of a 180° pulse, a bilobed refocusing

gradient is used in GRE.

20-6 T/F The longer the TR, the more T2* weighting will be
achieved.

20-7 Calculate the scan time for a GRE with TR = 30 msec, NEX = 2,
Ny = 256, for (a) a single slice and (b) 15 slices.

20-8 T/F Magnetic susceptibility is less with GRE than with CSE.

20-9 T/F The reason a 180° pulse is not used in GRE is to reduce the
scan time.

20-10 The SNR in 3D GRE is equal to the SNR in 2D GRE times the
factor:
(a) Nz
(b)
(c) Ny

(d)

20-11 T/F GRE sequences use a partial flip angle because the TR is
too small to allow adequate recovery of the longitudinal
magnetization to allow sufficient SNR.

1From Dixon WT. Simple proton spectroscopic imaging. Radiology.

1984;153:189-194.
21
Gradient Echo: Part II (Fast Scanning
Techniques)

Introduction
In the last chapter, the technique of gradientecho imaging was
introduced. In this chapter, several gradient-echo techniques will be
discussed, including GRASS (gradient-recalled acquisition in the steady
state)/FISP (fast imaging with steady-state precession), SPGR (spoiled
GRASS)/FLASH (fast low-angle shot), and SSFP (steady-state free
precession)/PSIF (opposite of FISP). Although every manufacturer uses
a different acronym, the underlying concepts are the same. We will
also discuss a multiplanar (MP) variant of these GRE techniques (e.g.,
MPGR-MP FISP, MPSPGR/MP FLASH). Finally, faster versions of these
techniques are introduced (e.g., Fast GRASS [FGR]/Turbo FISP, Fast
SPGR [FSPGR]/Turbo FLASH) as well as their MP versions (e.g.,
FMPGR/Fast MP FISP and FMPSPGR/Turbo MP FLASH).

Nomenclature
Table 21-1 contains a summary of the important acronyms used by three
major manufacturers: General Electric (GE), Siemens, and Philips. Also
refer to the list of abbreviations in the Glossary. As an example, GE
uses the prefix “Fast” and Siemens uses “Turbo” to denote similar fast
scanning techniques, be it gradient-recalled echo or spin echo. (The
acronyms are spelled out in the Glossary.)

GRASS/FISP
It was mentioned in the last chapter that, in contrast to SE, in GRE
there may be residual transverse magnetization at the end of each
cycle remaining for the next cycle. This residual magnetization reaches
a steady-state value after a few cycles and is denoted Mss.
This residual, steady-state magnetization is added to the transverse
magnetization created by the next α radio frequency (RF) pulse and
thus increases the length of the vector in the x-y plane (Fig. 21-1). This
then yields more T2* weighting. In other words, tissues with a longer T2
have a longer Mss than do tissues with shorter T2.
Actually, to preserve this steady-state component, an additional step
needs to be taken in the pulse sequence. A so-called rewinder gradient
is applied in the phase-encoding direction at the end of the cycle to
reverse the effects of the phase-encoding gradient applied at the
beginning of the cycle (i.e., it “unwinds” the former effect). In other
words, the rewinder gradient is nothing but the opposite of the phase-
encoding gradient (Fig. 21-2). For instance, if gradient +3 is applied for
phase encoding, the rewinding gradient would be −3.

SPGR/FLASH
The word “spoiling” refers to the elimination or “spoiling” of the
steady-state transverse

magnetization. There are various ways of accomplishing this task:

Table 21-1

GE Siemens Philips

GRASS FISP TFE

SPGR FLASH T1 FFE

SSFP PSIF T2 FFE

FSPGR Turbo FLASH T1 TFE

 Abbreviations: TFE, turbo field echo; FFE , fast field gradient
echo.

1. By applying RF spoiling
2. By applying variable gradient spoilers
3. By lengthening TR
RF Spoiling. RF spoiling is the method used in SPGR and is illustrated in
Figure 21-3. In this scheme, a phase offset is added to each successive
RF pulse. This causes a corresponding phase shift in successive Mss
vectors. By maintaining a constant phase relationship between the
transmitter and the receiver (achieved via a phase-locked circuit),
successive Mss vectors cancel each other out. A pulse sequence diagram
(PSD) for SPGR is shown in Figure 21-4. In this scheme, rewinder
gradients are naturally not used because their purpose is to preserve
the steady-state magnetization, thus defeating the purpose of spoiling.
An example is seen in Figure 21-5.
Variable Gradient Spoilers. Spoiling can also be achieved by using
gradient spoilers. This is accomplished by introducing an additional
gradient with variable strengths from cycle to cycle (Fig. 21-6).

Figure 21-1. A-D A residual transverse magnetization that reaches

a steady state Mss remains after a short TR.
Lengthening TR. The last method to achieve spoiling of Mss is by
lengthening TR. When TR is sufficiently large (generally over 200 msec),
there is enough time to allow complete dephasing of the spins in the
transverse plane (because TR » T2*). This is similar to the SE pulse
sequence.
Question: For a long TR (e.g., 500 msec), what is the difference
between GRASS (or FISP) and SPGR (or FLASH)?
Answer: There isn't any! A TR of 500 msec in GRASS effectively allows
transverse magnetization to decay away over each cycle, that is, it
eliminates the steady-state component (Mss). Consequently, GRASS/FISP
and SPGR/FLASH would have similar properties for any given TE and α
when TR is long.
SPGR/FLASH Tissue Contrast. By eliminating the steady-state
component, only the longitudinal component affects the signal in the
SPGR/FLASH technique. Thus, this technique lends itself to reduced T2*
weighting and increased T1 weighting. This is true provided α is also
relatively large. When α is small, however, the T1 recovery curves play
a minor role, and proton density (PD) weighting is increased.

In SPGR, a long TR and a large α yield T1

weighting. A large TR and a small α yield PD- or
T2*-weighted images depending on TE.
Figure 21-2. A PSD for GRASS/FISP. A “rewinder” gradient is
applied along the y-axis at the end of the cycle to reverse the
effect of phase-encoding gradient.

Figure 21-3. A-D Spoiling of the steady-state transverse

magnetization can be done via RF spoilers (as in SPGR), in which a
phase offset is added to each successive RF pulse.

Figure 21-4. A PSD for spoiled GRE by using RF spoilers (as in

SPGR).
Figure 21-5. Axial spoiled gradient-echo T1 images of the
 abdomen. A: Without contrast-showing periportal edema (arrow)
secondary to patient's hepatitis. B: Postgadolinium with chemical
(spectral) fat saturation in a different patient that shows a
multiloculated lesion in the medial segment of the left lobe
consistent with pyogenic abscess.

Figure 21-6. A PSD for spoiled GRE by using gradient spoilers.

Disadvantages of SPGR (FLASH)

1. Increased dephasing caused by inhomogeneities in B0
2. Increased magnetic susceptibility artifacts
3. Increased chemical shift artifacts (dark bands)

SSFP/PSIF
This technique is harder to comprehend. It yields heavily T2 (not T2*)-
weighted images. The PSD is shown in Figure 21-7. The idea here is that
each α RF pulse contains some 90° and some 180° pulses embedded in
it. Therefore, in Figure 21-7, α1 acts like a 90° excitation pulse and α2
like a 180° rephasing pulse. This yields a pulse sequence similar to spin
echo (SE) whereby an echo is formed at α3. Because it is difficult to
read a signal and transmit α3 at the same time, the echo is actually
recalled 9 msec prior to α3 by using an appropriate gradient. Note that
the echo corresponding to α1 is formed between α2 and α3.
Interestingly, in this scheme, TE is larger than TR (and smaller than 2TR
by 9 msec), which is somewhat counterintuitive. The rewinder
gradients are also shown in the diagram. The rewinder gradient is one
cycle ahead of the phase-encoding gradient due to the mechanism
described in the foregoing discussion. (In this technique, any two
successive RF pulses can be employed to create an SE.)

Figure 21-7. A PSD for SSFP/PSIF. Each α pulse contains some 180°
pulse embedded in it that acts like a refocusing pulse. This, in
turn, will result in a spin echo (SE) at the time of the next α pulse.
Hence, contrast is determined by T2 (not T2*).
 SSFP/PSIF Tissue Contrast. The SSFP sequence provides heavily T2 (not
T2*)-weighted images with increased scan speed without the use of
dedicated excitation and rephasing pulses.

Advantages of SSFP/PSIF
1. Decreased dephasing due to inhomogeneities in B0 compared with
GRASS and SPGR
2. Decreased magnetic susceptibility artifacts compared with GRASS and
SPGR
3. Decreased chemical shift artifacts (dark bands) compared with GRASS
and SPGR

Disadvantages of SSFP
1. Decreased signal-to-noise ratio (SNR) secondary to the use of longer
TEs (TE > TR)
2. Increased sensitivity to nonstationary tissue

Multiplanar Techniques
The GRASS and SPGR sequences can be performed using an MP
technique by selecting a long TR (several hundred milliseconds). These
are termed MPGR (multiplanar GRASS or multiplanar gradient
recalled)/MP FISP and MPSPGR (multiplanar SPGR)/MP FLASH. As
mentioned previously, this long TR causes spoiling of the steady-state
component in the transverse plane, thus making GRASS and SPGR
possess similar features.
We can still achieve both T1 and PD/T2* weighting depending on the
flip angle α, as discussed previously. To reiterate, a small α yields PD
weighting, whereas a large α yields T1 weighting. More specifically, at
small flip angles, MPGR and GRASS behave fairly similarly, whereas at
larger angles, MPGR tends to be more T1 weighted than GRASS because
 MPGR uses a long TR.

Advantages of Long TR
1. SNR is increased, because the longitudinal magnetization has more
time to recover completely.
2. MP scanning is feasible because a long TR allows acquisition of other
slices during the dead time within one TR period (similar to SE).
3. It is also possible to perform multiecho imaging (e.g., a short TE and
a long TE) similar to the multiecho, MP technique in SE. However, the
second echo tends to get degraded in GRE secondary to rapid T2*
decay.
4. A long TR, as discussed in Chapter 27, reduces saturation effects
(seen with very short TRs caused by incomplete recovery of the
longitudinal magnetization). Consequently, larger flip angles can be
used. Although larger flip angles cause more saturation effects, the
presence of a longer TR counterbalances it. Larger flip angles
obviously increase the length of the transverse magnetization and
thus the SNR.

Fast Gradient-Echo Techniques

This topic might at first appear puzzling. Perhaps you are saying that
you thought all GRE techniques were “fast.” It is true that GRE
techniques are generally faster than SE techniques, although the FSE
(fast spin echo)/TSE (turbo spin echo) technique may be equally fast.
However, there are additional methods that can further increase the
speed of scanning. These methods are called Fast GRASS/Turbo FISP,
Fast SPGR/Turbo FLASH, and so on.
The MP variants of these methods are also available and are referred to
as Fast MP GRASS/Turbo MP FISP, Fast MP spoiled GRASS/Turbo MP
FLASH, and so on. These render multiple slices with increased SNR in a
rapid time period.
How can you make an already fast GRE technique faster? The answer
lies in employing ultrashort TRs and TEs to reduce the sequence time,
 that is, the time it takes to excite, phase encode, and frequency
encode. This is achieved by the use of the following:
1. Fractional echo
2. Fractional RF
3. Fractional number of excitations (NEX)
4. Reduction in the sampling time Ts (vis-à-vis by increasing the
bandwidth [BW])
The first three items are discussed in Chapter 23 (Figs. 21-8, 21-9, and
21-10). Basically, by using a fraction of the echo and a fraction of the
RF pulse, we can in effect decrease the echo time TE. Increasing the
BW from ±16 kHz (i.e., BW = 32 kHz) to ±32 kHz (i.e., BW = 64 kHz)
results in a reduction of the sampling time Ts from 8 to 4 msec for 256
frequency-encoding

steps Nx (Fig. 21-11). The trade-off here is a reduction in SNR because

SNR is proportional to 1/ . A wider BW allows more noise through,
as shown in Figure 21-11. The active time is now given by Active time =
TE + Ts/2 + To where To is the “overhead” time and Ts is the total
sampling (readout) time. By minimizing TE and Ts, we can minimize the
active time and thus reduce the minimum TR. Employing a fractional
NEX decreases the overall scan time because the scan time is
proportional to NEX.
Figure 21-8. Fractional echo.
 Figure 21-9. Fractional RF.

In fast MP techniques, a longer TR is used, but multiple slices are

acquired within that TR period.

Figure 21-11. An increase in the BW results from a decrease in the

sampling time Ts, thus allowing a reduction in TE. The trade-off is
a decrease in SNR.
 Figure 21-10. Fractional NEX.

Example
1. Determine the scan time for obtaining 15 slices using Fast
SPGR/Turbo FLASH (one slice at a time) when TR = 10 msec, TE =
min, Ny = 256, NEX = 1: Time = (10)(256)(1)(15) = 38,400 msec = 38.4
sec
2. Determine the scan time for obtaining 15 slices using MP Fast
SPGR/Turbo FLASH (multislice) when TR = 100 msec, TE = min, Ny =
256, NEX = 1: Time = (100)(256)(1) = 25,600 msec = 25.6 sec

Applications. The technique is useful in applications requiring very fast

scanning such as the following:
1. Single breath-hold techniques in the abdomen
2. Imaging of a joint in motion (e.g., temporomandibular joints)
3. Cine imaging of the heart
4. Temporal scanning of the same slice after contrast administration
5. Perfusion images after bolus injection of contrast

Disadvantages
1. Decreased SNR and CNR caused by ultrashort TRs (less so in degree
with MP techniques)
2. Increased chemical shift artifacts of the second kind at very low TEs
(namely, at TE = 2.2 msec, 6.6 msec, etc.)

Tissue-Prepared Fast GRE Techniques

In fast GRE techniques, ultrashort TRs are employed, so the tissue
contrast may be suboptimal. To help improve the tissue contrast, a
technique called magnetization preparation or tissue preparation is
used (e.g., MP-RAGE). Before the α RF pulse (prep time), other RF
pulses (180° and/or 90°) are applied to the tissue. This prep time
allows the tissues to develop a certain contrast (T1 or T2 weighting,
depending on the type of application).
 Figure 21-12. In an IR (inversion recovery) preparation Fast GRASS
technique, a 180° pulse is applied before the GRE sequence,
allowing for better T1 discrimination of two different tissues. This
will increase T1 weighting.

Two types of tissue preparation methods are discussed:

1. IR prepared (inversion-recovery prepared)
2. DE prepared (driven-equilibrium prepared)
IR Prepared. First consider the IR prepared method. In this scheme
(Fig. 21-12), a 180° pulse is applied (prep time) before the α pulse.
This method is similar to an inversion-recovery (IR) technique, providing
increased T1 weighting and allowing suppression of various tissues
depending on the prep time (Table 21-2). Note that the table has prep
time values less than we have seen in typical IR sequences; this is
because there is less longitudinal magnetization in this partially
saturated technique, and hence the inversion time will be less (a
similar concept as seen with lower inversion times for lower field
strength systems versus longer times for higher
field strength). An example can be seen in Figure 21-13.

Figure 21-13. Short-axis IR prepared gradientecho sequence (TR

7/TE 3/TI 150 msec) of the heart 10 min postgadolinium
administration (delayed enhancement protocol) shows high signal
in the inferolateral wall (arrows) diagnostic of myocardial
infarction.
DE Prepared. The second method is the DE prepared technique. It
employs a 90°—180°—90° pulse sequence similar to SE to provide T2-
weighted contrast before the α pulse (Fig. 21-14). The longer the prep
time, the more T2 decay occurs and the more T2-weighted the tissue
contrast becomes. In other words, tissues with longer T2s are degraded
to a lesser degree compared with tissues with shorter T2s, thereby
enhancing T2 contrast.

Figure 21-14. In an SE (spin echo) preparation technique, a 90°—

180°—90° sequence (similar to SE) is applied before the GRE
sequence, allowing better T2 discrimination. Refer to the text for
more details.

Table 21-2

Suppressed Organ Prep Time (msec )

 Liver 200-400

Spleen 400-500

CSF 700-800

More specifically, referring to Figure 21-14, consider two tissues A and B

with different T2 values (tissue A having a longer T2 than tissue B).
After the first 90° pulse, both A and B will have similar transverse
magnetization vectors. Right before the 180° pulse, tissue B (having a
shorter T2) has decayed more than A and thus has a shorter vector in
the transverse plane. After the 180° pulses, their direction is reversed
by 180° in the transverse plane. The second 90° pulse causes these
vectors to be driven to the longitudinal axis (hence the name driven-
equilibrium), but with tissue A (having a longer T2) at a higher value
than B. This allows the GRE sequence to start out with a bias toward
tissues with longer T2s.

Flow Imaging
GRE scanning is generally performed one slice at a time (except in the
MP situation); therefore, each slice is an entry slice. Consequently,
flowrelated enhancement (FRE) applies to every single slice, and
vessels appear bright on GRE images.
The basic concept is that no saturated flowing protons enter the slice
so that flipping these protons yields maximum signal. This is the basic
concept behind 2D or 3D time of flight (TOF) MR angiography. This
topic is discussed at length in Chapters 26 and 27.

Key Points
1 Several GRE techniques are available:
GRASS/FISP, SPGR/FLASH, and SSFP/PSIF (Table 21-
1).
2 In GRASS, the residual transverse magnetization is
preserved via a rewinder gradient, contributing to
increased T2* weighting.
3 In SPGR/FLASH, the residual transverse
magnetization is “spoiled” by introducing phase shifts
in the successive RF pulses. This causes reduced T2*
and increased T1 weighting.
4 Spoiling can also be accomplished via gradient
spoilers or by lengthening the TR.
5 In SSFP/PSIF, heavily T2-weighted images are
obtained. GRASS/FISP and SPGR/FLASH represent
gradient-recalled FID sequences, whereas SSFP/PSIF
represents a gradient-recalled SE (spin-echo)
sequence. Interestingly, in this technique, TE is
larger than TR and usually is smaller than 2TR by 9
msec.
6 MP variants of the above are also available (e.g.,
MPGR/MP FISP and MPSPGR/MP FLASH) by
employing a longer TR (over 100 msec).
7 Fast GRE techniques (e.g., FGR/Turbo FISP and
FSPGR/Turbo FLASH) provide additional speed. This
is accomplished by using a fractional RF, fractional
echo, and fractional NEX, as well as a wider BW
(shorter sampling time Ts).
8 Combined fast and MP GRE techniques (e.g.,
FMPGR and FMPSPGR/Turbo MP FLASH) render
multiple slices with increased SNR in a fast mode.
9 The characteristics of GRASS/FISP, SPGR/FLASH,
and SSFP/PSIF are summarized in Table 21-3.

Table 21-3
 GRE
Technique SNR CNR Comments

Best
GRASS possible Preserves steady-
/FISP Highest T2* state component

Best
SPGR possible Spoils steady-
/FLASH Intermediate TI W state component

Gradient-
Provides recalled SE ; TR
SSPF/PSIF Lower T2W < TE < 2TR

Questions
21-1 Spoiling of the residual transverse magnetization can be
accomplished by
(a) gradient spoilers
(b) RF spoilers
(c) long TR
(d) all of the above
(e) only (a) and (b)

21-2 Fast GRE techniques may employ

(a) fractional echo
(b) fractional RF
 (c) fractional NEX
(d) narrow BW
(e) all of the above
(f) only (a)-(c)

21-3 T/F A rewinder gradient is applied in the phase-encode

direction to preserve the residual transverse magnetization Mss (as in
GRASS/FISP/TFE).

21-4 T/F Spoiled GRE techniques increase T1 weighting.

22
Echo Planar Imaging

Introduction
In the last three chapters, we discussed some of the fast scanning
techniques, namely, the fast spin echo (FSE) and gradient-recalled echo
(GRE) techniques along with their “fast” variants. In this chapter, we
will discuss the echo planar imaging (EPI) technique, which is the
fastest MRI technique available.

Basic Idea in EPI

Unlike other fast scanning techniques that can be achieved via software
updates, single-shot EPI requires hardware modifications. More
specifically, high performance gradients (discussed in Chapter 30) are
needed to allow rapid on-andoff switching of the gradients. The basic
idea is to fill k-space in one shot (single-shot EPI) with readout gradient
during one T2* decay (if longer, T2 blurring will occur) or in multiple
shots (multishot EPI) by using multiple excitations. As we shall see
shortly, single-shot EPI allows oscillating frequency-encoding gradient
pulses and complete k-space filling after a single RF pulse. Generally,
gradient strengths of over 20 mT/m and gradient rise times of less than
300 µsec are required. Furthermore, extremely fast computers are
needed to allow fast digital manipulations and signal processing.

Types of EPI
Two main types exist: single-shot EPI and multishot EPI. Earlier single-
shot EPI techniques used a constant phase-encode gradient. Newer
techniques use a “blipped” phase-encode gradient, referred to as
“blipped EPI.”1
Single-Shot EPI. In single-shot EPI, all the lines in k-space are filled by
multiple gradient reversals, producing multiple gradient echoes in a
single acquisition after a single radio frequency (RF) pulse, that is, in a
single measurement or “shot.”
To achieve this, the readout gradient must be reversed rapidly from
maximum positive to maximum negative Ny/2 times (e.g., 256/2 = 128
times) during a single T2* decay (e.g., 100 msec). Each lobe of the
readout gradient above or below the baseline corresponds to a separate
ky line in k-space. Therefore, the number of phase-encode steps Ny is
equal to the sum of the positive and negative lobes of the readout
gradient. The area under the Gx lobe determines the field of view (FOV)
—the larger the area, the smaller the FOV. Apparently, single-shot EPI
places tremendous demands on the gradients with respect to maximum
strength Gmax and minimum rise time tRmin (i.e., the maximum slew
rate Gmax/tR) as well as on the analogto- digital converter (ADC). In
general, ADCs with maximum bandwidths (BWs) on the order of MHz are
required rather than the kHz maximum BWs used for conventional spin
echo (CSE).
In earlier EPI methods, the phase-encode gradient was kept on
continuously (Fig. 22-1)

during the acquisition, resulting in the zigzag coverage of k-space (Fig.

22-2). This led to some artifacts during Fourier transformation
compared with conventional k-space trajectories. To rectify this
problem, the phase-encode gradient was subsequently applied briefly
during the time when the readout gradient was zero, that is, when the
k-space position was at either end of the kx axis (Fig. 22-3). This
method was referred to as blipped phase encoding, as its duration was
minimal (200 µsec). Therefore, the same phase-encode gradient is
applied briefly Ny number of times (e.g., 256 times). The technique was
called blipped EPI, and the k-space trajectory (Fig. 22-4) was much
easier on the Fourier transformation.
 Figure 22-1. A PSD for the original single-shot EPI technique in
which a constant phase-encode gradient is applied during readout.
The Gx gradient has a sinusoidal shape caused by rapid on-and-off
switching of the gradients.

As discussed previously, one of the problems with single-shot EPI is that

any phase error tends to propagate through the entire kspace. (This is
not a problem in CSE because of the presence of rewinder gradients,
which reset the phase at the end of each cycle.) The phase errors we
are talking about here are not the ones caused by motion (motion
artifact is not a problem in the ultrafast EPI) but the errors arising from
the variation in the resonating frequencies of protons (e.g., fat and
water protons), which lead to mismapping along the phase-encode axis.
Consequently, one of the technical problems of single-shot EPI is
magnetic susceptibility artifacts, particularly at air/tissue interfaces
around the paranasal sinuses. Also, because of this phase error
propagation along the phase-encode axis, chemical shift artifact in EPI
is along the phase-encode axis rather than the frequency-encode axis,
as in the case of CSE.
Figure 22-2. Single-shot EPI with constant phase gradient allows
traversal of k-space in a zigzag manner after a single RF pulse.
 Figure 22-3. A PSD for blipped EPI. The phase-encode gradient is
on briefly only while Gx is zero (blipped phase encoding). The
signal starts out lower and peaks where it does due to the initial
phase offset (Gy) (shaded area).

Multishot EPI. In multishot EPI, the readout is divided into multiple

“shots” or segments (Ns), so that Ny = Ns × ETL where ETL (echo train
length) is the number of lines in each segment. Because k-space is
segmented into multiple acquisitions, this technique is also called
segmental EPI.
Advantages of Multishot EPI (Compared with Single-Shot EPI)
1. This technique places less stress on the gradients compared with
single-shot EPI.
 Figure 22-4. The k-space trajectory for blipped EPI in an odd-
even manner, which is much easier on the Fourier transformation.

2. Phase errors have less time to build up compared with single-shot

EPI, thus reducing diamagnetic susceptibility artifacts.
Disadvantages of Multishot EPI (Compared with Single-Shot EPI)
1. Multishot EPI takes longer to perform than does single-shot EPI.
2. Because of this fact, multishot EPI is more susceptible to motion
 artifacts.

Pulse Sequence Diagram in EPI

Figure 22-1 illustrates an example of the original EPI pulse sequence
diagram (PSD). The major difference between this sequence and a
conventional sequence is the application of a series of sinusoidal-
shaped pulse sequences along the readout axis. This series requires
rapid on-andoff switching of the gradients to achieve a train of positive
and negative gradient pulses. It can be accomplished only with
advanced hardware that can support such high performance gradients.
The other difference is the application of a constant phase-encoding
gradient (remember that k-space is filled in one TR period). (The term
TR may not be suitable here because each slice is obtained after a
single RF pulse. However, TR could be defined as the time between
successive slice-select RF pulses.) The above scheme allows acquisition
of raw data for each slice after a single RF pulse (unlike routine SE, in
which one RF pulse is needed for each phase-encoding step).

Figure 22-3 provides a PSD for blipped EPI in which the phase-encode
gradient is applied briefly Ny number of times when the readout
gradient is zero.

k-Space Trajectory in EPI

Unlike CSE, in which data sampling is performed during a constant
readout gradient, in earlier single-shot EPI techniques, sampling was
carried out during the readout gradient's alternating positive and
negative lobes, which caused traversal of k-space in a zigzag or
sinusoidal manner (Fig. 22-2). In the case of blipped EPI, k-space
trajectory for even echoes is the opposite of that for odd echoes (Fig.
22-4). In multishot (segmental) EPI, data acquisition is carried out in
multiple segments in an interleaved manner (Fig. 22-5). Spiral imaging
in multishot EPI can also be formed (Fig. 22-6) by using two oscillating
gradients.
These trajectories are in contrast to CSE, in which each line of k-space
is filled during one TR period (and there are Ny number of such
periods).

Scan Time in EPI

Because the entire data acquisition for each slice is accomplished after
a single RF pulse, the scan time for a single slice in single-shot EPI is
limited by T2* or T2 decay (on the order of 100 msec).

Figure 22-5. The k-space trajectory for multishot EPI. k-Space is

divided into multiple segments, which are acquired in an
 interleaved manner.

Figure 22-6. A spiral trajectory for multishot EPI using two

oscillating gradients.

In general, if ESP (echo sampling period) is the time interval between

two successive echoes, Ny is the number of lines in k-space, and NEX is
the number of excitations (acquisitions), then for single-shot EPI, the
scan time is given by T (single-shot EPI) = ESP × Ny × NEX For multishot
EPI, the scan time is given by T (multishot EPI) = TR × Ns × NEX = TR ×
Ny × NEX/ETL which resembles the scan time for FSE (recall that Ny = Ns
× ETL).

Contrast in EPI
Contrast in EPI depends on the “root” pulsing sequence (which is
similar to preparation prepulses in GRE techniques). For instance, to
achieve SE contrast, a 90°-180° SE root sequence is applied before the
EPI module. Similarly, a partial flip RF pulse (<90°) before the EPI
module provides gradient-echo contrast. A 180°-90°-180° IR root prior
to EPI provides inversion recovery contrast. In addition, diffusion
gradients can be added for EPI-diffusion imaging.
 Figure 22-7. An SE-EPI PSD for the original EPI. A 90°-180° root
pulsating sequence is applied before the EPI module.

Therefore, we can summarize as follows:

1. SE-EPI (90°-180°-EPI) uses a 180° pulse to overcome external
magnetic field inhomogeneities (Fig. 22-7). This technique provides
T1 and T2 weighting. Because all the echoes in EPI are acquired with
the same value of the phase-encode gradient, contrast in SE-EPI is
determined from the temporal rephasing of the 180° RF pulse.
Contrast in a T2-weighted EPI is very similar to that of CSE, that is,
signal is determined by T2 decay and flowing blood is dark.
2. GRE-EPI (α°-EPI) does not employ the 180° pulse, thus providing T2*
 weighting. This technique is faster and best suited for, say, cardiac
cine imaging. Contrast here is determined by the time between the
negative phase-encode offset and the EPI readout module.
3. IR-EPI (180°-90°-180°-EPI) allows T1 contrast by applying a 180°
inversion prepulse.

Artifacts in EPI
N/2 Ghost Artifacts. Even with blipped EPI, phase errors may result
from the multiple positive and negative passes through k-space (i.e.,
alternating polarity of the readout gradient). “Ghost” artifacts of the
main image may appear along the phase axis, not caused by motion as
in CSE, but by eddy currents, imperfect gradients, field
nonuniformities, or a mismatch between the timing of the odd and even
echoes. Since the ghosts are derived from half the data (even or odd),
they are called N/2 ghosts. See Figure 22-8 for examples.
Remedy. Minimize eddy currents; proper tuning of the gradients.
Susceptibility Artifacts in EPI. Magnetic susceptibility effects in EPI
may result in variations in frequencies and phase errors. This effect is
reduced for multishot EPI because phase

errors have less time to build up. Advantages of multishot EPI over FSE
is that it has contrast much closer to CSE and it has greater sensitivity
to magnetic susceptibility effects such as hemorrhage compared with
FSE.
 Figure 22-8. Both (A) and (B) are b0 images from an EPI sequence
of the brain in different patients. (A) has a marked amount of “N
over 2” artifact, whereas (B) has more subtle findings (arrows).

Remedy. Minimized by proper shimming, TE shortening, or use of

multishot echo.
Chemical Shift Artifacts in EPI. Because of the presence of phase error
propagation along the phase-encode axis, chemical shift artifacts in EPI
occur along the phase-encode axis rather than along the frequency-
encode axis seen in CSE. These artifacts are much more pronounced
than with CSE, so an effective fat suppression technique is necessary.
Remedy. Apply fat suppression.

Functional Echo Planar Imaging

Diffusion Imaging. Diffusion is defined as the process of random
molecular thermal motion (also referred to as Brownian motion), which
plays an important role in, for example, cerebrovascular accidents
(CVAs). Diffusionweighted SE-EPI can be accomplished with a pair of
diffusion gradients applied before and after the 180° pulse (Fig. 22-9)
to dephase and eliminate signals caused by diffusing protons.
Diffusion Tensor Imaging. Diffusion tensor imaging (DTI) is an advanced
form of diffusion imaging that quantitates the anisotropy of white
matter. (Anisotropy is a property that is different along the three axes
in the image, e.g., diffusion.) Instead of a single gradient pulse, a
minimum of 6 (and sometimes 55) gradient pulses with a b value of
1000 is applied in addition to the b = 0 acquisition, that is, 7 total
acquisitions. Three of these acquisitions are the three that are normally
performed, that is, Dx, Dy, and Dz. In addition, three combination
gradient pulses are performed, corresponding to the off-diagonal
elements of the 3 × 3 tensor matrix (Fig. 22-10).
The theoretical advantage of DTI is that the measured ADC no longer
depends on the specific orientation of the x, y, and z gradients of the
scanner (true for a point, not a 2 × 2 × 5-mm voxel). The advantage of
DTI is that it

should allow “quantitative diffusion” measurements, which are not

dependent on head position in the magnet. In the DTI model, the x, y,
and z axes are modified so that the z-axis corresponds to the principal
direction of diffusion (typically along the major white matter tract in
the voxel). This direction is known as the principal eigenvector. The
diffusion coefficient in this direction is called the principal eigenvalue.
In addition to the principal eigenvector and eigenvalues, new
eigenvectors (new x and y axes) perpendicular to the new z-axis are
described (Fig. 22-11). Diffusion along the three eigenvectors is
described by eigenvalues D1, D2, and D3 (Fig. 22-12).
Figure 22-9. Diffusion-weighted SE-EPI. A pair of diffusion
gradients (shaded areas) is applied before and after the 180° pulse
to dephase and eliminate signals from diffusing protons.
 Figure 22-10. A diffusion tensor is a 3 × 3 matrix. By performing a
mathematical maneuver called a similarity transform, the
nondiagonal elements in the matrix are eliminated. This amounts
to reorienting the z-axis in the voxel so it points along the
direction of the principal white matter tract.

DTI images can be displayed as scalar maps of combinations of D1, D2,

and D3 variously called fractional anisotropy, relative anisotropy, and
anisotropy index (Fig. 22-13). Normal white matter has high diffusion
anisotropy, that is, diffusion is much greater parallel to than
perpendicular to the white matter. Abnormal white matter (due to
multiple sclerosis [MS], diffuse axonal injury, or gliosis) has reduced
anisotropy. In addition, DTI images can be displayed as principal
eigenvector maps, showing the orientation and integrity of white
matter in the image (Figs. 22-14 and 22-15).
Perfusion Imaging. Because contrast in GREEPI is T2* weighted, it is
perfect for performing first-pass perfusion sequences with gadolinium.
Flowing blood is bright, as with conventional GRE, allowing for EPI
magnetic resonance angiography.

Advantages of EPI
1. Scan time is approximately 100 msec or less.
2. Cardiac and respiratory motion won't pose problems any longer.
 Figure 22-11. Diffusion tensor showing reorientation of
coordinate system so new z-axis (z) is parallel to the main white
matter tract in the voxel.

3. Proton density-, T1-, and T2-weighted images free of motion artifacts

can be achieved.
4. It allows the study of organ function rather than the mere depiction
of organ anatomy.
5. Resolution can be improved because time is not a significant factor.

Disadvantages of EPI
1. Because fat-water chemical shift artifacts (of the second kind [Dixon
effect]) can be problematic with such short TEs, fat suppression with
presaturation techniques is always required for EPI.
2. Because of rapid on-and-off switching of the gradients, there is a
potential for generating an electric current or voltage in the patient,
thus causing peripheral nerve stimulation where the patient may
experience a crawling sensation in the skin also known as
“formication.” This shock is caused by the well-known fact in
electromagnetic theory that rapid changes in a magnetic field (i.e.,
 dB/dt for the mathematically oriented reader) induce an electric
current (E) in a conductor (in this case the patient).

Figure 22-12. A tensor characterizes diffusion in the brain not by

a single apparent diffusion coefficient but by three diffusion
eigenvalues (D1, D2, D3) describing diffusion along three
eigenvectors. The principal eigenvector is parallel to the main
white matter tract in the voxel, and diffusion in that direction is
characterized by D1.

3. Potential for phase error (caused by slight variations in resonant

frequencies) propagation exists. This effect is less for multishot EPI
as there is less time for phase errors to build up.
Figure 22-13. Fractional anisotropy map of normal brain. The
bright signal in the normal brain indicates the greater anisotropy.
 Figure 22-14. Tractography of corticospinal tracts based on plot
of diffusion eigenvectors. Lines show projection in plane, and
dots show projection perpendicular to plane in image. (Courtesy
of Ferenc Jolesz, MD, Boston, MA.)

Figure 22-15. Anisotropy map (A) demonstrates intact white

matter tracts subjacent to infiltrating glioma seen in
accompanying FLAIR image (B). (Courtesy of Shawn Ma, PhD,
Houston, Texas.)

4. Intrinsic nonuniformities in B0 and diamagnetic susceptibility effects

also result in variable resonance frequencies and increasing phase
errors. Again, this effect occurs less in multishot EPI.

Clinical Applications of EPI

1. Diffusion imaging of the brain (by looking at the molecular diffusion
of water). This is helpful in early diagnosis of acute CVAs, when
 routine images are unremarkable, and in distinguishing a CVA from
other processes (e.g., neoplasm) (Fig. 22-16).
2. Dynamic perfusion studies of the brain.
3. Cardiac and abdominal imaging free of motion artifacts in an
extremely fast mode (Fig. 22-17).
4. Imaging the coronary arteries free of cardiac motion artifacts.
5. Cine cardiac imaging within a single heart beat.
6. Dynamic perfusion study of the myocardium after intravenous
administration of gadolinium contrast to assess for ischemic regions
(Fig. 22-18).
7. Single breath-hold imaging of the abdomen, providing T1, T2, or PD
weighting.

Figure 22-16. Axial FLAIR (A) and DWI (diffusion-weighted image)

(B) show bright signal on FLAIR in the left occipital lobe with a
trace of increased signal in the left temporal lobe. The DWI image
confirms that the left occipital lobe signal is due to an old stroke,
while showing more clearly the significant new stroke in the
temporal lobe.

Figure 22-17. Axial EPI DWI image of the liver with a low b value
(b = 50 sec/mm2) demonstrates a bright T2 hemangioma (arrow).
Also note that the low b value provides excellent suppression of
signal in the liver vessels. This is a classic example of “T2 shine
through.”
Figure 22-18. Short-axis segmented GRE-EPI dynamic
postgadolinium adenosine stress study image shows decreased
signal in the inferolateral wall consistent with ischemia (normal
signal was seen on the resting study).

Key Points
Echo planar imaging (EPI) is the fastest MR
 technique that is widely available. It has become
particularly important in the imaging of acute CVAs.
It employs a train of oscillating frequency-encoding
gradients, thus rendering a sinusoidal k-space
traversal after a single RF pulse. Consequently, each
slice can be imaged in a matter of milliseconds (free
of any motion artifact), and the entire study can be
accomplished in a matter of seconds.

Questions
22-1 T/F Contrast in EPI depends on the root pulsing sequence.

22-2 T/F Phase errors are eliminated with blipped EPI.

22-3 Regarding multishot echo, all the following are true except
(a) It places less stress on the gradients compared with single-
shot EPI.
(b) Phase errors have less time to build up compared with single-
shot EPI.
(c) It takes longer to perform than singleshot EPI.
(d) It is less susceptible to motion artifacts.

22-4 T/F In blipped EPI, the phase-encode gradient is constant

during the acquisition.

22-5 T/F In multishot EPI, k-space is filled after a single excitation

pulse.

22-6 T/F In single-shot EPI, k-space is traversed in a zigzag fashion.

22-7 N/2 ghosts are generally seen with

(a) constant phase EPI
 (b) blipped EPI
(c) multishot EPI
(d) none of the above

1For more details, refer to Edelman RR et al. Echoplanar MR imaging.

Radiology. 1994;192:600-612.
 23
New Scanning Features

Introduction
In this chapter, we discuss some of the more recent techniques used by
newer MR scanners with more advanced software. The following is a
summary of these features and their functions:
1. To increase speed
a. Fractional number of excitations (NEX)
b. Fast spin echo (FSE)
c. Fast gradient-echo techniques
d. Parallel imaging (discussed in the next chapter)
2. To reduce TE
a. Fractional echo
b. Fractional radio frequency (RF)
3. To increase resolution (without time penalty)
a. Asymmetric field of view (FOV)
4. To reduce aliasing
a. No phase wrap
b. No frequency wrap
5. To increase coverage
a. Phase-offset RF pulses
6. To achieve contiguous slices
a. Contiguous slices
 b. 3D acquisition
7. To achieve saturation
a. Spatial saturation
b. Spectral (chemical) saturation
8. To increase signal-to-noise ratio (SNR)
a. Low bandwidth
9. Reduce motion
a. Periodically rotated overlapping parallel lines with enhanced
reconstruction (PROPELLER)
Remember that FSE and fast gradient-echo techniques are separate
pulse sequences, whereas the other features can be added to any pulse
sequence.

Fractional NEX
See Figure 23-1, and also refer to Chapter 13.
Mechanism
1. Only a portion of k-space is used (e.g., ½ NEX, ¾ NEX [actually, the
number of phase encodes Ny is reduced, not NEX]). Reconstruction is
based on inherent symmetry of k-space along the phase axis.
2. Slightly more than half of k-space is used (called overscan) for phase
correction.
3. The center of k-space is usually included because it contains the
strongest signals.
Advantages
1. Increased speed (caused by reduced Ny)

Disadvantages
1. Decreased signal-to-noise ratio (SNR)
2. May increase artifacts
Applications
1. Used for localizer (scout) images

Figure 23-1. Fractional NEX.

2. Used when speed is more important than SNR

3. Body imaging and magnetic resonance cholangiopancreatography
(MRCP)

Fast Spin Echo

For more detailed discussion, refer to Chapter 19.
Mechanism
1. Use of multiple 180° refocusing pulses
2. Filling the k-space with multiple lines per TR (per “shot”)
3. Echo train length (ETL) denotes the number of 180° pulses (e.g., 2,
4, 8, 16, etc.)
Advantages
1. Reduces scan time by a factor of ETL (2, 4, 8, 16, etc.)
2. Spin-echo contrast without reduction in SNR (SNR can actually be
increased by using a very large TR)
3. Decreased magnetic susceptibility—useful in imaging near metal
(particularly at higher bandwidth [BW])
Disadvantages
1. Reduced coverage (caused by the presence of long TEs).
2. Cerebrospinal fluid may be bright on proton density images. This is a
result of weighted averaging effect of later echoes. To reduce this
undesirable effect, use a shorter ETL (e.g., ETL = 4), shorter TE
(higher BW).
3. Fat is bright on T2-weighted images because of elimination of
diffusion-mediated dephasing caused by the closely spaced 180°
pulses as spins diffuse through regions of different magnetic field
strength (e.g., fat and water).
4. Magnetic susceptibility effects (e.g., hemorrhage) are reduced
because of decreased dephasing from closely spaced (refocusing)
180° pulses, which leave little time for spins to dephase as they
diffuse through regions of magnetic nonuniformity.
5. May increase heating as a result of rapid train of RF pulses.
Applications
1. Fast scanning
2. High resolution (e.g., internal auditory canals)
3. Increase SNR with reasonable acquisition time
4. Single breath-hold technique
5. Isotropic T2-weighted data set (e.g., 3D FSE)

Fractional Echo
See Figure 23-2.
 Mechanism
1. Only a fraction of the received echo is sampled (feasible because of
the symmetry of the echo about TE and symmetry of k-space along
frequency axis).
Advantages
1. TE can be reduced
2. SNR is improved in early echoes (less T2 decay)
3. Improves T1 weighting (reduces T2 effect)
4. May decrease flow artifacts and susceptibility effects

Figure 23-2. Fractional echo. Note outer dashed box is the full
echo while the inner dashed box represents the fractional echo.

Applications
1. T1-weighted images
2. To reduce flow artifacts and magnetic susceptibility effects
NOTE: Using a higher bandwidth can also result in a lower minimum TE,
however, at a SNR loss.

Fractional RF (90°, 180°, or Partial Flip)

See Figure 23-3.
Mechanism
1. Same principles as in fractional echo (a fraction of the RF pulse is
included in the pulse cycle because of the symmetry in the pulse).
2. TE can be reduced accordingly.
Features. The features are similar to those of fractional echo.

Figure 23-3. Fractional RF (90° or 180° or partial flip).

Asymmetric FOV
See Figure 23-4.
Mechanism
1. Rectangular FOV is used (FOV is typically reduced in phase direction
 because Ny is typically less than Nx).
2. May get square or rectangular pixels.
Advantages
1. With rectangular FOV, we can obtain resolution of, say, 512 × 512
matrix in the time it takes to perform a 512 × 256 acquisition.
2. Increases speed while maintaining resolution.
3. Useful when the anatomy being imaged is asymmetric (smaller) in
phase direction (e.g., the spine).
Disadvantages
1. Reduced SNR compared with full FOV
2. May cause wraparound in phase direction
 Figure 23-4. Asymmetric FOV. When a rectangular FOV is used,
FOV typically is reduced in the phase direction. By acquiring more
phase-encoding steps, a higher resolution can be achieved. In this
example, you can acquire 256 phase-encoding steps and still get a
512 resolution.

Applications
1. Spine
2. Extremities
3. Abdomen

No Phase Wrap (Phase Oversampling)

 See Figures 23-5, 18-7 and 18-8.
Mechanism
1. Doubles the number of phase-encoding steps and the FOV.
2. Discards one half of FOV at each end to maintain the specified FOV.
3. NEX is reduced by one half to keep time constant.
Advantages
1. Reduces or eliminates wraparound (aliasing)
2. No change in SNR or scan time
Disadvantages
1. If 1 NEX is used, scan time is doubled.
Applications
1. Small FOV (may cause wraparound)
2. Extremities

No Frequency Wrap (Frequency Oversampling)

Mechanism
1. The echo is oversampled, thus ensuring that the Nyquist frequency
requirement is satisfied (e.g., 512 samples are obtained by the
analog to digital convertor [ADC] even though Nx may read 256).
Remember that the Nyquist law requires that the sampling frequency
ωs be at least twice as high as the maximum frequency present in the
signal (ωmax). Stated differently, at least two samples are needed per

cycle corresponding to the highest frequency waveform.

 Figure 23-5. No phase wrap. To avoid wraparound in the phase-
encoding direction, you can select this feature to double the FOV
in that direction and automatically discard the unwanted portions
at the end. In this example, you can acquire 512 phaseencoding
steps and get a 256-step resolution.

2. Various low-pass filters (LPFs) and band-pass filters (BPFs) are also
 used to get rid of unwanted high frequencies in the signal.
Advantages
1. Avoids wraparound in the frequencyencoding direction
2. No increase in the scan time
Disadvantages
1. May decrease SNR because, by increasing the number of samples, the
sampling interval is reduced and thus the BW is increased (remember
BW = 1/δTs and SNR ∝ ).
2. This is done internally, that is, the operator has no control over it.
Applications
1. Almost always automatically turned on during routine scanning

Phase-Offset RF Pulses
See Figure 23-6.
Mechanism
1. Simultaneously excites two slices with two RF pulses that have a
phase offset
Advantages
1. Doubles the number of slices per TR (without increasing time)
2. Can achieve more T1 weighting by reducing TR
Disadvantages
1. May increase wraparound artifact
 Figure 23-6. Phase-offset RF pulses. Two RF pulses with different
phases excite two slices simultaneously.

Applications
1. When a short TR is desired, and a large area is studied.
2. Gadolinium-enhanced studies in which shorter TR enhances
paramagnetic effects of contrast.

Contiguous Slice Option

See Figure 23-7.
Mechanism
1. Employs longer RF pulses that have a more rectangle-shaped
frequency transform, reducing cross-talk.
2. This eliminates/minimizes the need for interslice gaps.
Advantages
1. Can achieve (almost) contiguous slices
 Figure 23-7. Contiguous slices can be achieved with an improved
RF profile that better approximates a rectangle.

2. Eliminates the need for interleaving (which doubles the scan time
when correctly performed with two acquisitions)
Disadvantages
1. Increases TE (longer RF pulses)
2. Fewer slices per TR
Applications
1. When gaps are not desirable

3D Acquisition
Mechanism
1. Gradient-echo (GRE) technique over a 3D volume.
2. 3D FSE T2 is an option.
3. Requires a phase-encoding gradient along the slice-select z-axis.
4. Contiguous slices (zero gaps).

Advantages
1. Increased SNR (due to acquisition from larger volume)
2. Allows thin, high-resolution contiguous slices and/or overlapping
slices
3. Allows for high-quality reformation in any plane
Disadvantages
1. May introduce wraparound artifacts at each end of the volume (end
of the slab) caused by the presence of phase encoding in the z
direction (Figs. 18-3 and 18-4).
2. Scan time now incorporates the number of phase-encoding steps
along the z-axis (e.g., 28, 60) into the formula as well. That is to say,
Acquisition time(3D GRE) = T(3D GRE) = TR · NEX · Ny · Nz = Nz · T(2D
GRE) but because TR is very short, this is acceptable. For 3D FSE, this
formula becomes Acquisition time(3D FSE) = T(3D FSE) = TR · NEX · Ny
· Nz/ETL = Nz · T(2D FSE) where ETL can be chosen to be very large.

Applications
1. C-spine
2. MR angiography (e.g., circle of Willis)
3. Dynamic gadolinium abdomen

Spatial Saturation Pulses (Sat Pulses)

See Figure 23-8.
Mechanism (also see Chapter 25)
1. Ninety-degree saturation pulses are applied on either side of selected
volume (can be applied in any direction: anterior/posterior,
superior/inferior, and right/left).
2. Sat bands are usually applied to suppress phase ghosts caused by
flow-related phenomena.
Advantages
1. Minimizes phase ghosts
2. Minimizes flow artifacts
 Disadvantages
1. May cause signal suppression in the remainder of the FOV
2. May lengthen TR, thus increasing the scan time
Applications
1. Imaging of the spine: Saturation band is placed anterior to the
vertebral bodies to suppress artifacts arising from the heart and great
vessels.
2. MR angiography: Saturation pulses are placed at one end of a vessel
to suppress either venous flow (to obtain MR arteriography) or
arterial flow (to obtain MR venography).
3. Abdominal imaging (minimizing artifacts from the aorta or inferior
vena cava).
4. Brain (minimizing artifacts from the internal carotid arteries and
dural sinuses).

Chemical (Spectral) Presaturation

See Figure 23-9.
Mechanism (also see Chapter 25)
1. A frequency-selective presaturation pulse is applied before the RF
excitation pulse, thus eliminating the longitudinal magnetization for
a specific tissue.
2. The presaturation pulse is applied immediately before the 90° pulse
of the SE. The tissue whose Larmor frequency matches the frequency
of the presat pulse first sees a 90° spectral presat pulse that flips it
into the x-y plane. A short time later, the longitudinal component Mz
has really not had time to recover. Thus, the subsequent 90°
excitation pulse that flips this minimal Mz into the x-y plane
generates a minimal signal.
Applications
1. To suppress fat or water by appropriate selection of the frequency
 Figure 23-8. A: Spatial saturation pulses are placed immediately
before or after each slice (outside the FOV) in the case of flow
imaging of arteries or veins. B: Spatial saturation pulses are placed
in front of the spine (within the FOV) to reduce artifacts from
perivertebral vessels. C: Sagittal T1 image of the spine
demonstrates an anterior saturation pulse over the heart resulting
in a band of dark signal (arrows) and suppression of artifacts from
the heart.

Figure 23-9. Chemical (spectral) presaturation pulses.

Advantages
1. It resolves tissues with similar T1 values (such as fat and gadolinium-
enhanced tumors).
2. This technique does not have any influence on the signal from tissues
other than the one being suppressed (in contrast, IR affects the
contrast of all tissues).
Disadvantages
1. Since this approach employs a frequencyselective technique, it
 suffers from sensitivity to magnetic field inhomogeneities (see Figs.
18-42, 18-43, and 20-7).
2. Requires extra time (thus lengthening TR and increasing the scan
time).
3. Increases the number of RF pulses, causing extra RF heating.

Low Bandwidth (Variable BW)

Mechanism
1. On double-echo SE sequence, the second echo has a lower BW
Advantages

1. Increases SNR (remember that SNR ∝ of second echo

Disadvantages
1. Increased chemical shift artifacts (see Fig. 18-14)
2. Decreases allowable number of slices (due to increased TE)
Applications
1. Any study requiring a higher SNR
2. Example: Dual SE of brain—a narrower BW is used on the second echo
(to increase SNR, which is decreased because of T2 decay)

Periodically Rotated Overlapping Parallel Lines

with Enhanced Reconstruction
Mechanism
1. Periodically rotated overlapping parallel lines with enhanced
reconstruction (PROPELLER) (GE) and BLADE (Siemens) cover k-space
in a radial (vs. Cartesian) fashion with several lines of k-space
included in each fast spin-echo blade. The centers of k-space can be
realigned if there is motion between acquisition of the blades.
2. By combining FSE-based PROPELLER with diffusion, the diamagnetic
 susceptibility artifacts over EPI-based diffusion can also be
eliminated.
Advantages
1. Reduces gross motion (Fig. 23-10)
2. Increases SNR
Disadvantages
1. Requires 50% to 60% more time to perform

Figure 23-10. A: Sagittal T1 of the brain in a 3-year-old patient

evaluated for seizures demonstrates multiple curvilinear lines from
motion artifact. B: Sagittal T1 with PROPELLER (BLADE) was then
performed to prevent this child from being sedated with resultant
elimination of motion artifact and a diagnostic image.

Key Points
Many new MR techniques are reviewed in this
chapter. These, along with their major trade-offs, are
 summarized below.
Advantage Feature Disadvantage

Increase speed Fractional NEX ↓ SNR , ↑ artifacts

FSE ↓ Coverage, ↑ contrast
Fast gradient-echo techniques averaging
Parallel imaging (discussed further ↓ SNR
in next chapter) ↓ SNR

Reduce TE Fractional echo Both: may ↑ artifacts

Fractional RF

Increase resolution (without Asymmetric FOV ↓ SNR and ↑ potential for

time penalty) wraparound

Reduce aliasing No phase wrap May slightly ↑ scan time

No frequency wrap May ↓ SNR

Increase coverage Phase-offset RF pulses May ↑ wraparound

Contiguous slices Contiguous slices (CS) May ↑ TE

3D acquisition May ↑ scan time

Saturation Spatial saturation Both: slightly ↑ scan time

Spectral (chemical) saturation Chem Sat: sensitivity to field
inhomogeneity

Increase SNR Low bandwidth (variable BW ) Increased chemical shift

artifacts
Slightly fewer slices

Reduce motion PROPELLER or BLADE ↑ Scan time by about 50%-60%

Questions
23-1 Speed can be increased by
(a) FSE or GRE techniques
(b) fractional NEX
 (c) reducing BW
(d) all of the above
(e) only (a) and (b)

23-2 An asymmetric FOV may result in all of the following except

(a) reduced SNR
(b) increased potential wraparound
(c) increased resolution
(d) increased scan time

23-3 A lower BW leads to all of the following except

(a) reduces minimum TE
(b) increased chemical shift artifacts
(c) decreases coverage
(d) increased SNR

23-4 T/F Spatial saturation pulses are used to minimize phase ghosts
and flow artifacts.

23-5 T/F The SNR is increased in 3D GRE.

23-6 T/F Minimum TE can be reduced by using fractional NEX.

24
Parallel Imaging

Introduction
Parallel imaging is an approach for reducing scan time and includes
techniques such as sensitivity encoding (SENSE) and generalized
autocalibrating partially parallel acquisition (GRAPPA). Vendor
implementations of these techniques are also known as array spatial
and sensitivity encoding technique (ASSET), autocalibrating
reconstruction for Cartesian sampling (ARC), integrated parallel
acquisition technique (iPAT), and rapid acquisition through parallel
imaging design (RAPID).

Vendor Name Technique Sensitivity Calibration

SENSE (Philips) SENSE Prescan

ASSET (GE ) SENSE Prescan

ARC (GE ) GRAPPA Auto

GRAPPA (Siemens) GRAPPA Auto

mSENSE (Siemens) SENSE Auto

RAPID (Hitachi) SENSE Auto

Parallel imaging requires the use of phasedarray coils. Depending on

how many coils are used, time savings (or acceleration factors) of
between 2 and 3 are easily achievable, albeit with a similar signal-to-
noise ratio (SNR) cost as accompanies other “fast imaging techniques”
based on decreasing phase steps or fewer excitations. As a result,
parallel imaging is most useful for high SNR situations, for example, CE-
MRA and 3T.

Concepts
The “parallel” in parallel imaging refers to the fact that each coil in
the phased array receives data at the same time (i.e., in parallel).
Recall from Chapter 2 that a phased array is made up of many small
coils, with 32 or more available on modern systems.
Parallel imaging works by knowing the local sensitivity of each coil in
the phased array. The field of view (FOV) is made intentionally too
small in the phase-encoding direction, and the resulting aliasing is
unwrapped using this information. The basic concepts were described in
the early 1990s, and the first successful clinical implementation was
simultaneous acquisition of spatial harmonics (SMASH). SMASH has now
been superseded by more sophisticated techniques and the main
approaches in clinical use today are SENSE and GRAPPA.
A parallel imaging acquisition with an acceleration factor of 2 acquires
data only for every other line of k-space. The corresponding images
have a ½ FOV and exhibit aliasing artifact. The SENSE and GRAPPA
approaches work either by unwrapping the aliasing in the images or,
equivalently, by filling in the missing lines of k-space (Fig. 24-1).
Both approaches make use of the different spatial sensitivities of the
coils in the phased array. Due to their distinct spatial locations, each
coil has different view of the object (Fig. 24-2). Knowing their spatial
locations and sensitivity profiles allows the aliasing to be undone by
postprocessing.
Figure 24-1. Image domain (top) and the associated k-space
(bottom), which indicates how aliasing in the image domain is
equivalent to undersampling in k-space. The SENSE method
unwraps the aliasing in the image, but the same result can be
obtained using GRAPPA to fill in lines of k-space.
 Figure 24-2. Aliased images obtained from an 8-coil phased array.
Note each image has the characteristic spatial sensitivity of the
coil, such that signals near the coil are strong and those far away
are weak. This basic information allows us to work out which
signals are in the right location and which signals have been aliased
from ½ FOV away.

Prescan or Autocalibration
There are two common ways to determine the spatial sensitivities of
the coils in the phased array. SENSE and GRAPPA differ in the way that
spatial sensitivity maps are determined and how the postprocessing is
done.
 Prescan. A large 3D volume acquisition lasting a few seconds run
immediately after the localizers. This gives low-resolution images of
the entire region inside the scanner and provides spatial sensitivity
maps for all subsequent scans.
Autocalibration. A few extra phase-encoding lines are acquired within
each parallel imaging scan. These are used in postprocessing to train
the algorithm how to generate missing k-space lines for that specific
scan only.

Figure 24-3. Description of the SENSE parallel imaging method.

Coil sensitivity maps (left) from two coils in the phased array and
aliased images (right). The equations express the aliased pixels as
a sum of two distinct spatial locations weighted by the coil
sensitivities.

Advantages/Disadvantages
1. Prescan only needs to run one time so there is little time spent on
 sensitivity measurements.
2. Autocalibration calibrates on every scan so it can tolerate patient
repositioning between scans.
3. Autocalibration costs a few extra phaseencodes so acceleration
factors will be slightly lower (e.g., 1.9 rather than 2). Note the SNR
decrease is proportionate (e.g., decrease rather than ).
SENSE. Sensitivity encoding generally uses a prescan to measure
sensitivity of the coils. The individual images are divided by a body coil
image to remove “contaminating” structure, leaving just the spatial
sensitivity of the coils (or sensitivity maps). These are shown in Figure
24-3; for simplicity, only two coils in the phased array are shown.
Since we used an acceleration factor of 2, we know the aliased pixels
come from two locations at a distance of exactly ½ an FOV apart. We
also know the pixels are weighted by the sensitivity of the coil, for
example, pixels close to the coil will have a higher weighting than
pixels further away. By knowing the aliasing pattern and the spatial
sensitivities, we can write a linear equation and solve it to obtain the
values of the two pixels (Fig. 24-3). The process is repeated for all
pixels.
Because the coil sensitivity maps and aliased images are represented in
the spatial domain, SENSE is often described as an image-domain
parallel imaging technique.
GRAPPA. GRAPPA uses autocalibration to provide information about the
coil sensitivities. For an acceleration factor of 2, alternate lines of k-
space are skipped but a few additional lines are acquired at the center
for calibration. The overall acceleration factor is slightly reduced
although the calibration data contribute to the SNR so the SNR is
proportionate to the overall acceleration factor. A schematic of the k-
space data is shown in Figure 24-4.
In GRAPPA, we seek to recreate the missing k-space lines using a
combination of its neighbors, often just two neighboring lines. Note the
SMASH technique mentioned above uses just one neighboring line.
The key is to work out how to generate missing lines, and this is where
the calibration data are needed. If the calibration data consist of (say)
line 2, then we need to find what linear combination of lines 1 and 3
can give the closest approximation to line 2. We must also specify a
coil, so the full procedure is to determine what

combination of lines 1 and 3 from all coils gives the best approximation
to line 2 from coil 1. Similarly, we must determine what combination
gives the best approximation to line 2 of coil 2.

Figure 24-4. Description of the GRAPPA parallel imaging method.

The undersampled k-space from two coils are shown (solid lines),
along with the calibration data (dotted lines). By a computational
search, we look for coefficients (C and D) that minimize the
difference between our estimate of the calibration lines,
generated by the equations, and the actual acquired calibration
line.

Once the coefficients are determined, the missing lines anywhere in k-

space can be generated from a combination of the neighboring lines.
The final result is a full k-space for each coil, which are then Fourier
transformed and combined into a single image using standard methods
(e.g., the square root sum of squares).
Because the coil calibration and data are represented in k-space,
GRAPPA is often described as a frequency-domain parallel imaging
technique.
SNR and g-factor. An image produced with half as many phase-encodes
has an SNR that is decreased by ; however, the SNR is often
significantly worse at higher acceleration factors. This is the result of
unfavorable coil geometries, or the “g-factor.” The overall loss of SNR
in a parallel imaging acquisition is the square root of the acceleration
factor multiplied by the g-factor. So ideally the g-factor should be close
to 1.
The g-factor is a numerical measure of how difficult it is to solve the
SENSE (or GRAPPA) equations. Benign g-factors (i.e., close to 1) occur
when the coils have very different views of the object along the phase-
encoding direction. Likewise, poor g-factors occur when coils have
virtually the same sensitivity. In terms of the SENSE equation, the latter
would occur when S11 = S21 and S12 = S22 so there is effectively only one
equation. In this case, it is impossible to find a solution and the gfactor
would be infinite. In practice, coils always have some similarities and
some differences. These vary for each pixel in the image and this gives
rise to the spatially varying SNR characteristics of parallel imaging.
Figure 24-5 shows images for SENSE and GRAPPA reconstructions. Some
points to note are
The overall SNR is in inverse proportion to the square root of the
acceleration factor.
GRAPPA acceleration factors are slightly lower due to auto
calibration.
Spatially varying SNR is observed, particularly, with higher
acceleration factors.
Generally, little difference is seen between SENSE and GRAPPA
images.
There are many ways to alter the SNR in parallel imaging by
postprocessing. Mathematically, it is straightforward to constrain the
SENSE and GRAPPA images to appear smooth, or to incorporate prior
knowledge about how the object should look. Although these may
increase the SNR, there is always a trade-off to be made! Often spatial
resolution may be compromised or subtle artifacts may be introduced
that are hard to understand.

Figure 24-5. 3T images produced by SENSE and GRAPPA at two

comparable acceleration factors, reconstructed from virtually the
same data. For most applications, accelerations of 2 perform well,
 and the SENSE and GRAPPA methods are interchangeable. At higher
accelerations, structured noise and artifacts start to become a
problem for both.

Key Points
We have looked at the concepts behind parallel
imaging, which uses phased-array coils to accelerate
scans, and taken a close look at two popular
methods, SENSE and GRAPPA. We saw that each coil
in a phased array produces an image with a
characteristic spatial sensitivity and that
postprocessing can undo the aliasing introduced by
reducing the FOV in the phase direction, for a
corresponding saving in scan time. We noted that
acceleration factor of 2 is readily obtainable on
current systems; this is likely to increase as special
purpose coils for high accelerations are developed.

Questions
24-1 What does the word “parallel” refer to in parallel imaging?
(a) The need to use fast computers for postprocessing
(b) The simultaneous data acquisition by the coils
(c) The blend of phase-encoding and phased-array coils
(d) The arrangement of the coils needed to make it work

24-2 With 8 coils and an acceleration factor of 2, what will be the

SNR change relative to a nonaccelerated scan?
(a) 1/2
(b)
 (c) 1/4
(d) No change

24-3 Which imaging situation is incompatible with parallel imaging

(a) Diffusion-weighted EPI
(b) 3T high field
(c) Large FOV body coil
(d) 128 channel phased array
(e) All of the above

24-4 T/F Higher acceleration factors can be obtained by using a more

powerful gradient system.

24-5 T/F Parallel imaging is most compatible with applications that

have inherently high SNR.
 25
Tissue Suppression Techniques

Introduction
One of the beauties of MR, especially with some of the new features, is
the ability to image a body part while “suppressing” the signal coming
from a certain selected tissue. This suppression allows perturbation of
tissue contrast to enhance the signal coming from tissues of greater
interest (such as a pathology). Two types of tissues are commonly
suppressed in clinical practice: fat and water.

Suppression Techniques
Several suppression techniques are available, some of which are listed
below.
1. Inversion recovery techniques
2. Chemical (or spectral) saturation or frequency-selective
presaturation
3. Spatial presaturation in the field of view (FOV)
Inversion Recovery (IR) Techniques. This technique was discussed at
length in Chapter 7. The IR pulse sequence diagram (PSD) is shown in
Figure 25-1. By appropriate selection of TI (time to inversion), we can
nullify or suppress a certain tissue. In fact, as we saw in Chapter 7, if TI
= (ln 2) [T1(tissue x)] = 0.693 T1(tissue x) then tissue x is nulled. Thus,
TI can be chosen to null fat or water or any other desired tissue
depending on the application (Fig. 25-2).
STIR (Short TI Inversion Recovery). This is the IR technique that is used
to suppress fat.
Example
What is the TI used in STIR? At 1.5 T, T1 of fat is approximately 200
msec. Then TI = 0.693 × 200 ≅ 140 msec FLAIR (Fluid-Attenuated
Inversion Recovery). This is an IR technique that nulls fluid. For
example, this sequence is used in the brain to suppress cerebrospinal
fluid (CSF) to bring out the periventricular hyperintense lesions, such as
multiple sclerosis (MS) plaques (Fig. 25-3).
Example
What is the TI used in FLAIR? At 1.5 T, T1 of CSF is approximately
3600 msec. Then TI = 0.693 × 3600 ≅ 2500 msec Fast FLAIR. STIR
sequences are typically performed with a fast spin-echo (FSE)
technique; however, there are some novel changes that must be
performed to apply FSE technique to the normally slow FLAIR sequence
and achieve CSF suppression in a fast manner.1 Figure 25-4 depicts a
schematic representation for Fast FLAIR. In this scheme, the following
parameters are used:
TR = 10,000 msec
TI = 2500 msec
FSE with ETL = 8
TEeff = 112 msec
 Figure 25-1. Inversion recovery (STIR shown). The recovery curves
following the 90° and 180° pulses are illustrated for fat and water.

In Figure 25-4, there are two packets of 15 slices each. During the first
5000 msec, 15 slice-selective 180° inverting pulses are followed 2500
msec (TI) later by 15 slice-selective FSE readout in 2500 msec. During
the second 5000 msec, recovery occurs in the first 15 slices (for a total
TR of 10,000 msec), and the process is repeated on the second set of 15
interleaved slices. In all, 30 slices are acquired.
A TR of 10,000 msec allows almost complete longitudinal recovery of
CSF. This relatively long time period also makes it possible to perform
multislice interleaving during inversion and readout. The inversion
“period” is the time during which 15 slice-selective 180° pulses are
applied. It is the time TI (2500 msec in this case) from the first 180°
inverting pulse to the 90° pulse at the beginning of the readout period.
FSE readout takes 136 msec (8 × 17 msec) for each slice. The readout
“period” is the time (also 2500 msec) during which 15 sliceselective
readouts are performed. The recovery period is the time from the slice-
selective 90° pulse

beginning an FSE readout to the next slice-selective 180° inverting

pulse, which is TR-TI (10,000 − 2500 = 7500 msec in this case).

Figure 25-2. TI required to null fat, brain, or CSF.

Figure 25-3. Axial T2 (A) and FLAIR (B) images of the brain
demonstrate the different conspicuity between the images of the
bright multiple sclerosis lesions.
 Figure 25-4. A schematic representation for Fast FLAIR. There are
two packets of 15 slices each. During the first 5000 msec, 15 slice-
selective 180° inverting pulses are followed 2500 msec (TI) later by
15 sliceselective FSE readout, in 2500 msec. During the second
5000 msec, recovery occurs in the first 15 slices (for a total TR of
10,000 msec) and the process is repeated on the second set of 15
interleaved slices. In all, 30 slices are acquired in 8 min.

The maximum number of slices that can be acquired in one TR is

determined either by the time period TI or by (TR-TI), whichever is
shorter. The longer the TR, the more complete recovery of
magnetization will be before the next excitation. A long TR will also
result in better signal-to-noise ratio (SNR) and increased contrast-to-
noise ratio (CNR) between the lesion and gray and white matter.
Unfortunately, when TR is increased, the TI necessary to null CSF does
not increase to the same degree. This difference results in inefficient
usage of time because (TR-TI) becomes larger and larger, and the
number of slices is still limited by TI. Given the TR of 10,000 msec and
the TI of 2500 msec used in this study, 5000 msec (i.e., 10,000 − 2 ×
2500) would be “wasted” before the next cycle. To increase the
efficiency, the sequence takes advantage of this dead time to acquire a
second multislice inversion-readout slab.
 Figure 25-5. Coronal T1 (A) of pelvis shows bright right adnexal
lesion. Coronal STIR (B) shows complete loss of signal, which may
imply fat, but is nonspecific since the FSE T2 with chemical
(spectral) fat saturation (C) shows that there is some signal within
the lesion, indicating that it is not a fat-containing lesion. Note the
chemical shift artifact in the T1 (A) lending further evidence that
this is not a fat-containing lesion. Diagnosis: endometrioma.

In this arrangement, the number of slices acquired can be doubled in

the same time duration with no dead time when the TR to TI ratio
equals 4. As the ratio approaches 6, yet another multislice section can
be added, and so on. This approach allows the flexibility to choose TR
and TI with little concern for time efficiency.
 Advantages of IR
1. No significant extra RF heating (unlike spectral presaturation—see
later text—due to the longer TI time vs. the very short time between
the spectral presaturation pulse and the root RF pulses)
2. No variability caused by magnetic field inhomogeneities (like spectral
presaturation—see below)

Disadvantages of IR
1. Tissues with similar T1 values are all suppressed and thus cannot be
differentiated (e.g., fat and blood or gadolinium-enhanced tumors
—Figs. 25-5 and 25-6)
 Figure 25-6. Coronal T1 postgadolinium image with fat saturation
(A) shows a large enhancing hemangioma in the left shoulder of a
young patient with Kesselbach-Merritt syndrome. Pregadolinium
coronal FSE T2 image (B) shows areas of bright signal; however, a
postgadolinium STIR image (C) shows signal dropout from the T1
shortening effects of gadolinium that make the lesion's T1
equivalent to that of fat.

2. Long acquisition times caused by long TRs

3. Low SNR
 Chemical (Spectral) Presaturation. In this technique, a frequency-
selective presaturation pulse is applied very shortly before the RF
excitation pulse, thus eliminating the longitudinal magnetization for a
specific tissue, for example, fat. It can be used to suppress fat or water
by appropriate selection of the frequency (based on the Larmor
equation, keeping in mind that at 1.5 T, water protons precess 220 Hz
faster than fat protons). This scheme is illustrated in Figure 25-7. Here,
the presaturation pulse is applied immediately before the 90° pulse of
the SE. The tissue whose Larmor frequency matches the frequency of
the presat pulse first sees a 90° spectral presat pulse that flips it into
the x-y plane. A short time later, the longitudinal component Mz has
really not had time to recover. Thus, the subsequent 90° excitation
pulse that flips this minimal Mz into the x-y plane generates a minimal
signal (Fig. 25-7). An example is in Figure 25-8.

Advantages
1. Resolves tissues with similar T1 values (such as fat and gadolinium-
enhanced tumors, or fat and blood products—Fig. 25-5)
2. No influence on the signal from tissues other than the one being
suppressed (in contrast, IR affects the contrast of all tissues)
 Figure 25-7. Spectral presaturation. A frequency-selective
presaturation pulse is applied before the excitation pulse to
eliminate longitudinal magnetization for a specific tissue such as
fat or water.

Disadvantages
1. Suffers from sensitivity to magnetic field inhomogeneities because it
employs a frequency-selective technique (Fig. 25-9)
2. Requires extra time (thus lengthening TR and increasing the scan
time)
3. Increases length of RF pulses, causing extra RF heating
Spatial Presaturation. Spatial presaturation is done generally to reduce
motion-and flowrelated artifacts of structures adjacent to—or actually
within the FOV next to—the region of interest. Examples include
 Figure 25-8. Axial T2 FSE image with chemical (spectral) fat
saturation shows suppression of fat signal. There is an intermediate
brightness left adrenal ganglioneuroma (arrows), which extends
toward the midline.

1. Imaging of the spine: a saturation band is placed within the FOV

anterior to the vertebral bodies to suppress artifacts arising from the
heart and great vessels (Fig. 23-8).
2. MR angiography: saturation pulses are placed outside the FOV at one
end of a vessel to suppress either venous flow (for achieving MR
arteriography [MRA]) or arterial flow (for achieving MR venography
[MRV]).
 Figure 25-9. Sagittal T2 FSE image with fat saturation (A) of the
knee shows marked field distortion and inhomogeneous fat
saturation due to metallic susceptibility artifact. Sagittal fast STIR
(B) shows homogeneous fat saturation and less distortion and
reveals a mild amount of bright signal in the distal femur. There
are also advanced patellofemoral osteoarthritis changes and a
large loose body posterior to the tibia (arrow).

Spatial presaturation is accomplished by applying an extra 90° pulse

before the 90° pulse of the SE sequence. This additional sat pulse
saturates tissues, including moving structures, within the slice,
eliminating signal (and thus associated artifacts). For more details,
refer to the discussion on saturation in Chapter 23.
Magnetization Transfer (MT). Magnetization transfer (MT) is a
technique that suppresses protein-bound water. The idea behind MT is
that protons in protein-bound water exhibit a resonant frequency that
is approximately 500 to 2500 Hz away from that of bulk water protons
(Fig. 25-10). MT saturation pulses are simply off-resonant pulses with
their center frequency 1000 to 2000 Hz removed from the Larmor
frequency of protons and a bandwidth (BW) of several hundred to
several thousand hertz, as shown in Figure 25-10A, thus allowing
 suppression of these protons. The result is shown in Figure 25-10B. Note
that the off-resonant protein-bound water is saturated. Because the
protein-bound protons and the bulk water protons are in rapid
exchange, the saturation is transferred to the bulk phase protons. Thus,
the peak amplitude of bulk water is reduced. MT is somewhat similar to
spectral fat suppression techniques except that here, the off-resonant
frequency is up to 2000 Hz (2 kHz) as opposed to 220 Hz in the case of
fat suppression.
This technique is used, for example, in time of flight (TOF) MR
angiography (see Chapter 27) as a means of suppressing the background
brain tissue to enhance visualization of smaller, more peripheral
vessels.
Magnetization Transfer Effect in FSE. As discussed in Chapter 19, MT is
inadvertently present in FSE because of the presence of multiple, rapid
180° pulses. These rapid 180° pulses have a broad BW that contains
frequencies off the bulk water resonance frequency that tend to
suppress protein-bound water.

Clinical Applications of Fat Suppression

1. To differentiate between fat and methemoglobin (chemical
saturation).
 Figure 25-10. Magnetization transfer. A: Protein-bound water
has a resonant frequency of about 500 to 2500 Hz off that of bulk
water. B: Spectral saturation of these offresonant frequencies
will result in suppression of protein-bound water. This technique
is used, for example, in MRA for background suppression.

2. Musculoskeletal: to minimize fat signal in the marrow to bring out

the signal from marrow edema (e.g., due to bone contusion, tumor,
and infection; chemical saturation or inversion recovery—Fig. 25-11).
3. Orbits: to suppress the retro-orbital fat to allow detection of
enhancing retro-orbital pathology on a contrast-enhanced study
(chemical saturation—Fig. 25-12).
4. Neck: to suppress fat in order to detect, and better evaluate the
extent of, a mass (chemical saturation or inversion recovery).
5. Body: to identify macroscopic fat-containing lesions such as
angiomyolipomas, mature teratomas, and myelolipomas (chemical
saturation).

Clinical Applications of Water Suppression.

Brain: to suppress CSF to bring out periventricular high intensity lesions
such as MS plaques, thus increasing their detectability (FLAIR).
Figure 25-11. Sagittal T2 FSE (A) and sagittal fast STIR (B) images
demonstrate better conspicuity of the bone marrow edema with
the STIR fat suppression technique. (Chemical saturation could be
used as well.)
Figure 25-12. Axial postgadolinium T1 images without (A) and with
fat saturation (B) show marked increased conspicuity of the
enhancing right optic nerve (arrow in B) in this patient with optic
neuritis.

Key Points
Tissue suppression is an important feature of MRI—it
allows improved tissue contrast and enhanced lesion
detectability. Generally, two tissues are subject to
suppression: fat and water. The two major
suppression techniques are inversion recovery (IR)
and chemical (spectral) saturation. Each technique
has its own advantages and disadvantages. The
selection of the type of technique depends on the
clinical application. Other tissues may be saturated
(e.g., protein-bound water and flowing blood).
Major fat suppression techniques include
1 STIR
2 Chemical (spectral) fat sat
Major fluid suppression techniques include
1 FLAIR
2 Chemical (spectral) water suppression
Techniques for suppression of protein-bound water
include
1 MT technique for background suppression
2 MT effect in FSE
Effects of presaturation technique in the FOV include
1 Saturation bands reduce flow artifacts (e.g., in
spine imaging)
2 Saturation pulses remove venous or arterial flow
(e.g., in MRA or MRV)
Questions
25-1 In an IR sequence, the TI to null a tissue is given by
(a) 0.693 T2
(b) 0.693 T1
(c) (1/0.693) T1
(d) (1/0.693) T2

25-2 The T1 of CSF at 1.5 T is about 3600 msec. What is the

approximate TI to null CSF?
(a) 2500
(b) 5000
(c) 140
(d) 249.48

25-3 T/F Protons in protein-bound macromolecules have a resonant

frequency that is about 220 Hz removed from that of bulk water.

25-4 T/F Magnetization transfer techniques saturate the off-resonant

protein-bound water protons.

25-5 T/F Fast or Turbo FLAIR is currently the most sensitive MR

sequence for detection of periventricular white matter disease.

25-6 Major water suppression techniques include

(a) STIR
(b) Spectral water suppression
(c) FLAIR
(d) all of the above
(e) only (b) and (c)
25-7 Major fat suppression techniques include
(a) STIR
(b) Spectral fat suppression
(c) FLAIR
(d) all of the above
(e) only (a) and (b)

25-8 T/F FSE demonstrates an inherent magnetization transfer

property.

1Hashemi RH et al. Suspected multiple sclerosis: MR imaging with a

thin-selection fast FLAIR pulse sequence. Radiology. 1995;196:505-510.

 26
Flow Phenomena

Introduction
Unlike computerized tomography (CT) (with or without contrast), in
which the appearance of flowing blood in a vessel is predictable, the
appearance of flowing blood in MRI is much more complicated. Flowing
blood or cerebrospinal fluid (CSF) can appear dark or bright depending
on numerous factors, including, but not limited to, the following:
1. Velocity
2. Pulse sequence (e.g., spin echo [SE] vs. gradient echo [GRE])
3. Position of the slice containing the vessel relative to the rest of the
slices
4. Contrast (TR and TE)
5. Echo number (even or odd)
6. Slice thickness
7. Flip angle
8. Gradient strength and rise time
9. Use of gradient moment nulling techniques
10. Use of spatial presaturation pulse
11. Use of cardiac gating
12. Chance of cardiac gating (pseudogating)

Types of Flow
In Chapter 18, we discussed two types of motion: random and periodic.
Blood and CSF flow have a periodic-type motion. Furthermore, flow of
blood can be divided into the following:
1. Laminar flow
2. Plug flow
3. Turbulent flow
4. Flow (stream) separation/vortex flow
These types of flow are depicted in Figure 26-1. Let's discuss these
separately.
Laminar Flow. This type of flow is seen in most normal vessels and has
a parabolic profile. If the lumen radius is R, then the velocity v at
position r would be
v(r) = Vmax (1 - r2/R2)
where Vmax is the maximum velocity in the center of the lumen.
Therefore, the average velocity in the lumen is given by
Vave = Vmax/2
Plug Flow. This type of flow is usually seen only in large vessels (aorta)
with high velocities. The velocity across the lumen is constant, thus
yielding a flat velocity profile:
v(r) = Vmax = Vave = constant
Turbulence. This phenomenon is observed in abnormal vessels (e.g.,
distal to stenosis) or at bifurcations during which a random motion of
fluid elements is observed. Vortex flow and large-scale recirculation
zones (i.e., eddies) are other terms that refer to such random motions.
Flow (Stream) Separation. This phenomenon is observed near the wall
of a vessel (e.g., in the proximal internal carotid artery) in which some
of the flow is separated from the main streamline.
Question: What determines laminar versus turbulent flow?
 Figure 26-1. Types of flow. A: Laminar flow. B: Plug flow. C:
Turbulent flow (distal to stenosis). D:Flow (stream) separation.

Answer: A dimensionless number called the Reynolds number (Re) can

predict the type of flow. It is given by
Re = (density × velocity × diameter)/viscosity
where density is in g/cm3, velocity is in cm/sec, diameter is in cm,
viscosity is in centipoise or g/cm-sec, and Re has no units. If Re < 2100,
then flow is laminar. If Re > 2100, then flow is turbulent.
Question: What is a typical blood flow velocity?
Answer: It depends on the vessel. Table 26-1 gives some examples.
The blood velocity may be much higher in abnormal conditions, such as
in an arteriovenous fistula (AVF).
Question: What is the difference between flow and velocity?
Answer: A mathematical relationship exists between bulk or volumetric
flow in a vessel lumen and the velocity of flow:
v = Q/A
where v is the average velocity (in cm/sec), Q is the bulk or volumetric
flow (in cm3/sec), and A is the cross-sectional area of the vessel (in
cm2).
 Normal Appearance of Flowing Blood. Most flow effects can be
attributed to one of the following:
1. Time-of-flight (TOF) effects
2. Motion-induced phase changes
TOF effects can lead to the following:
1. Signal loss (high-velocity signal loss or TOF loss)
2. Signal gain (flow-related enhancement [FRE])
Flowing blood can appear bright or dark. Three main independent
factors result in each case:
1. ↓ SI of flowing blood
a. high velocity
b. turbulent flow
c. dephasing
2. ↑ SI of flowing blood
a. even-echo rephasing
b. diastolic pseudogating
c. FRE

Let's discuss each of these separately.

Table 26-1

Vessel Blood Flow Velocity (cm/sec)

Aorta 140 ± 40

Superficial femoral artery 90 ± 13

Vertebral artery 36 ± 9

Venous flow <20

 Figure 26-2. To form an echo, protons must be exposed to both a
90° and a 180° pulse. As velocity increases, a larger portion of
flowing protons is exposed only to the 90° but not the 180° pulse,
thus causing signal loss.

High-Velocity Signal Loss

In SE imaging, the protons must be exposed to both a 90° and a 180°
radio frequency (RF) pulse to give off a signal. High-velocity signal loss
or TOF loss occurs when flowing protons do not remain within the
selected slice long enough to be exposed to both RF pulses.
Figure 26-2 illustrates how the magnitude of signal loss is a function of
the velocity. To see how this works, first recall the simple relationship
among time, distance, and velocity:
Velocity = distance/time

or
v = d/t
The distance each proton in flowing blood has to travel within a slice is
the slice thickness δz. The time between a 90° pulse and a 180° pulse
is one-half TE.
Thus, if the velocity of flowing blood is
v = δz/(1/2 TE)
then protons flowing into the slice would be exposed only to the 90°
pulse, and not to the 180° pulse, thus forming no signal or SE. Let's give
this velocity an arbitrary name vm, that is,
vm = δz/(1/2 TE)
However, if the velocity were 0 (i.e., for stagnant blood), then an SE
would be formed. If the velocity falls in between, only a fraction of the
protons would form an SE. The fraction of protons escaping the 180°
pulse is given by
v/vm = v(1/2 TE)/δz
Thus, the fraction of protons receiving both RF pulses is
1-v/vm = 1-v(1/2 TE)/δz
The signal intensity is, therefore, proportional to
I α 1-v(1/2 TE)/δz
Figure 26-3 illustrates the linear relationship between signal intensity
(I) and velocity v. It is clear from this graph that if the velocity of
flowing blood is at least equal to vm (i.e., v ≥ vm), then a flow void is
observed.
 Figure 26-3. A plot of signal intensity versus velocity. The higher
the velocity, the more the signal loss.

Figure 26-4. Axial PD image shows TOF losses in the right internal
carotid and basilar arteries (black arrows); however, no TOF loss or
“flow void” is seen in the left internal carotid artery (white
arrow). Patient had left internal carotid artery occlusion.

Example
Suppose that the slice thickness δz = 1 cm and TE = 50 msec. What is
the velocity at which flow void is observed?
Using the above formula, the velocity should be at least
vm = Δz/(1/2 TE) = 1 cm/(25 msec) = 1000 cm/25 sec = 40 cm/sec
which is seen in an artery.
Therefore, vm is larger for thicker slices and smaller TEs, and vice
versa (see Fig. 26-4 for an example).
N.B.: Remember that TOF losses only apply to SE imaging and not to
GRE imaging

because, in GRE, the echo is formed via a single RF pulse and

refocusing gradients (without a 180° pulse).

Figure 26-5. Sagittal 1.5-mm T2 image of the brain shows signal

loss inferior to the aqueduct of Sylvius (arrow). This was due to
high velocity of CSF and subsequent turbulent flow resulting in
intravoxel dephasing. Patient had normal pressure hydrocephalus
(NPH).
 Turbulent Flow
In turbulent flow (with Re>2100), random flow exists that contains
different velocity components (i.e., with different speed and direction
of flow). Consequently, each of these components has different phases
that tend to cancel each other out and result in no signal (flow void).
This result can occur with low- or high-velocity flow (Fig. 26-5).

Dephasing
There are many causes of dephasing. One important cause is the so-
called odd-echo dephasing. This phenomenon results in signal void on
the first and other odd echoes. The protons in a voxel do not move at
the same velocity across the lumen in laminar flow; thus, each
precesses at a different frequency and accumulates a different phase.
(Also see the discussion below on evenecho rephasing.)
Another cause of dephasing is intravoxel dephasing. Because of
laminar flow, different velocities may exist within a voxel, thus leading
to phase dispersion (incoherence) and signal loss. How to decrease
intravoxel dephasing and increase SNR:
1. Decrease the voxel size (increase spatial resolution), either by
increasing the matrix (trade-off: will reduce signal-to-noise ratio
[SNR]) or by reducing the FOV (trade-off: may cause wraparound).
2. Reduce the TE (e.g., use a fractional echo).
3. Add flow compensation (FC) techniques (see below).

Even-Echo Rephasing
This phenomenon is somewhat the opposite of odd-echo dephasing. It
only occurs in SE imaging with symmetric echoes (i.e., when the second
echo delay time is twice the first one: when TE2 = 2 TE1, e.g.,
30/60/90/120 and 40/80/120/160). The result is higher signal intensity
for even echoes compared with odd echoes. (We shall see later that in
GRE or SE, FC techniques basically yield “even-echo rephasing” on the
first echo to minimize signal losses.)
To see how this works, we first need to learn about the relationship
 between phase and velocity. Recall that along the readout gradient,
ω = γBX = γGX
Now, for a constant velocity v, the position at time t is given simply by
x = vt
Thus, combining the above two, we get
ω = γGvt
Now, phase change Δφ and angular frequency ω are related by
δφ = ωδt
so that
φ = ∫ωdt = ∫(γGvt)dt = γGv∫t dt = γGv(t2/2)
From this we can see the following:
1. Phase φ and velocity v are proportional.

2. A quadratic relationship exists between phase φ and time t, that is,

φ = kt2, where k is a constant (k = 1/2 γGv).
On the other hand, for stationary tissue (with v = 0), the above
relationship becomes
φ = k′t
that is, a linear relationship exists between phase and time for
stationary tissues.
Now consider an SE sequence with symmetric echoes as shown in Figure
26-6. The phase changes over one cycle for both stationary tissues and
flowing blood are plotted. This graph reveals that stationary tissues
have zero phase on the first echo (TE) and the second echo (2TE). The
situation for flowing blood, however, is different. On the first echo, the
flowing blood has a positive phase. However, on the second echo, the
phase is back to zero!
 Figure 26-6. Phase accumulation for stationary and flowing
protons.

MATH: Let's try to prove the above fact mathematically. Assume that τ
= 1/2 TE. The phase gain at time TE/2 is kτ2, where k is a constant (k =
1/2 γGv). Right after the 180° pulse, the phase is -kτ2. Now, at time TE
(i.e., at 2τ), the phase gain will be k[(2τ)2 - (τ)2] = 3kτ2. (The
mathematically oriented reader will recognize this as the integral of ∫
2kτ from point τ to 2τ.) Thus, the net phase gain will be 3kτ2 + (-kτ2) =
2kτ2. Similarly, at 3/2 TE (or 3τ), that is, at the second 180° pulse, the
phase gain is k[(3τ)2 + (2τ)2] = 5kτ2, with a net gain of 5kτ2 + 2kτ2 =
7kτ2. Immediately after this second 180° pulse, the phase will be -7kτ2.
In a similar fashion, the phase gain at the time of the second echo (i.e.,
4τ) will be k[(4τ)2 - (3τ)2] = 7kτ2. Therefore, the final net phase gain
will be 7kτ2 + (-7kτ2) = 0. That is to say, the net phase shift of flowing
blood on the second echo is zero. This is illustrated in Figure 26-6.

Thus, protons in flowing blood lose their coherence on the first echo
(causing odd-echo dephasing) and subsequently regain coherence on
the second echo (causing even-echo rephasing). This pattern yields a
higher signal of flowing blood on the second echo. It only works for
symmetric echoes, not for asymmetric ones. (Interested readers can go
through the previous math and prove this fact for themselves.)

Diastolic Pseudogating
During a cardiac cycle, blood flow is more rapid during systole and
slower in diastole. Thus, in diastole, a higher intravascular signal is
observed. (Remember that higher flow results in more TOF losses.)
When cardiac gating is used, each slice is (theoretically) acquired at a
fixed point in the cardiac cycle (although different slices in a multislice
acquisition are obtained at different points in the cycle). Consequently,
a vessel traversing several slices will demonstrate varying signal
intensities at different slices. In a cardiac-gated sequence, TR must be
a multiple of the heart rate (HR). For example, if HR = 60 bpm (1 beat
per second = 1 Hz), then TR = 1000, 2000, 3000, etc. In general,
TR = 1/HR
with appropriate units.

Flow-Related Enhancement (FRE)

 This phenomenon usually refers to the first slice that the flowing blood
enters. Hence, FRE is also called the entry phenomenon. FRE is a type
of TOF effect in which the fresh inflowing blood that enters the first
slice is totally unsaturated, that is, the protons within it have not as
yet been subjected to any prior RF pulse and thus yield full
magnetization, whereas the adjacent stationary tissue remains partially
saturated because of prior RF pulses.
Figure 26-7 illustrates the relationship between FRE and the blood
velocity. This diagram demonstrates that when v = 0 (i.e., stagnant
blood), the protons are partially saturated. However, when the velocity
is v = δz/TR, then the unsaturated inflowing protons completely replace
the previously partially saturated protons. Again, we shall give this
velocity an arbitrary name vM. The fraction of inflowing protons is then
v/vM = v(TE/δz)
Thus, the relationship between the intravascular signal intensity and
velocity is
I α I0 + (TE/δz)v
where I0 is the signal intensity of stagnant blood (i.e., at v = 0). This
relationship is illustrated in Figure 26-8.
Example
At what velocity do you observe maximum FRE when δz = 1 cm and
TR = 1000 msec = 1 sec? From the previous formula,
vM = δz/TR = 1 cm/1sec = 1cm/sec
which is consistent with a slow venous flow (Figs. 26-9 and 26-10).
Why Is Flowing Blood Bright on GRE Images? You may have noticed
that on most GRE images, vessels appear bright on all the slices. There
are three main reasons for this:
1. GRE imaging is usually performed in a sequential mode (i.e., one
slice at a time). Consequently, every slice is an entry slice. Thus, FRE
applies to every slice in the volume.
Due to lack of a 180° refocusing pulse and the nonslice selectivity of
 the refocusing gradient, TOF losses are usually not significant in GRE
imaging.
2. In GRE imaging, TE is usually very short, which minimizes signal losses
caused by dephasing.
See Figure 26-11 as an example.
Question: Is FRE only limited to the first (entry) slice?
Answer: The answer is no. If the velocity of flow is higher than δz/TR
(but not too high for TOF losses to take place), then unsaturated

protons can penetrate adjacent slices and yield FRE. Obviously, as these
inflowing protons travel through the imaging volume, they are
subjected to more and more RF pulses and become more and more
saturated. Thus, FRE is always maximum at the entry slice and becomes
gradually fainter in deeper slices (a good distinguishing point from an
intraluminal thrombus). Now, how far FRE can penetrate the imaging
volume has to do with the direction of flow and the direction of slice
excitation.
Figure 26-7. Effect of flow-related enhancement (FRE) for slow
flow. Unsaturated protons produce more signals. As the velocity
increases, a larger portion of unsaturated inflowing protons will
replace the previously partially saturated protons, thus increasing
signal intensity.

Figure 26-8. The plot of signal intensity versus velocity in FRE.

Figure 26-9. Axial T1 (A) through the pelvis shows flow-related
enhancement within the slower flowing external iliac veins
(arrows); the faster flowing arteries show time-of-flight loss. By
increasing the TE in the accompanying T2 FSE image with fat
saturation sequence (B), more time-of-flight losses are seen within
the right external iliac vein in addition to the arteries
(arrowheads). However, incomplete time-of-flight losses are seen
within the left external iliac vein (arrow) due to a component of
even slower flow within this vein.

Figure 26-10. Sagittal T1 image (A) of the brain demonstrates a

tangle of flow voids in a patient with an arteriovenous
malformation (AVM). However, an axial T1 image (B) shows some
vessels with flow-related enhancement (arrow). This demonstrates
the phenomenon of in-plane saturation of protons (the sagittal
image) yielding signal void and out-of-plane unsaturated protons in
the axial image resulting in flow-related enhancement.
 Figure 26-11. Gradient-echo axial image from 2D time of flight
shows bright signal from flow-related enhancement in the internal
carotid and vertebral arteries. The patient also has an ectatic
course of his left vertebral artery, accounting for his left hemifacial
spasm.

Cocurrent and Countercurrent Flow. The slice-excitation wave (SEW)

is the direction of successive 90° excitation pulses. If flow is
perpendicular to the slices, then flow is either in the direction of the
SEW (called cocurrent) or against it (called countercurrent).
Intuitively, in a countercurrent setting, the flowing protons are
subjected to fewer 90° pulses than in a cocurrent setting (Fig. 26-12).
Consequently, FRE demonstrates deeper penetration in a
countercurrent setting (Figs. 26-13 through 26-15).
 Table 26-2

Manufacturer Acronym Description

GE FC Flow compensation

Philips FC Flow compensation

Siemens GMR Gradient-motion rephasing

Combined Flow Phenomena. What would happen if very high-velocity

flow enters the entry slice? Although FRE affects the entry slice
increasing intraluminal signal, TOF signal losses will also have an effect,
thus somewhat offsetting FRE. In other words, in reality, flowing blood
demonstrates a combination of flow phenomena, and the intensity of
blood is, in the final analysis, determined by the phenomena that
predominate.

Gradient Moment Nulling (Flow Compensation)

Gradient moment nulling (GMN) is one method of minimizing flow
motion artifacts. It is based on the principle of even-echo rephasing.
Evenecho rephasing is accomplished by adding extra gradient pulses to
produce the even-echo rephasing effect on the first echo (thus
eliminating firstecho dephasing). You can therefore achieve rephasing
without having to use a double-echo sequence. There are several
synonyms for GMN (Table 26-2).
Figures 26-16 and 26-17 demonstrate FC for both GRE and SE imaging.
In both cases, protons in flowing blood will have a net phase of zero at
the center of the echo. The mathematics is similar to the one for even-
echo rephasing (see previous section), and the interested reader may
wish to go through the exercise. The extra gradients associated with FC
are referred to as gradient lobes. For instance, for GRE, the relative
gradient strengths of these

lobes demonstrate a ratio of 1:2:1. Remember that this type of FC only

corrects for firstorder1 (i.e., constant velocity) flow. To correct for
higher order motion (e.g., acceleration or jerk), additional gradient
lobes are required (Fig. 26-18).2 As you can see, the addition of these
extra lobes lengthens the cycle, thus lengthening TR and minimum TE
(thus reducing the number of slices).
Figure 26-12. A: Cocurrent flow. The flow and slice-excitation
wave (SEW) are in the same direction. B: Countercurrent flow. The
flow and SEW are in opposite directions.

Figure 26-13. FRE penetrates deeper in countercurrent flow than

in cocurrent flow.
 Figure 26-14. Axial T1 image of the abdomen (A) shows bright
flow-related enhancement (FRE) in the IVC (arrow) on only the
most inferior three cuts in this sequence, consistent with
countercurrent flow-related enhancement. Comparison T1 image
(B) at the superior slice shows no signal within the IVC; however,
there is FRE within the aortic aneurysm secondary to slower flow
within the aneurysm. There is also motion artifact on image B in
the phaseencode direction (anteroposterior) from the aneurysm
(arrows).

FC can be applied along each and all the three coordinates x, y, and z.
Figure 26-15. Series of T1 images of the femoral vessels acquired
superior to inferior demonstrates the phenomena of countercurrent
flow-related enhancement. Note that on the first few images (the
most inferior slices) there is very bright signal in the
countercurrent venous flow (white arrows), whereas the arteries
remain dark (black arrow in first image). Notice that as the slices
become deeper within the volume, there is less signal until it is
nearly dark on the last image. Note that the periphery is brighter
than the center as opposed to the schematic in Figure 26-13. This
is because higher velocities in the center preferentially result in
more TOF loss in this patient, while if the flow is slower, the center
will have higher signal as seen in Figure 26-13.

Figure 26-16. Flow compensation in GRE. With this scheme,

flowing protons (with constant velocity) are in phase (i.e., zero
phase difference) at the center of the echo.
Figure 26-17. Flow compensation in SE. With this scheme, flowing
protons are in phase at the center of the echo.
Figure 26-18. A: Secondorder flow (acceleration compensation).
With this scheme, protons with secondorder flow (i.e.,
acceleration) get in phase at the center of the echo. B: Third-order
flow (jerk) compensation. With this scheme, protons with third-
order flow (i.e., jerk) get in phase at the center of the echo.
Key Points
1 Several types of flow exist: laminar flow, plug flow,
turbulent flow, and flow separation/vortex flow.
2 Intravascular flow is usually laminar, that is, it has
a parabolic profile.
3 Turbulent flow occurs distal to stenoses and on
vessel turns.
4 The Reynolds number can predict laminar versus
turbulent flow.
5 Most flow effects can be attributed to one of two
phenomena: TOF or motion-related phase changes.
6 TOF can cause either signal loss (TOF loss) or
signal gain (FRE).
7 Causes of intravascular signal loss include high
velocity, turbulent flow, and dephasing.
8 Causes of intravascular signal gain include FRE,
even-echo rephasing, and diastolic pseudogating.
9 Dephasing may be caused by intravoxel dephasing
or odd-echo dephasing.
10 FRE is based on the entry phenomenon. FRE is
the basis for TOF MR angiography (MRA).
11 FRE penetrates deeper slices when flow is
countercurrent to the slice-excitation wave (SEW)
rather than when it is cocurrent.
12 Even-echo rephasing causes bright intravascular
signal on the second echo (when the echoes are
symmetric).
13 Even-echo rephasing is the basis for FC (GMN).
GMN, however, is accomplished via gradients that
cause rephasing of flowing spins on the first echo and
 thus increased signal.

Questions
26-1 Normal intravascular flow is usually:
(a) turbulent flow
(b) plug flow
(c) laminar flow
(d) all of the above
(e) none of the above

26-2 T/F Turbulent flow is expected proximal to a stenosis.

26-3 T/F FRE is synonymous with the entry phenomenon.

26-4 T/F Time-of-flight effects only cause signal loss.

26-5 Causes of intravascular signal loss include

(a) high velocity
(b) turbulent flow
(c) dephasing
(d) all of the above
(e) only (a) (b)

26-6 Causes of intravascular signal gain include

(a) FRE
(b) even-echo rephasing
(c) diastolic pseudogating
(d) all of the above
 (e) only (a) and (c)

26-7 T/F The Reynolds number (Re) can predict laminar versus
turbulent flow.

26-8 T/F FRE penetrates deeper slices when flow is cocurrent rather
than countercurrent.

26-9 T/F FRE is only observed in the first (entry) slice.

26-10 T/F Even-echo rephasing is the basis for flow compensation

techniques.

26-11 Laminar flow is given by the formula:

(a) v(r) = Vmax (1 - r2/R2)

(b) v(r) = Vmax (1 - R2/r2)

(c) v(r) = (1 - r2/R2)/Vmax
(d) There is no formula for such a flow.

26-12 Plug flow is given by the formula:

(a) v(r) = Vmax (1 - r2/R2)
(b) v(r) = Vmax (1 - R2/r2)
(c) v(r) = constant = Vave
(d) none of the above

26-13 TOF effects can lead to

(a) signal loss
(b) signal gain
(c) both
(d) none of the above
26-14 T/F Given the parameters TR 2000, TE1 20, TE2 80 msec,
even-echo rephasing would occur.

26-15 T/F Laminar flow has a parabolic profile.

26-16 Flow effects include

(a) time-of-flight (TOF)
(b) motion-induced phase changes
(c) both
(d) none of the above

1Zero-order motion is stationary (v = 0). First-order motion (v = dx/dt =

constant) is constant velocity. Second-order motion (a — d2x/d2t) is

acceleration. Third-order motion (j — d3x/d3t) is jerk motion or
pulsatility.
2For a more complete discussion, refer to Stark DO, Bradley WG, eds.

Magnetic Resonance Imaging, 3rd ed, vol 1. St. Louis: Mosby; 1999.
 27
MR Angiography

Introduction
In this chapter, we will discuss the topic of MR angiography (MRA). As
in the last chapter, MRA may at first appear very complicated, but we'll
try to present the major concepts in a simplified fashion. There are
three main MRA techniques:
1. TOF (time-of-flight) MRA
2. PC (phase contrast) MRA
3. CE (contrast-enhanced) MRA
TOF and PC techniques can be performed using two-dimensional Fourier
transform (2DFT) or three-dimensional FT (3DFT). CE MRA is performed
with the 3D technique. Thus, there are a total of five different
methods:
1. 2D-TOF MRA
2. 2D-PC MRA
3. 3D-TOF MRA
4. 3D-PC MRA
5. 3D-CE MRA
Each of these techniques lends itself to a different type of clinical
application.

TOF MRA
TOF MRA is based on flow-related enhancement (FRE; discussed in the
previous chapter) in a 2D or 3D gradient-echo (GRE) technique.
(Remember that in GRE imaging, TOF losses do not play an important
role.) Usually, flow compensation is used perpendicular to the vessel
 lumen.
2D-TOF MRA. Figure 27-1 depicts a typical pulse sequence for 2D-TOF
MRA. A presat (presaturation) pulse is applied above or below each slice
to eliminate the signal from vessels flowing in the opposite direction.
Usually a short TR (about 50 msec), a moderate flip angle (45° to 60°),
and a short TE (a few millisecond) are used. An example is seen in
Figure 27-2.
3D-TOF MRA. Figure 27-3 depicts a pulse sequence diagram (PSD) for a
3D-TOF MRA. Here, a slab of several centimeters (usually about 5 cm) is
obtained that contains up to 60 slices.

Advantages of 2D-TOF MRA

1. Faster scanning
2. Maximized FRE because each slice is an entry slice

Disadvantages of 2D-TOF MRA

1. In plane saturation effects (discussed later)

Advantages of 3D-TOF MRA

1. Higher signal-to-noise ratio (SNR) because signal is acquired from a
larger volume
2. Improved spatial resolution

Disadvantages of 3D-TOF MRA

1. 3D techniques more susceptible to saturation effects (see below)
2. Less sensitive to slow flow
 Figure 27-1. A PSD for 2D-TOF MR angiography.

PC MRA
PC MRA is based on the fact that the phase gain of flowing blood
through a gradient is proportional to its velocity (assuming constant
velocity). We saw in the previous chapter that phase (φ) and velocity
(v) are related by φ = ∫ωdt = ∫(γGvt)dt = 1/2γGvt2 Therefore, knowing
the phase at any point in time allows us to calculate the velocity.
The most common method to employ PC MRA is by the use of a bipolar
gradient (Fig. 27-4A). This process is called flow encoding. Because the
two lobes in this bipolar gradient have equal area, no net phase change
is observed by stationary tissues (Fig. 27-4A). However, flowing blood
will experience a net phase shift proportional to its velocity (assuming
a constant flow velocity) (Fig. 27-4B). This is how flow is distinguished
from stationary tissue in PC MRA. Figures 27-5 and 27-6 illustrate the
PSDs for 2D-PC and 3D-PC MRA, respectively.
There are several features unique to PC MRA, as the following
discussions demonstrate.
Question: What are the “magnitude” image and the “phase” image?
Answer: In PC MRA, you not only get an image of the blood vessels
(magnitude image); you also get an image that shows you the direction
of flow (phase map). The phase image would tell you whether the flow
is right-left, superior-inferior, or anterior-posterior. An example is
determination of hepatofugal versus

hepatopedal flow in the portal vein of a patient with cirrhosis.

Figure 27-2. Axial source image from a 2D-TOF demonstrates no
flow-related enhancement in the left internal carotid artery
(arrow), consistent with occlusion from arterial dissection in this
young patient. Normal flow-related enhancement is seen in all the
other vessels.
Figure 27-3. A PSD for 3D-TOF MRA.
 Figure 27-4. A: A bipolar flow-encoding gradient. Because the two
lobes have equal areas with opposite polarities, no phase change is
observed by stationary spins. However, flowing spins will yield a
net phase change proportional to their velocity, which is the
principle behind PC MRA (B).

Question: What is VENC?

Answer: VENC stands for velocity encoding. It is a parameter that is
selected by the MR operator when using PC MRA. VENC represents the
maximum velocity present in the imaging volume. Any velocity greater
than VENC will be aliased according to the following formula: Aliased
velocity = VENC − actual velocity

For instance, if VENC = 30 cm/sec, then a vessel with a flow velocity of

40 cm/sec will be represented as a flow of v = 30 − 40 = −10 cm/sec
that is, a flow of 10 cm/sec in the opposite direction (Fig. 27-7). A
smaller VENC is more sensitive to slow flow (venous flow) and to
smaller branches but causes more rapid (arterial) flow to get aliased. A
larger VENC is more appropriate for arterial flow. Sometimes you may
image the same thing with two different VENCs—a small VENC and a
large VENC—to image all the flow components accurately (an example
is imaging an AVM or an aneurysm).

Figure 27-5. A PSD for 2D-PC MRA.

 Figure 27-6. A PSD for 3D-PC MRA.

Advantages of PC MRA
1. The capability to generate magnitude and phase images
2. Superior background suppression
3. Less sensitive to intravoxel dephasing or saturation effects
 Figure 27-7. Two phase contrast images (phase maps) through the
root of the aorta in a patient with aortic stenosis. A: VENC = 150
cm/sec; B: VENC = 300 cm/sec. Note in (A) the bright central
signal (arrow) that may indicate regurgitant flow; however, the
signal is surrounded by darker signal with abrupt transition from
black pixel to white pixel diagnostic of aliasing. Increasing the
VENC in (B) eliminated the aliasing. No bright signal indicative of
regurgitation was observed and an accurate maximal velocity
determination was now possible.

Disadvantages of PC MRA
1. Takes longer to do
2. More sensitive to signal losses caused by turbulence and by dephasing
on vessel turns (e.g., carotid siphon)
3. The need to guess the maximum flow velocity in order to select an
optimum VENC
 Figure 27-8. Coronal (A) and axial (B) thick slab (5 cm) 2D-PC MRV
(magnetic resonance venogram) of the brain in normal patients.

2D- versus 3D-PC MRA

1. 2D techniques are faster
2. 3D techniques have better SNR
Examples of magnitude phase contrast images are seen in Figures 27-8
through 27-10.
 Figure 27-9. 2D sagittal phase contrast MRV (magnetic resonance
venogram) (A) shows markedly decreased flow in the superior
sagittal and straight sinuses (arrows) in this patient with sagittal
sinus thrombosis. B: Normal exam. Note the phase-induced
ghosting (arrows).

CE MRA
CE MRA is different than either TOF or PC imaging since CE MRA is
primarily dependent on the T1 properties of gadolinium in the
vasculature rather than the flow properties per se. This technique has
been made possible due to the advent of high performance gradients
(more in Chapter 30) that permit very rapid GRE imaging and the use of
the paramagnetic contrast agent gadolinium. This allows imaging to be
achieved during the transit time and hence peak T1 shortening of the
gadolinium. This technique is therefore very dependent on the precise
timing of the arrival of the bolus of gadolinium in the vessel of interest.
The plane of imaging is usually in the plane of the vessel (usually
coronal) as opposed to 2D-TOF technique in which the imaging plane is
usually orthogonal to the vessel of interest. Imaging this way increases
coverage while maximizing resolution. Since this technique is more
reliant on the T1 properties than any flow properties, it is very
resistant to dephasing artifacts that are seen in some of the other
techniques.
 Figure 27-10. Axial 3D phase contrast MIP (maximum intensity
projection) image of the renal vessels shows normal flow without
stenosis. This is a magnitude image reflecting higher velocities in
the renal arteries represented by brighter signal.

There are two principal CE-MRA techniques: elliptical-centric and

multiphase. In the former, acquisition begins as contrast enters the
artery of interest, filling the center of k-space (containing most of the
SNR) first and the veins later (during filling of the low SNR periphery of
kspace). This technique requires automatic bolus detection software
(GE SmartPrep, Siemens CARE Bolus, or Philips BolusTrak) or a timing
run to determine the time it takes the gadolinium to reach the artery
of interest. The former is performed by placing a cursor over the
upstream portion of the artery of interest; the latter is performed by
injecting 2 cc of gadolinium and noting the time it takes the artery to
turn maximally bright. In multiphase techniques, gadolinium is injected
and multiple acquisitions are made, one of which is bound to

be in the arterial phase. An additional benefit of multiphase

acquisitions is the ability to provide pathologically delayed vessel filling
of contrast material. But one important limitation to multiphase
imaging is the tradeoff of spatial resolution for temporal resolution.
One approach to reduce loss of spatial resolution in multiphase imaging
relies on the fact that much of the information to form an MR image is
present in the central region of k-space. A faster time-series of 3D
images can be reconstructed by acquiring a multiphase exam in which
the central phase encoding values are acquired more often than the
outer regions of k-space. This technique, time-resolved imaging of
contrast kinetics (TRICKS), divides the 3D Cartesian kspace into several
subvolumes located at increasing distance from the k-space center and
oversamples the central region of k-space relative to the sampling rate
of the outer regions. In this way, TRICKS is able to consistently capture
an arterial phase free of venous overlay.
The resampling of the center of k-space results in a reduction in spatial
resolution in TRICKS exams compared with single-image acquisitions
acquired in the same scan time. Undersampled projection
reconstruction has proved to preserve the spatial resolution and speed
up the acquisition with limited streak artifacts. The combination of
undersampled PR (projection reconstruction) acquisition in the kx-ky
plane with a TRICKS encoding in the slice direction can significantly
increase the temporal resolution without spatial resolution
degradation.
 Figure 27-11. Coronal source image (A) and selected MIP
(maximum intensity projection) (B) of the left carotid artery from
a CE MRA shows focal high-grade stenosis at the origin of the left,
internal carotid. Notice that on the source the entire course of
either artery is not seen secondary to the thin slice.

Advantages
1. Rapid technique
2. Resistant to dephasing (e.g., from turbulent flow)
3. Large field of view with good resolution
4. Excellent SNR

Disadvantages
1. Dependent on timing (artifacts or venous contamination may occur)
2. Requires intravenous injection for administration of gadolinium
3. No directional information
See Figures 27-11 through 27-14 for examples.
Table 27-1 contains a summary of some of the major clinical
applications of the five methods of MRA discussed above.
Figure 27-12. Coronal source images (A and B) show a focal high-
grade stenosis of the left proximal main renal artery (arrow in A),
whereas the origin of the right main renal artery is normal (arrow
in B). Note that because these are thin slices, the images do not
usually show the entire vessels. The MIP (C) image in this patient
shows the left renal artery stenosis (arrow) and an accessory right
renal artery (arrowhead).

Figure 27-13. TRICKS imaging of the renal artery.

Figure 27-14. Asymmetric filling of the contrast agent is
demonstrated with PR-TRICKS acquisition with a high spatial
resolution of 1.1 × 1.1 × 1.5 mm3 and temporal resolution of 3.8
sec per frame.

Table 27-1

Technique Clinical Applications

2D -TOF Carotid and vertebral arteries in the neck

MRA Venous structures (due to slow flow)

3D -TOF Intracranial vasculature (circle of Willis)

MRA Intracranial vascular malformations and
aneurysms

2D -PC Portal vein

MRA CSF flow study
 Localizer for determining VENC (velocity
encoding)

3D -PC Intracranial vasculature

MRA Intracranial vascular malformations and
aneurysms

3D -CE Carotid and vertebral arteries of the neck

MRA Aortic arch, renal arteries, and upper or lower
extremity runoff
Figure 27-15. Maximum intensity projection (MIP). A channel of
similar pixels in every slice is selected, and the pixel with the
maximum intensity (provided its intensity is above a certain
threshold) is projected onto an image. This process is repeated for
all pixels to form an image.
Maximum Intensity Projection
We can finally explain how we can image just the blood vessels (in a
way that looks 3D) and not the stationary tissue. This imaging is
accomplished via an algorithm called maximum intensity projection
(MIP). MIP can be used as a noun, verb (the raw data are mipped), or
adjective (mipped image). Mipping is done as follows: because flowing
blood in MRA techniques has high intensity, the intensity of a pixel in a
slice is compared with the corresponding pixels in all the other slices
(as in a channel), and the one with maximum intensity is selected. For
example, pixel (1,1) in slice 1 is compared with pixel (1,1) of all other
slices. This process is repeated for all the pixels in the slice. In other
words, the high-intensity dots in space are connected to generate an
MRA image. Thus, the mipped image represents the highest intensities
(hopefully all caused by flowing blood) in the imaging volume. This
image is illustrated in Figure 27-15. Obviously, a certain internal
threshold is used, below which no pixel in the channel falls. An example
of a CE-MRA MIP with a comparison 2D-TOF MIP is seen in Figure 27-16.

Figure 27-16. CE MRA MIP image (A) and comparison 2D-TOF MIP
(B) show occlusion of the left, internal carotid artery in the same
 patient as Figure 27-2.

Disadvantages of MIP
The major drawback of MIP is that bright structures other than flowing
blood may potentially be included in the mipped image. Examples are
fat, subacute hemorrhage, and the posterior pituitary gland (Fig. 27-
17). This problem is mainly with TOF MRA and not with PC MRA.

(The latter is a subtraction technique based on velocity-induced phase

shifts rather than on tissue T1 and T2.)

Figure 27-17. Sagittal MIP (A) image shows worrisome signal for
aneurysm near the anterior communicating artery and the internal
carotid terminus. Additional sagittal T1 (B) shows the high signal to
be pituitary hemorrhage and not flow-related enhancement.

Saturation Effects
Saturation effects refer to the gradual loss of longitudinal
magnetization caused by repeated excitation radio frequency (RF)
 pulses. This, in turn, leads to loss of signal (and thus reduced SNR). This
problem usually arises in a 2D acquisition in which flowing blood has to
travel within (rather than through) a slice or in a 3D acquisition in
which the blood travels through a thick imaging volume (or slab). In
such a situation, saturation effects may cause the distal portion of a
vessel not to be included in the image.

Figure 27-18. Saturation effects. A longer TR (A, for a fixed flip

angle α) causes better recovery of longitudinal magnetization than
a shorter TR (B) thus reducing saturation effects.

There are two main causes of saturation effects:

1. Decreasing TR
2. Increasing α
Decreasing TR. As shown in Figure 27-18, a shorter TR causes less
recovery of longitudinal magnetization from one cycle to the next,
causing gradual loss of the Mz component. This effect is less
pronounced with a longer TR.
Increasing α. Next consider the case of a large α. A larger α causes
more loss of longitudinal

magnetization. Therefore, for a given TR, there is more gradual loss of

Mα with a smaller α than with a larger one (Fig. 27-19).

Figure 27-19. Saturation effects. A smaller flip angle α (A, for a

fixed TR) also causes more recovery of longitudinal magnetization
than a larger flip angle α (B) thus leading to reduced saturation
effects.

In GRE technique, saturation effects become problematic because very

short TRs are used. The use of small flip angles counteracts this effect.
These saturation effects become especially important in 2D or 3D in-
plane flow or in 3D imaging in which volume imaging is performed over
a slab, and signal losses might be significant from one end of the slab to
the other. Multislice GRE techniques that use longer TRs decrease these
saturation effects and allow for larger flip angles (which improves the
SNR).
We have already discussed another mechanism for reducing these
saturation effects: using a paramagnetic contrast agent, such as
gadolinium (CE MRA). The use of this agent causes T1 shortening of
blood. Consequently, the T1 recovery is much faster with fewer
saturation effects (Fig. 27-20).
 Figure 27-20. Saturation effects. Use of gadolinium (Gad) causes
T1 shortening, which reduces saturation effects. A: Without Gad.
B: With Gad.

Two other techniques are available to reduce saturation effects: MOTSA

(multiple overlapping thin-slab acquisition) and TONE (tilted optimized
nonsaturating excitation).
MOTSA. Motsa is a combination of 2D-TOF and 3D-TOF techniques for
the purpose of reducing the saturation effects associated with a thick
slab. In this method, multiple thin slabs are used, which are
overlapping by 25% to 50% (Fig. 27-21). The final imaging volume is
created by extracting the central slices of each slab and discarding the
peripheral slices (which are more affected by saturation effects). The
main drawback of this technique is the potential for Venetian blind
artifact at the points where the slabs overlap. See Figures 27-22 and
27-23 for examples.
 Figure 27-21. MOTSA (multiple overlapping thin-slab acquisition).
To reduce saturation effects in deeper slices, multiple thinner
slices are processed (in this example, 7). Next, the peripheral
slices in each slab are discarded and the remaining slices are
extracted and combined to form an image.

TONE. In this scheme, the flip angle α is increased progressively as the

flowing spins move into the imaging volume by using increasing RF
pulses. Recall that a larger α yields a higher SNR. Thus, increasing α
counteracts saturation effects of slowly flowing blood in deeper slices.
This allows better visualization of distal vessels and slow-flowing
vessels. This scheme is illustrated in Figure 27-24 in which a ramped
flip angle excitation pulse is used. In this example, the center flip angle
is 30° and the flip angle at each end varies by 30% (i.e., 20° at the
entry slice and 40° at the exit slice).
Figure 27-22. MOTSA MRA of the circle of Willis shows a left
parietal lobe AVM.

There are five main ways of reducing saturation

effects:
1. Decreasing the flip angle α
2. Increasing the repetition time TR
3. CE MRA
 4. MOTSA
5. TONE

Question: How does magnetization transfer (MT) affect MRA?

Answer: Magnetization transfer (MT) saturation was discussed in
Chapter 25. MT is based on suppression of the off-resonant protein-
bound water protons (e.g., brain tissue). This technique, combined
with TOF MRA, helps to suppress the background signal (e.g., it can
reduce the signal from brain parenchyma by about 30%), increasing
conspicuity of small and distal

branches, vessels with slow flow, and aneurysms. MT can also be

combined with TONE for further visualization of small vessels.
Figure 27-23. The MOTSA technique can also be used for
extracranial carotid evaluation, as seen in this normal examination.
The exam was cardiac gated.

Figure 27-24. TONE (tilted optimized nonsaturating excitation). In

this scheme, a ramped flip angle is used, which is larger in deeper
slices. Because a larger α yields a larger transverse component in
the x-y plane, this scheme would improve SNR in deeper slices,
counteracting the saturation effects of slow-flowing blood.
 Question: Why do MRA techniques overestimate the degree of stenosis?
Answer: Because accelerated flow through the stenotic area leads to
dephasing during TE. To reduce this effect, use a shorter TE. Also,
turbulent flow and vortex flow (flow eddies) as well as stream
separation distal to stenosis and at vessel turns (e.g., carotid siphon)
may cause dephasing and flow void, overestimating the length of
stenosis (in the case of post stenosis) or mimicking stenosis (in the case
of vessel turns). CE MRA minimizes these effects.

Black-Blood MRA
Black-blood MRA is a subset of TOF techniques, which accentuates the
TOF losses resulting in flowing blood appearing dark rather than bright.
Rapidly flowing blood (arterial flow), as discussed in the previous
chapter, demonstrates TOF signal losses. Slowly flowing blood (venous
flow) has higher intensity. Various flow presat pulses and dephasing
methods via gradients are employed in this technique to render flowing
blood black. Note that the maximum intensity projection is replaced by
a minimum intensity projection algorithm.

Advantages
1. This technique does not overestimate the degree of stenosis as much
as bright-blood TOF MRA.
2. Dephasing in vessel turns that mimic stenosis with bright-blood TOF
MRA is not a problem here.

Disadvantages
1. Calcified plaques may also be dark and therefore invisible. Thus, this
technique may underestimate the degree of stenosis.
2. Other black materials (such as air or cortical bone or calcification)
may mimic flow.

Fresh-Blood Imaging
 3D-fresh blood imaging (FBI) MRA is a new technique based on the fact
that vessel signal intensity is dependent on blood flow (or cardiac
phase) in T2-weighted images. The early systolic phases (0 ˜ 200 msec
after the R wave) show low signal intensity for arteries (high velocity
with TOF losses) and high signal intensity for veins (lower velocities
with no significant TOF loss), whereas the diastolic phases (400 ˜ 600
msec after the R wave) show high signal intensity for both arteries and
veins (no significant TOF loss). Bright-blood MRA is achieved by
subtracting systolic images from diastolic images. The 3D-FBI method
employs an electrocardiography (ECG)-gated 3D half-Fourier fast
spinecho sequence triggered for systolic and diastolic acquisitions. The
ECG triggering time is an important factor influencing the blood signal
intensity in the vessel of interest. An “ECG-prep scan” is typically used
to produce 2D half-Fourier FSE single-slice images at various triggering
delay times. An appropriate ECG delay time is determined for the
vessel of interest and later applied in the 3D half-Fourier FSE
acquisition synchronized by ECG gating for every slice encoding. An
example is seen in Figure 27-25.

Advantages
1. Non-contrast technique; gadolinium not required
 Figure 27-25. Coronal FBI MIP image of the lower extremity
shows abrupt occlusion of tibioperoneal trunk, peroneal and
tibial arteries, as well as geniculate collaterals.

2. Arterial and venous imaging in all areas of the body

3. Insensitive to off-resonance artifacts

Disadvantages
1. Sensitive to selection and timing of the triggered acquisitions
2. Motion artifacts and blurring
3. Longer scan time than CE-MRA

Key Points
1 Three main MR angiography (MRA) techniques
exist: (i) time-of-flight (TOF), (ii) phase contrast
(PC), and (iii) contrast enhanced (CE).
2 Both TOF and PC MRA can be performed in 2D or
3D.
3 TOF MRA is based on FRE.
4 PC MRA is based on velocity-induced phase
changes.
5 PC MRA techniques allow acquisition of magnitude
and phase images.
6 Phase images provide information regarding the
direction of flow (not provided by TOF MRA).
7 VENC (velocity encoding) is a parameter that needs
to be input by the operator when doing PC MRA. It
represents the maximum flow velocity before aliasing
occurs.
8 If VENC is too low, aliasing is more likely to occur.
If VENC is too high, slow flow and small vessels are
not well visualized. Therefore, low VENC is good for
visualization of slow (venous) flow and small
branches. High VENC is good for highvelocity
(arterial) flow.
9 CE MRA is based on a rapid 3D GRE sequence that
is timing sensitive and takes advantage of the
powerful T1 shortening of gadolinium.
10 CE MRA is the most resistant to dephasing of the
bright-blood techniques.
11 Time-resolved imaging of contrast kinetics
(TRICKS) provides rapid multiphase imaging while
retaining much of the SNR. TRICKS is useful for high-
flow malformations or differential flow conditions
(e.g., lower extremity run offs).
12 MIP (maximum intensity projection) is an
algorithm used in MRA in which the highest intensity
dots in space are connected to generate a three-
dimensional-looking image of the vessels.
13 3D imaging is subject to saturation effects (as
blood enters deeper slices).
14 Saturation effects can be minimized by several
methods: (i) decreasing the flip angle, (ii) increasing
TR, (iii) using gadolinium, (iv) MOTSA, or (v) TONE.
15 MOTSA (multiple overlapping thin-slab
acquisition) uses several overlapping thin slices to
reduce saturation effects. A potential problem is
Venetian blind artifacts.
16 TONE (tilted optimized nonsaturating excitation)
uses a ramped RF flip angle (small α at the entry
 slice and larger α at the exit slice) to reduce
saturation effects.
17 TOF MRA tends to overestimate the degree of
stenosis (due to dephasing effects).
18 Black-blood MRA is based on TOF signal losses.
Instead of maximum intensity projection, a minimum
intensity projection algorithm is used. This technique
overcomes the problem of overestimating the degree
of stenosis.
19 Fresh-blood imaging (FBI) is a new TOF bright
arterial signal MRA technique that takes advantage of
the differing velocities and TOF losses between
systole and diastole.

Questions
27-1 The main MRA techniques include
(a) TOF MRA
(b) PC MRA
(c) CE MRA
(d) all of the above

27-2 The parameter VENC must be set for

(a) TOF MRA
(b) PC MRA
(c) both
(d) none

27-3 Match (i) low VENC (ii) high VENC with

(a) more sensitive to slow flow
 (b) aliasing
(c) poor visualization of small branches
(d) arterial flow

27-4 T/F TOF MRA may underestimate the degree of stenosis.

27-5 Saturation effects can be reduced by

(a) decreasing the flip angle
(b) increasing TR
(c) using a gadolinium chelate
(d) MOTSA
(e) all of the above
(f) only (a)-(c)

27-6 T/F TOF MRA is based on FRE.

27-7 T/F MT allows better visualization of smaller vessels with slow

flow in the brain.

27-8 Match
i. ramped RF
ii. Venetian blind artifact
iii. deeper FRE
iv. magnitude and phase images
v. low VENC
with
(a) MOTSA
(b) TONE
 (c) aliasing
(d) countercurrent flow
(e) PC MRA

27-9 T/F In the MIP algorithm, the pixel with the highest intensity is
selected for each slice.

27-10 Compared with conventional brightblood MRA, black-blood

MRA
(a) uses a minimum intensity projection algorithm
(b) does not overestimate the degree of stenosis as much
(c) both (a) and (b)
(d) none of the above
28
Cardiac MRI

Introduction
Cardiac MRI is arguably the most difficult MRI examination to perform.
Cardiac MRI must overcome not just respiratory motion, but also
cardiac motion that cannot be suspended for the image. Additionally,
there are a variety of nomenclature approaches to the pulse sequences,
some of which are fairly unique to cardiac imaging while others are
used in other organ systems all contributing to difficulty in
understanding cardiac MRI. These different nomenclature systems are
superimposed on the option of static versus cine imaging, and
functional or physiologic imaging which all combine to challenge not
just the patient and MRI technologist, but the MR physicist and
radiologist.

Motion Effects and Compensation

Motion is a perpetual challenge to high-quality images in cardiac MRI.
Body imagers are very knowledgeable of respiratory motion in
abdominal imaging, but cardiac MRI has the additional challenge of
cardiac motion. Respiratory and cardiac gating techniques are used
frequently in cardiac MRI to compensate for motion. Gating techniques
use an electrical impulse based on a physiologic marker (e.g., R wave
from an electrocardiographic [ECG] or diaphragm position indicator) to
accept, reject, or reorder data in k-space contributing to an image.
Gating techniques fall into two broad categories: (i) prospective and
(ii) retrospective. Prospective gating or EKG- or plethysmographic-
triggering uses the impulse and based on previous preset or calculated
parameters determines prospectively how k-space will be filled prior to
signal acquisition. On the other hand, retrospective gating runs the
pulse sequence and collects the signal regardless of the electrical or
pressure impulse marker, and then either “on the fly” or after the
signals have all been obtained uses certain parameters to either accept
or reject signals for inclusion into k-space and subsequent Fourier
transformation.
Respiratory Motion. Respiratory motion can be compensated by either
breath-hold imaging or respiratory gating. The average individual can
breath-hold for 15 to 25 sec, while individuals with cardiopulmonary
disease will be able to breath-hold for shorter periods, which limits this
technique. Alternatively, respiratory gating techniques track the motion
of the diaphragm either indirectly by using a bellows device around the
chest/abdomen or directly through a navigatorecho pulse. Both
techniques track the depth and direction of the respiration and then
either eliminate or minimize k-space acquisition when the position of
the diaphragm falls outside of prescribed limits.
A bellows device can utilize either (i) respiratory gating (triggering) or
(ii) respiratory compensation (respiratory-ordered phase encoding
[ROPE] or centrally ordered phase encoding [COPE]) of k-space.
Respiratory gating can be either prospective or retrospective. Gating
will accept signals into k-space if the bellows detects the diaphragm
position at or near a predetermined position in the respiratory cycle
(usually

end-expiration). However, respiratory gating utilizes only about 20% of

the respiratory cycle since most of the time the diaphragm position will
lie outside the ideal position. Respiratory compensation (ROPE and
COPE) improves efficiency by reordering the sequence of k-space filling
based on the fact that the outer portions of k-space are less sensitive
to changes due to motion than the central portion. ROPE gradually
varies the phase-encode steps along the entirety of the respiratory
cycle, while COPE acquires the lower phase-ordered steps (central k-
space) around the longest respiratory cycle dwell time (usually end-
expiration) with ever-increasing higher phase steps the further away
from endexpiration.
Figure 28-1. Navigator-echo gating. A: Coronal True FISP scout
demonstrates column of excited tissue (rectangle) that the
navigator-echo pulse will use on the right hemidiaphragm. B: Axial
True FISP showing path of the intersecting RF pulses (arrows)
generating the navigator-echo column at the intersection (dashed
arrow). C: Tracing of navigator echo data showing the lung/liver
interface and movement of the diaphragm. Time along x-axis and
distance in millimeters along y-axis. Narrow window at “145”
demonstrates acceptance window.
The newest respiratory gating method is navigator-echo gating. This
technique does not require the bellows, but instead uses either a single
spiral radio frequency (RF) pulse or two intersecting RF pulses, usually
on the right diaphragm, to track movement. Navigator-echo gating data
then usually prospectively triggers the acquisition of kspace if the
object tracked is within a certain prescribed window of movement
(usually 3 to 5 mm) (Fig. 28-1). Newer variants of navigator gating

include a phase-reordered sequence that triggers the filling of the

center of k-space when the diaphragm is within the ideal range while
contributing to the outer portions of k-space when the diaphragm is
outside the ideal range. This adaptation fills k-space during more of the
respiratory cycle, resulting in a shorter scan time similar to respiratory
compensation techniques. Lastly, navigator information can be used
prospectively to correct for motion by adjusting the slice position if the
diaphragm is detected outside the preferred window. This technique is
known as slice-tracking or motion-correction.
Cardiac Motion. Cardiac motion is complex with various contributions
from longitudinal shortening (long-axis), radial contraction (shortaxis),
and rotational motion. Although cardiac motion can be compensated
for by using a pulse oximeter device or plethysmograph peripheral pulse
monitor, ECG gating is more precise and usually yields a superior result.
ECG gating allows the signal to be acquired in the same phase (e.g.,
systole and mid-diastole) of the cardiac cycle, resulting in less cardiac
motion. However, there are additional complexities of the cardiac cycle
with resultant R-R interval variability due to (i) normal beat-to-beat
variability, (ii) premature contractions, and (iii) changes due to
respiration especially breath-hold. Prior to scanning, the patient's
cardiac tracing is monitored and certain parameters are calculated
based on the patient's R-R interval range. The R-R interval range and
frequency of premature contractions vary between patients and, if
significant enough, can negatively impact image quality. ECG gating
approaches these limitations and variables in two fundamental ways: (i)
prospective and (ii) retrospective gating (Figs. 28-2 and 28-3).
Additionally, for highly variable R-R interval patients, an arrhythmia
reject window can be used in both gating techniques that further
prevents filling k-space if R waves fall too far outside expected
parameters. The arrhythmia reject window length may be either
symmetric or asymmetric around the expected R wave.

Figure 28-2. Prospective gating for a single slice/single phase

(mid-diastole) fast spin-echo sequence with an echo train length =
6. Notice the long trigger delay (TD) between the R wave and the
commencement of the pulse sequence with the initial 90° RF
pulse. Dashed boxes around the QRS waves indicate the built-in
arrhythmia reject window intrinsic to prospective gating.
 Figure 28-3. Retrospective gating for a single slice/multiphase cine
sequence using a GRE pulse sequence. Notice the oversampling of
the R-R interval of about 125% denoted by the different dashed or
solid lines denoting a complete train of RF pulses contributing to a
segment of k-space for the separate cardiac phases. There is no
trigger delay as in prospective gating. Notice that the TR is much
shorter in the GRE sequence than in the FSE sequence in Figure 28-
2. Lastly, the dashed box represents an optional asymmetric
arrhythmia rejection window (− 10% to + 50% expected R wave)
whereupon an R wave detected outside of this box would result in
rejection of the prior R-R interval's signals.

Prospective gating uses R wave detection with a variable trigger delay

and then begins collecting k-space. The k-space is then filled over a
certain prescribed percentage of the average R-R interval (usually 80%
to 90% for cine imaging) whereupon no k-space is filled during the last
10% to 20% of the R-R interval. This results in a constant number of
lines of k-space that are filled per R-R interval (Fig. 28-2). The last 10%
to 20% of the R-R interval that does not fill k-space is designed in
arrhythmia reject window that prevents acquisition of k-space during
an early R wave due to beat-to-beat variability. However, this may or
may not exclude premature contractions, and this can then be solved
by superimposing a broader arrhythmia rejection window. Prospective
gating is used in both static and cine imaging.
Retrospective gating, on the other hand, does not have any periods
within the cardiac cycle where k-space is not being filled. This
technique over samples the R-R interval usually at 125% and then
retrospectively goes back and based on the detected R waves
determines which line of k-space corresponds with each specific cardiac
phase (Fig. 28-3). This technique results in a variable number of k-
space lines filled per R-R interval since the R-R interval has beat-to-
beat variability and sometimes premature contractions. After the
collection of signals, the computer then uses a weighted interpolation
technique and determines the phase of the cardiac cycle that each
signal belongs to. This technique is most commonly used in cine
sequences such as gradient-recalled echo (GRE), true fast imaging with
steady-state precession (True FISP, Siemens), FIESTA (fast imaging
employing steady-state acquisition, General Electric), or b-FFE
(balanced fast field echo, Philips) and phase contrast imaging.
If arrhythmias are too frequent, then ECG triggering can become
impossible, and one can salvage the study by taking off the cardiac
gating and increasing the NEX to 4. The idea behind this is that systole
is typically a small

percentage of most patients' cardiac cycles, and thus by increasing the

NEX, it will average out the smaller percentage of the signal acquired
during systole with its greater motion. If the patient's heart rate is fast,
especially over 100, then this technique will be limited as well. The
drawback is that scan time will increase and breath-hold techniques
will be impossible, so a respiratory-gated technique must be used.
Lastly, a single-shot technique (discussed later) can be used.
Blood Motion. Blood movement within the heart and great vessels is
 pulsatile with complex flow characteristics, resulting in motion-induced
intravoxel dephasing from differing velocities within a voxel
contributing to signal loss. Valvular stenosis and regurgitation, vessel
stenosis, and cardiac shunts can all have marked effects on the blood
flow velocities, direction, and subsequent dephasing and signal loss.
Flow compensation or gradient moment nulling can help mitigate these
problems, but at a small sacrifice of a higher minimum time to echo
(TE).
Faster Imaging. The faster the imaging sequence, the less chance for
any type of motion-related effects. GRE or True FISP imaging with very
short TRs or half-Fourier acquired single-shot turbo spin-echo (HASTE)
sequences all acquire k-space in a shorter period of time versus fast
spin-echo (FSE) sequences. A motion-reducing option for GRE or True
FISP sequences is to acquire the k-space in a singleshot mode where
the entire image's k-space is acquired within a single R-R interval.
HASTE sequences are by nature a single-shot sequence.
Additionally, parallel imaging can be used with any pulse sequence (see
Chapter 24) and commonly reduces the time of acquisition by two to
fourfold or increases spatial resolution two to four times without a time
penalty. The major drawback to parallel imaging is decreased signal-to-
noise ratio (SNR), so parallel imaging works best with sequences that
have high SNR such as True FISP or postgadolinium imaging.
Motion in Cardiac Imaging and Solutions
1. Gross patient movement: instruct patient to lie still, mild sedation
2. Respiratory movement: breath-hold techniques, respiratory gating,
navigator-echo gating
3. Cardiac motion: ECG gating, pulse oximeter gating, increase NEX,
single-shot technique
4. Blood motion: flow compensation/gradient moment nulling, pulse
sequences insensitive to dephasing, for example, True FISP
5. Parallel imaging: two to fourfold decrease in acquisition time
however decreased SNR
General k-space Filling Strategies
Segmented versus Single Shot. Early in cardiac MRI, only one line of k-
space was filled for a single image during a single R-R interval. This was
a very inefficient way of filling k-space and resulted in long scan times.
Subsequently, all cardiac imaging fills k-space in a segmented fashion
that fills more than one line of k-space for a single image during a
single R-R interval. One can conceptualize k-space “segmented” into
separate areas based on different R-R interval contributions of k-space.
For example, if there were 160 phase-encode steps and eight lines of k-
space were filled for a single image during each R-R interval, then
there would be 160/8 = 20 segments of k-space. The eight lines of k-
space filled per R-R interval is termed views per segment (VPS). If all
k-space is filled in a single R-R interval, then this is equivalent to a
single segment, and it is referred to as a single shot. It should be noted
that a “single shot” outside of cardiac MRI usually refers to a pulse
sequence where a single RF pulse tips all the required longitudinal
magnetization into the transverse plane for k-space to be completely
filled for a single image such as in a HASTE or single-shot EPI sequence.

Static Imaging
All of the previously described pulse sequences in this book have been
or are used to produce static cardiac images. These images are usually
divided into two categories: (i) bright blood and (ii) dark blood.
Unfortunately, the appearance of blood as either bright or dark is a
consequence of the specific pulse sequence and not a particular
sequence one can select on

the scanner. Blood signal has dark T1 and bright T2 signal intrinsically.
Knowledge of the intrinsic signal of blood combined with superimposed
time-of-flight effects (time-of-flight [TOF] loss or flow-related
enhancement [FRE]) in a particular pulse sequence determines whether
blood is bright or dark.
Fast Spin Echo and HASTE. Spin-echo imaging has generally been
abandoned due to excessive scan times, and it is now replaced by
either FSE or HASTE that provides good anatomic detail, intrinsic dark-
blood signal due to TOF loss, and appropriate T1 or T2 weighting based
on TR and TE. FSE with ECG gating and either breath-hold or navigator-
echo gating result in good image quality; however, the drawback is
lengthy scan times. HASTE sequences have shorter scan times and are
usually obtained in a single R-R interval; however, the SNR will be less
due to 1/2NEX signal averages.
Although FSE and HASTE sequences are usually dark blood, some areas
of slow blood flow from physiologic, pathologic, or in plane flow can
result in minimal expected TOF loss and unexpected bright-blood signal
confounding interpretation. This bright signal can be minimized by
using a double-inversion recovery (DIR) sequence. This sequence uses
a nonslice-selective 180° RF pulse followed immediately by a slice
selective 180° RF pulse and an appropriate T1 that is influenced by the
R-R interval (T1 usually 600 msec) that nulls the signal of inflowing
blood, resulting in a more robust black blood effect (Figs. 28-4 and 28-
5). Due to some intrinsic changes in the position of the slice from the
initial 180° RF pulses and the actual filling of k-space about 600 msec
later, the slice-selective 180° RF pulse is usually twice the nominal
thickness of the slice to ensure that all the myocardium is reinverted.
Additionally, to allow full recovery of the longitudinal magnetization, k-
space is usually filled every other R-R interval unless the patient is
bradycardic.
 Figure 28-4. Prospective gated single slice/single phase (mid-
diastole) fast spin-echo DIR sequence with an echo train length = 6.
Notice the shorter trigger delay (TD) driven by the requirement to
initiate the pulse sequence with the nonselective 180° inversion
pulse (dashed line in pair) immediately followed by a slice-
selective 180° inversion or “reinversion” pulse (solid line in pair).
This schematic shows acquisition of echoes every R-R interval,
which occurs only in bradycardic patients. The R-R interval is
usually shorter requiring echo acquisition every other R-R interval
or heartbeat.

Fast Spin-Echo Gating. Static imaging including FSE uses prospective

gating. Optimal gating with minimal motion artifact is achieved when

each image is acquired in the same cardiac phase. Static images are
either acquired as a single slice/single phase (usually mid-diastole) or a
multislice/single phase where multiple slices are acquired in an
interleaved fashion. The single image/single phase technique fills k-
space during mid-diastole and results in less motion since it minimizes
changes of cardiac phase due to beat-tobeat variability; however,
usually only one slice or at most two slices can be acquired during a
single breath-hold (Fig. 28-2). Multislice/single phase is more efficient,
but has more motion effects due to beat-to-beat variability and
misregistration between slices due to the changing position of the heart
at different phases of the cardiac cycle.
Figure 28-5. A: Direction of longitudinal magnetization
immediately after the nonselective and slice-selective 180° RF
pulses. These pulses are usually applied during late diastole
immediately after the R wave is detected. Solid arrows represent
longitudinal magnetization in the myocardium, while dashed lines
 represent blood. B: Direction of longitudinal magnetization
immediately before the acquisition of echoes and filling of k-space.
The TI is usually chosen to coincide with acquisition of k-space
during mid-diastole. Solid arrows represent amplitude of
longitudinal magnetization in the myocardium, while dashed lines
represent blood. Notice that the dashed arrows that were in the
imaging slice in (A) have now moved to the left with the direction
of flow and will not give off signal due to their movement outside
the imaging slice. The prior dashed arrows that were inverted have
now become nearly zero in longitudinal magnetization (gray dashed
arrows).

All FSE sequences have time of repetition (TR) that approximates or

exceeds the R-R time. This requires the TR for an FSE sequence to be a
multiple of the patient's R-R interval. For example, if a patient's pulse
is 75 beats/min, then the R-R interval is 800 msec [(60 sec/min)/(75
beats/min) = 0.8 sec/beat], and the TR would have to be a multiple of
800 msec. A T1 FSE TR is equal to 800 msec, and a T2 FSE TR would be
2400, 3200, or 4000 msec. Faster sequences with TRs that are much
shorter than the R-R interval such as GRE and True FISP sequences are
not tied to this principle and thus can keep their best TR for the pulse
sequence (see later discussion).
Gradient-Recalled Echo. A faster alternative to FSE imaging for static
images is GRE sequences. These sequences may not give quite the T1 or
T2 weighting quality of an FSE sequence, but they are acquired more
quickly. Additionally, spoiled GRE sequences typically have bright-blood
signal due to FRE (see TOF MRA in Chapter 27). Since GRE sequences
are dependent on FRE, ultrashort TRs are not practical because the TR
must be long enough to allow unsaturated protons to enter the imaging
slice (Fig. 28-6). Flow compensation techniques are customarily used
during bright-blood GRE techniques to maintain as much signal as
possible. Postgadolinium GRE sequences can be performed, which
further increase the blood

signal; however, the contrast between the blood pool and the
myocardium will be decreased due to the enhancing myocardium.

Figure 28-6. Four chamber bright-blood GRE sequence TR 7.8/TE

4.3 msec of the heart demonstrates bright-blood signal in the heart
and great vessels. Area of darker signal along left ventricular free
wall (arrow) was a pericardial hematoma. Also note the
susceptibility effects from midline sternotomy wires (dashed
arrow).

True Fast Imaging with Steady-state Precession (True FISP). The

current workhorse of fast cardiac imaging is the True FISP (Siemens)
sequence, which is also known as FIESTA (General Electric), b-FFE
(Philips), or sometimes generically as b-SSFP (balanced steady-state
free precession). True FISP utilizes a high flip angle, very short TR/TE,
and a completely balanced pulse sequence that preserves transverse
magnetization, resulting in extremely fast T2/T1-weighted high SNR
images with insensitivity to dephasing and signal loss (Figs. 28-7 and 28-
8). True FISP does not rely on FRE for bright-blood effects, but rather
on the intrinsic contrast difference in T2/T1 signal of blood versus
myocardium. True FISP's TR/TE is usually 3 to 4 msec/1.5 to 2 msec
enabling this sequence to be performed faster than most other pulse
sequences.

Figure 28-7. True FISP pulse sequence diagram demonstrates

completely balanced RF pulses and magnetic gradients. The time
to echo (TE) is exactly ½ of the time of repetition (TR).

GRE and True FISP Gating

GRE and True FISP TR times are much less than an R-R interval time,
and thus their optimal TR can be used regardless of the R-R time.
Prospective gating is frequently used with GRE/True FISP static imaging
obtained in middiastole. True FISP's TRs are 3 to 4 msec allowing for a
single slice/single phase image acquired in a single R-R interval, also
known as a single shot. GRE sequences have longer TRs and may require
two R-R intervals to create a single image during mid-diastole.
Alternatively, prospective or retrospective gating can be used for a
multislice/single phase acquisition in an interleaved method. This will
result in several different slices obtained at various times in the cardiac
cycle.

Figure 28-8. Vertical two chamber or left ventricular outlet view

(same patient as in Fig. 27-7) demonstrates bright signal not only in
the blood, but also within the mediastinal and abdominal fat due
to the intrinsic T2/T1 bright signal. Patient had significant aortic
stenosis, which is seen here as mild signal loss from marked
turbulent flow (arrow). A standard bright-blood GRE sequence
would have more signal loss.
Inversion Recovery Preparation GRE/True FISP or
Delayed Enhancement Sequences
A 180° RF inversion recovery (IR) preparation pulse can be applied to a
pulse sequence following gadolinium administration, usually either a
GRE or a True FISP sequence, to take advantage of the fact that there
are different temporal T1 postgadolinium curves for normal
myocardium versus scar. These IR-prepared GRE or True FISP sequences
are frequently referred to as delayed enhancement or late gadolinium
enhancement images and have proven very useful in diagnosing
abnormal myocardium due to ischemic events, a variety of
cardiomyopathies (CM) (e.g., sarcoid, right ventricular dysplasia,
myocarditis, amyloidosis, or hypertrophic CM), or masses. These pulse
sequences are performed following a 10 to 15 min delay after
intravenous gadolinium that results in selective nulling out of normal
enhancing myocardium while abnormal myocardium remains bright (Fig.
31-3). These IR-prepared GRE or True FISP sequences have a single IR
pulse for a whole train or series of TRs (usually 20 to 25). Thus, there is
a sliding time to inversion (TI) for each signal acquisition (see Chapter
21) which is analogous to the “TE effective” in FSE imaging with the
central lines of k-space acquired around the desired TI (usually
between 200 and 300 msec) to optimize the contrast contribution of
the nulled normal myocardium at the optimal TI (Fig. 28-9). Again there
is some cardiac motion during the TI; therefore, the 180° RF pulse is
nonslice selective to ensure complete nulling of all myocardium.
Finally, k-space is usually collected every other heartbeat allowing
enough time for full longitudinal magnetization recovery.
A frequently encountered difficulty with IRprep GRE/True FISP
sequences is poor nulling of normal myocardium on later images. This is
due to gadolinium continually washing out of tissues, resulting in a
longer T1 times and a subsequent need for longer TI times with later
images. A phase-sensitive version has recently been introduced to
mitigate this problem. Phase-sensitive IR-prepared GRE sequences
restore the polarity of the magnitude of the signal, which increases the
contrast between abnormal and normal enhancing myocardium (Fig. 28-
10).

Cine Imaging
Cine imaging is similar to static imaging; however, instead of getting a
single image for a single slice, you obtain a series of images obtained at
different phases within the cardiac cycle for a single slice resulting in a
single slice/multiphase acquisition. This permits visualization of the
motion of the heart, valves, and any abnormal structures within the
heart. GRE and True FISP sequences

provide this capability since FSE sequences take too long to acquire the
required multiple phases per slice. In cine imaging, the user defines
how many phases within the cardiac cycle per slice (usually 15 to 25)
are acquired. The goal for temporal resolution between different
phases should be around 50 msec. For example, a patient with an R-R
interval of 1000 msec (60 beats/min) with a cine sequence with 20
phases will have a temporal resolution of 1000 msec/20 phases = 50
msec/phase. The benefit of increased phases is better temporal
resolution and more precise physiologic calculations (see later
discussion) with smoother cine loop viewing; however, the drawback is
that it will take longer to scan. The relationship between phases, heart
rate, and views per segment is given below.
Figure 28-9. The multiple TRs in the IR-prepped GRE or True FISP
delayed enhancement sequences are fewer in this example for
clarity, but usually number 20 to 25; in the case of a single-shot
sequence, the number of TRs would equal all the lines of k-space.
The time to inversion is variable or sliding along the train of TRs;
however, the nominal TI is in the middle of the TR train where the
lowest phase-encode step is applied. The solid line in the
longitudinal magnetization (Mz) indicates abnormal enhancing
myocardium, while the dashed line reflects normal myocardium.
Notice that ideally the normal myocardium crosses the null point
exactly in the center of the train of TRs. Finally, there is no
acquisition of k-space in the next R-R interval due to the need to
allow time for longitudinal magnetization recovery.
Example: How long will it take to perform a prospective cine sequence
using a True FISP sequence with 20 phases in the cardiac cycle, R-R
interval, or time of segment acquisition = 1000 msec with a 20%
arrhythmia reject window (800 msec of scanning per R-R), TR = 4 msec,
and 160 phase-encode steps or views, and NEX = 1.
800 msec/20 phases = 40 msec/phase (temporal resolution)
(40 msec/phase)/(4 msec/TR) = 10 TRs or 10 views/segment
160 views/10 views/segment = 16 segments
Total scan time = (16 segments)(1000 msec/segment) = 16,000 msec
or 16 sec
If the TR is longer (GRE for instance) or if more phases are required,
then the total scan time will

also be longer which may fall outside a patient's breath-hold ability. To

solve this problem, another technique known as view sharing can be
used. View sharing is similar to the “shared echo” approach in FSE (see
Chapter 19) in that echoes or views are shared between the phases.
View sharing typically shares 50% of the echoes, and thus the new
phases will be double minus one (Fig. 28-11).
 Figure 28-10. Short-axis segmented IR-prep GRE delayed
enhancement imaging after gadolinium in a patient with
myocarditis. A: Amplitude image demonstrates mildly bright
midmyocardial signal in the lateral wall (arrow). Notice sparing of
subendocardium consistent with myocarditis. B: Phase-sensitive
reconstruction demonstrates increased contrast between the
abnormally enhancing areas of myocarditis (arrow) in the lateral
wall and the normal myocardium (compare best with
interventricular septum).

Cine Gating Techniques

Cine imaging can be acquired with either prospective or retrospective
gating. Prospective techniques are generally faster; however, they omit
filling the last 10% to 20% of the cardiac cycle or late diastole and thus
result in incomplete cardiac data that is especially important for
determining physiologic or functional imaging (see later discussion).
Retrospective gating is used when all phases of the cardiac cycle are
required to be imaged.
Myocardial Tagging. An option in cine imaging used for enhanced
detection of wallmotion abnormalities is the application of a tagging-
prep pulse usually used with a GRE or True FISP sequence. The tagging
bands are generated by a series of RF pulses separated by magnetic
gradients that result in SPAtial Modulation of the Magnetization or
SPAMM. The resultant image has a series of parallel dark lines or a dark
grid of lines (Fig. 28-12). These lines or grid intersections are then
subsequently deformed or move with contraction of the myocardium,
but areas of hypo- or akinesis of the myocardium have persistent
straight lines. A limitation of SPAMM is that these saturation bands fade
after around 400 msec. A newer technique called Complementary
SPAtial Modulation Magnetization (CSPAMM) improves the duration of
the saturation bands.
Perfusion Imaging. Perfusion imaging is a subtype of cine imaging. This
sequence uses a first pass of gadolinium that enters the blood pool and
right ventricle initially followed very rapidly in sequence by filling of
the left ventricle, coronary arteries, and perfusion of the myocardium.
This is a cine technique, but due to the single pass nature of the
gadolinium, a multislice/multiphase acquisition is required (Fig. 31-6).
This requires very fast imaging sequences in order to achieve both the
temporal and spatial resolution. Perfusion

imaging is typically done both at rest and at stress (usually using

adenosine) to determine areas of decreased perfusion related to
ischemia. The profound T1 shortening effects of gadolinium allow for
shorter TRs in GRE sequences since we no longer rely on FRE effects,
but rather the intrinsic T1 shortening properties of enhancement due to
the gadolinium. Fast single-shot GRE with minimum TR, a hybrid
segmented GRE-EPI or True FISP sequence is used with a saturation or IR
prep pulse to suppress the signal from background structures and to
provide more T1 weighting.

Figure 28-11. Prospective gating of GRE or True FISP sequence

without and with view sharing. Notice that the trigger delay is
nearly 0. The “dead” time or intrinsic arrhythmia rejection window
(dashed line) is characteristic of prospective gating where no signal
is acquired at the end of a predetermined number of echoes. A:
Notice that with four views per segment only eight phases can be
imaged without view sharing. B:With 50% view sharing, the number
of phases increased from 8 to 15. The phase increase is double the
original phases minus one for 50% view sharing.

Figure 28-12. Myocardial tagging using a True FISP sequence. A:

End-diastolic short-axis image with straight grid lines in the heart.
B: End-systolic image with deformed grid lines consistent with
movement of the myocardium. Also notice the darkness of the grid
lines has decreased from (A) due to some T1 relaxation.
 Figure 28-13. Quantitative left ventricular parameters. Upper two
images from short-axis True FISP cine sequence with epicardial and
endocardial contours drawn. Table below indicates ejection
fraction, volumes, and cardiac output. The addition of epicardial
contours allowed calculation of myocardial mass as well. Notice
low ejection fraction.

Functional Imaging. Calculation of functional or physiologic parameters

is a common request either to confirm echocardiography abnormalities
or to evaluate parameters that echocardiography has difficulty with.
Functional or physiologic parameters are obtained in one of two ways:
(i) derived from bright-blood cine images such as True FISP, GRE, or
tagged variants, or (ii) phase contrast imaging (details in Chapter 27).
 A stack of short-axis bright-blood cine images are used in determining
ejection fractions, myocardial mass, and volumes (Fig. 28-13).

However, these cine images must be manipulated on a postprocessing

workstation either one that comes with the scanner or a stand alone
postprocessing workstation equipped for cardiac MR calculations.
Identifying end-diastole and end-systole, and then delineating
endocardial contours provides ejection fraction and volume
measurements. Addition of an epicardial contour to the endocardial
contour also provides myocardial mass and contractility measurements.
Phase contrast imaging can determine velocities, regurgitant fractions,
and shunt or Qp/Qs ratios (pulmonary to systemic flow ratio). Phase
contrast imaging by its very nature provides velocity and direction data.
Determining an appropriate velocity encoding or VENC value is essential
to provide the best imaging quality while eliminating aliasing. Phase
contrast imaging is typically, but not always, performed orthogonal to
the vessel or chamber evaluated.

Clinical Applications and Pulse Sequences

1. Infarction or cardiomyopathy: cine True FISP and delayed
enhancement IRprepared GRE or True FISP
2. Arrhythmogenic right ventricular dysplasia: cine True FISP, T1 with
and without fat saturation (DIR preferred), and delayed enhancement
IR-prepared GRE or True FISP
3. Mass: cine True FISP, T1 with and without fat saturation (DIR
preferred), T2 (DIR preferred), T1 postgadolinium, and delayed
enhancement IR-prepared GRE or True FISP
4. Coronary MRA: True FISP, GRE, or postgadolinium MRA for bright
blood; DIR FSE for black blood
5. Valvular disease: GRE or True FISP, phase contrast with calculations
for maximum velocity and regurgitant fraction, and DIR FSE for valve
anatomy
Key Points
1 Cardiac and respiratory motion adversely impact
cardiac MR images.
2 Respiratory motion is compensated for by using a
bellows or navigator-echo pulse both of which track
the motion of the diaphragm.
3 Cardiac motion is compensated for by ECG gating.
4 ECG gating can be either prospective or
retrospective.
5 Prospective gating is faster but omits 10% to 20%
of the cardiac cycle.
6 Retrospective gating is slower but covers the entire
cardiac cycle.
7 FSE/HASTE sequences provide dark-blood static
imaging for morphology.
8 Most FSE/HASTE sequences improve dark-blood
effects by employing a doubleinversion-prep pulse.
9 IR-prepared GRE or True FISP delayed
enhancement sequences demonstrate infarcted
myocardium and are useful in demonstrating various
cardiomyopathies and even tumors.
10 GRE and True FISP sequences provide bright-
blood appearance for either static or cine imaging.
11 Functional imaging can be obtained either
through postprocessing of cine imaging (ejection
fraction, volume, and myocardial mass) or
performance of phase contrast imaging (velocity and
shunt data).
Questions
28-1 Which techniques are useful for respiratory compensation?
(a) Navigator echoes
(b) Bellows
(c) Breath-holding
(d) All of the above

28-2 Which technique does not typically have “bright blood”?

(a) FSE
(b) True FISP
(c) GRE
(d) Postgadolinium GRE

28-3 Which gating technique best evaluates the entirety of the

cardiac cycle in a cine sequence?
(a) Prospective gating
(b) Retrospective gating
(c) Neither, they are equivalent

28-4 To calculate myocardial mass, what contours need to be drawn?

(a) Endocardial
(b) Epicardial
(c) Both

28-5 True FISP sequences cannot produce which of the following?

(a) Cine images
(b) Static images
(c) Ejection fractions
 (d) Flow velocities

28-6 Which sequence is usually obtained in a “single shot”?

(a) DIR FSE
(b) Cine True FISP
(c) HASTE
(d) Phase contrast

28-7 What feature improves temporal resolution?

(a) Lower TR
(b) Higher views per segment
(c) Shorter R-R interval
(d) Higher phase-encode steps
29
MR Spectroscopy in the Brain

Overview
Proton magnetic resonance spectroscopy (MRS) is an MR-based chemical
analytical technique that can be performed on any high field magnet.
Rather than providing images, it usually provides spectra consisting of
individual peaks, the chemical shift of which from a universal standard
helps identify the species comprising the peak. The normal spectrum
from gray matter is different from the normal spectra from white
matter (Fig. 29-1). MRS is the equivalent of the nuclear magnetic
resonance (NMR) used in organic chemistry—only now performed on a
patient in a whole-body MR magnet.
MRS is a functional technique. It can identify abnormalities not seen on
conventional MRI, for example, invasion of a glioblastoma multiforme
(GBM) into adjacent brain where there is no enhancement or T2
abnormality on MRI. It can distinguish between two or more
abnormalities that appear the same on MRI, for example, recurrent
GBM versus radiation necrosis.
As in NMR, the area under a given peak is proportional to the number of
protons contributing to the peak. So, taking ethanol (CH3— CH2—OH) as
an example, the area under the methyl (CH3) peak would be 3 (in
relative units), the area under the methylene (—CH2—) peak would be
2, and the area under the hydroxyl (—OH) peak would be 1. Since the
peaks generally have the same shape, this ratio of peak areas also holds
for peak heights. In general, peak height is proportional to the
concentration of a given species. MRS requires a species to be present
in at least a 1 mM concentration to be seen.
The above discussion is somewhat simplified. In actuality, the protons
on adjacent carbons influence the net magnetic field on each other. So
the magnetic field experienced by the methyl protons in ethanol would
be influenced by the adjacent methylene protons. In quantum
mechanical sense, these two protons can either both point up, both
point down, or one up and one down or one down and one up. When
they both point up the net magnetic field experienced by the adjacent
methyl protons is slightly higher than B0 and when they both point
down, it is slightly lower than B0, which shifts the peaks. Up-down and
down-up cancel each other and results in the same peak position as one
would have without any adjacent protons. This is known as “spin-spin
splitting,” and it results in a big peak with area or peak height of 3 to
be split into three peaks (called a “triplet”) of area 1-2-1 (Fig. 29-2).
Spin-spin splitting or “J-coupling” smears the chemical shift, making
peak identification more difficult. It increases with increasing TE.
MRS in the brain is generally performed in conjunction with MRI which
identifies the abnormality. For single voxel techniques, a volume of 8 cc
is generally recommended at 1.5 T. A smaller voxel can be used but the
signal-to-noise ratio (SNR) will be reduced (Fig. 29-3). Since peak height
is generally proportional to field strength, a smaller voxel can be used
at 3 T, reducing partial volume averaging. Generally, a second spectrum
is acquired in the contralateral normal brain.
 Figure 29-1. Normal spectra for white matter (1) and gray matter
(2). Note: the higher Cho level in normal white matter (third
highest peak).

Figure 29-2. Spectrum from ethanol. Note: spin-spin splitting of

methylene and methyl peaks.

Occasionally, the voxel is not placed on a lesion but in gray matter or

white matter. Figure 29-1 shows the recommended voxel placement and
resulting normal spectra. For gray matter, it is both calcarine cortices,
spanning the posterior interhemispheric fissure. A standardized white
matter voxel is in the high parietal lobe.
Once the voxel is placed on a lesion, the magnetic field in that voxel is
made more uniform or “shimmed.” Shimming results in improving the
uniformity from 1 part per million (ppm) in the main magnetic field to
0.1 ppm inside the voxel (Fig. 29-4). For multivoxel techniques, the
shimming must be performed throughout a slice, which now provides
multiple 1 cc voxels.

Figure 29-3. Signal-to-noise ratio as a function of voxel volume.

In addition to shimming, the technologist must also suppress the signal

coming from water—which has 100,000 times more signal than that
coming from the metabolites comprising the peaks in the spectrum.
Water suppression is accomplished by exposing the voxel to an RF pulse
that is 4.7 ppm from the universal standard, which, by definition, is at
0 ppm. This universal standard is TMS (tetramethylsilane). Having 4
methyl groups (12 protons) all in the same chemical environment gives
it a peak height of 12! Peak position is always expressed in ppm relative
to TMS because it doesn't change

between field strengths, whereas the actual frequency shift would go

linearly with field strength. Thus data obtained at 1.5 T can be applied
at 3 T as long as it is expressed in ppm.

Figure 29-4. Effect of shimming. When full width at half maximum

(FWHM or W1/2) is 6 Hz (0.1 ppm at 1.5 T Larmor frequency of 64
MHz), Cr (3.0 ppm) can be distinguished from Cho (3.2 ppm). When
the shimming is only to 0.25 ppm (15 Hz at 1.5 T), the Cr and Cho
peaks start to merge.
MRS requires setting the repetition time (TR) and the echo delay time
(TE) prospectively. The shorter the TR, the more T1-weighted the
acquisition and the shorter the acquisition time, as in MRI. Choline
(Cho) has a shorter T1 than the other species so it will appear relatively
higher at shorter TR (Fig. 29-5). As Cho is used to diagnose cancer, it is
important to standardize the TR. The longer the TE, the more T2-
weighted the acquisition and the lower the SNR of the peaks (Fig. 29-
6). Most institutions use a TR of 1500 msec and the shortest possible TE
of 30 or 35 msec to maximize the SNR. This also allows the detection of
short T2 species (like myoinosotol [mI] and lipid), which would
otherwise have already decayed at longer TE.
Peak width is proportional to 1/T2, thus short T2 hemorrhage will lead
to peak broadening (Fig. 29-7). Peak broadening can also be seen with
motion and with a poorly shimmed voxel (Fig. 29-4).
Figure 29-5. Effect of TR. At shorter TR (1500 ms), Cho is
relatively increased compared to the other peaks due to its shorter
T1. At longer TR (5000 ms), all the peaks have greater signal-to-
noise ratio and have “caught up” to Cho similar to a proton
density-weighted MR image.
Visible Metabolites
Spectra are read from right (0 ppm) to left (4 ppm). As noted above the
peaks are identified on the basis of their chemical shift (Fig. 29-1). In a
normal brain, the first peak encountered is the main peak of N-acetyl
aspartate (NAA) at 2 ppm. This is a marker of normal neurons. It is
reduced whenever normal brain is displaced by a spaceoccupying lesion
or in diffuse processes like dementia or diffuse axonal injury.
There are small peaks on the left-hand shoulder of NAA at 2 to 2.5 ppm
(Fig. 29-1). These consist of glutamate (Glu) and glutamine (Gln) as
well as some secondary NAA peaks. On commercial systems with a
minimum TE of 30 msec, Glu and Gln cannot be distinguished due to
spin-spin splitting so they are collectively called “Glx.” Glx is elevated
in ischemia and hepatic encephalopathy.
The next peak in a normal brain is creatine (Cr) at 3 ppm. It is a marker
of energy metabolism and is the most constant of the peaks. Thus peak
heights can be compared relative to Cr when reading spectra. Cr is
reduced in near-drowning.
 Figure 29-6. Effect of TE and technique. PRESS (based on spin
echo) has twice the SNR of STEAM (based on stimulated echo) at a
given TE. However, with longer TE, there is more T2 decay and
lower SNR.

Cho is arguably the most important peak because it is elevated in

cancer due to membrane turnover (Fig. 29-8). It occurs at 3.2 ppm.
On short TE (30 to 35 msec) techniques, mI occurs at 3.5 ppm. It is a
sugar that appears to be the “prime osmolyte.” Although all these
species contribute to the osmolality of the brain, mI is the first to
become elevated when the serum becomes hyperosmolar, limiting the
amount of water shifted from the brain to the bloodstream. mI is also
elevated in glial tumors compared to NAA (Fig. 29-9) as well as in
Alzheimers.

Figure 29-7. Effect of magnetic susceptibility. A: EPI diffusion

image shows low signal due to short T2 deoxy-hemoglobin from an
acute hemorrhage. B: Spectrum shows extreme peak broadening
(FWHM is proportional to 1/T2).

Finally, there is a secondary Cr peak at 3.9 ppm. If the baseline rises on

the left side of the spectrum, this is usually a sign of incomplete water
suppression as water resides at 4.7 ppm.
Figure 29-8. A: Nonspecific T2 prolongation in 39-year-old woman
with fluctuating visual symptoms for several weeks, suggesting
ischemia. B: MRS shows elevated Cho/NAA, suggesting tumor.
Biopsy was positive for anaplastic astrocytoma.
Figure 29-9. Radiation necrosis and recurrent GBM. A: Enhancing
mass shows no elevation of Cho in anterior portion (B) but
elevation in posterior portion (tallest peak on the left hand side of
the spectra) (C).
 Figure 29-9. (continued) Four months later, the posterior portion
has increased in size, supporting the diagnosis of recurrent GBM (D
and E).

In certain disease states, for example, radiation necrosis, lipid is

elevated due to membrane destruction (Fig. 29-9B). Since there are
many lipid species, they have a range of chemical shifts from 0.9 to 1.2
ppm. Lipids also tend to have short T2s so they are only seen on short
TE techniques.
Lactate is adjacent to lipid at 1.3 ppm. It is elevated in ischemia, for
example, stroke (Fig. 29-10), and when anaerobic glycolysis occurs in
tumors that have outgrown their blood supply (Fig. 29-11). When it is
necessary to separate lactate from lipid, an intermediate TE of 135 to
144 msec can be used to flip the lactate doublet upside down using the
standard PRESS (point-resolved spectroscopy) technique (Fig. 29-11B).
Figure 29-10. A: Infarct in azygous anterior cerebral artery (ACA)
supplying both ACA territories. B: Elevated Lactate (doublet at 1.3
ppm on TE = 270 ms PRESS).
 Figure 29-11. A: TE = 35 ms PRESS shows positive lactate doublet.
B: TE = 144 ms PRESS shows negative lactate doublet in this patient
with GBM (Note: elevated Cho/NAA).

Patterns of Disease
Solid tumors have elevated Cho and decreased NAA. When they become
necrotic, lipid and lactate become elevated as well (Fig. 29-11).
Although elevated Cho/NAA is a sensitive marker for cancer, it is
nonspecific. The same pattern is seen in tumefactive multiple sclerosis
and acute disseminated encephalomyelitis.
MRS can distinguish recurrent GBM (elevated Cho/NAA) from radiation
necrosis (no elevated Cho but elevated lipid) (Fig. 29-9B). It can
distinguish necrotic lymphoma (elevated Cho) from toxoplasmosis (no
elevated Cho).
By placing the voxel outside the area of enhancement or T2
prolongation, MRS can distinguish a metastasis (no elevated Cho) from
an infiltrating glioma (elevated Cho) (Fig. 29-12).
Figure 29-12. A-C: Gastric metastasis. Cho/Cr is not elevated
outside area of enhancement. D-F: GBM. Cho/Cr is elevated both
in (D) and outside (E) enhancement (Cho is the tallest peak in the
spectra; compare with normal Cho/Cr in F). Elevated Cho/Cr in (E)
 represents infiltration of tumor outside area of enhancement.

Key Points
1 MRI images any mobile proton-containing species.
MRS shows up to 10 different proton-containing
chemical species, allowing more specific diagnoses.
2 MRS shows abnormalities, for example, infiltration
of a GBM, not seen on MRI.
3 J-coupling decreases peak height linearly with TE,
for example, glutamate and glutamine, which are
seen jointly as Glx on the left-hand shoulder of NAA
(2.0 to 2.5 ppm).
4 Species are identified based on their chemical shift
in parts per million (ppm) relative
P.351
to tetramethylsilane at 0 ppm (by definition).
5 In a normal brain, NAA (2.0 ppm) indicates normal
neurons, Glx (2.0 to 2.5 ppm) represents glutamate
and glutamine, Cr (3.0 ppm) represents energy, Cho
(3.2 ppm) represents membrane turnover, and mI
(myoinosotol at 3.5 ppm) is elevated in glial tumors
and hyperosmolar states. Choline is elevated in all
tumors and a few other states, for example, multiple
sclerosis. Lactate (1.3 ppm) indicates anaerobic
glycolysis seen in stroke and GBMs. Lipid (0.9 to 1.2
ppm) is seen with membrane breakdown seen in
infection, radiation necrosis, and tumors.

Questions
29-1 Which of the following metabolites is normally not seen?
(a) NAA
(b) Lactate
(c) Cho
(d) Cr
(e) mI

29-2 Which of the following metabolites is not seen at long TE?

(a) mI
(b) NAA
(c) Cho
(d) Cr
(e) Lac

29-3 Which metabolite is best to distinguish recurrent GBM from

radiation necrosis?
(a) NAA
(b) Cr
(c) Cho
(d) mI

29-4 Which metabolite is the best to distinguish necrotic lymphoma

from toxoplasmosis in AIDS?
(a) NAA
(b) Cr
(c) Cho
(d) mI

29-5 mI is elevated in
 (a) Alzheimers
(b) Hyperosmolar states
(c) Glial tumors
(d) All of the above

29-6 T/F Elevated Cho/NAA is specific for tumor.

29-7 T/F Performing MRS after giving gadolinium raises the Cho/NAA
ratio.

29-8 Peak broadening in MRS is caused by

(a) Motion
(b) Short T2 hemorrhage
(c) Local metallic artifact
(d) Poor shimming
(e) All of the above
30
High Performance Gradients

Introduction
This chapter briefly discusses the new technology of high performance
gradients. As you know by now, gradients have many purposes,
including slice selection, spatial encoding, flow compensation (FC), and
spoiling, rewinding, and presaturation. It should be fairly obvious that
every time you use a gradient, you are lengthening the pulse cycle
(thus increasing minimum TE).
Consider the two gradients in Figure 30-1A and 30-1B. The gradient in
Figure 30-1A has half the strength of the one in Figure 30-1B but twice
the duration. Thus, both these gradients have the same area (shaded
area). They both achieve exactly the same result (e.g., phase shift) on
stationary spins, but the second one is twice as fast and allows one to
reduce the echo delay time TE. Therefore, the first requirement of a
high performance gradient is a higher maximum strength.
When we discuss high performance gradients, not only do we want to
achieve a stronger magnitude or strength, but we also want the
maximum strength to be achieved in as short a time as possible (i.e., a
short rise time) to minimize the duration of the gradient. Therefore,
the second issue is how fast a gradient can reach its plateau (Fig. 30-2).
The ratio of the maximum gradient (Gmax) to the rise time (tR) is called
the slew rate (SR).

Early gradients had a Gmax of 3 to 6 mT/m and a tR of 1.5 to 2 msec

(i.e., SR of 1.5 to 4 mT/m/msec).

The requirements for high performance gradients are

 the following:
1. High gradient strength (Gmax)
2. Short rise time (tR)
i.e., a high slew rate (Gmax/tR).

In the mid-1980s, GE introduced shielded gradients with Gmax of 10

mT/m and tR of 675 µsec = 0.675 msec (i.e., SR of 15 mT/m/msec). The
new high performance gradient systems (Siemens' Sonata or Symphony
with Quantum Gradients, GE's Twin Speed, Philips' Intera, etc.) have
Gmax as high as 40 mT/m and tR as low as 180 µsec with SR as high as
200 mT/m/msec .

Advantages
1. Shorter cycles are possible. As discussed previously, a stronger
gradient can be applied over a shorter duration. Consider the
example of FC. Figure 30-3 demonstrates two FC gradients that
achieve the same thing, but the second one has higher strength (G′ >
G) and is thus applied over a shorter period (T′ < T). (As an aside,
because of the quadratic nature of phase accumulation for flowing
spins, there is no linear or inverse linear relationship between
gradient strength and application time, i.e., doubling the gradient
strength does not halve the time of gradient application.) Higher
order motion (e.g., acceleration, jerk) requires

the addition of even more gradient lobes. You can see how high
performance gradients can save a lot of time during each cycle,
allowing for shorter TEs (reducing dephasing) and TRs (for fast
scanning).
Figure 30-1. The gradient in (B) has twice the strength of the
one in (A) but half the duration. Consequently, they both have
exactly the same area and thus achieve the same results for
stationary spins. However, the one in (B) has the advantage of
being faster, a condition that is required for fast scanning.

Figure 30-2. In reality, it takes a certain amount of time (tR) for

a gradient to rise to its plateau Gmax. The ratio Gmax/tR is called
the slew rate (in mT/m/msec).
 Figure 30-3. With high performance gradients, flow
compensation can be accomplished faster (again because by
increasing the strength and reducing the duration, the same area
can be achieved).

2. Smaller field of views (FOVs) are possible. As we saw in Chapter 15,

an inverse relationship exists between the gradient strength along
one axis and the FOV: FOV = BW/(γG) where G is the gradient
strength and BW is the bandwidth defined as BW = 2fmax where fmax is
the Nyquist frequency (see Chapter 12). Thus, increasing G allows us
to reduce the FOV, leading to higher spatial resolution (an example is
high-resolution imaging of a small part such as the pituitary gland).
That is FOVmin = BW/(γGmax) Because spatial resolution is defined as
the FOV divided by the number of encoding steps: Spatial resolution =
 FOV/N then minimizing FOV (while keeping N fixed) results in
improved spatial resolution.
Example
For a standard high field system, the BW = 32 kHz (±16 kHz) and
Gmax = 10 mT/m. Then FOVmin = 32 kHz/(42.6 MHz/T × 10 mT/m) ≅ 7.5
cm Now if Gmax is increased to 25 mT/m, then FOVmin = 32 kHz/(42.6
MHz/T × 25 mT/m) ≅ 3 cm If you reduce the BW to 16 kHz (±8 kHz),
then the minimum FOV will be reduced even further to about 1.5
cm.
3. Faster imaging is possible, including faster FSE (fast spin echo),
GRASE (gradient and spin echo), and EPI (echo planar imaging).
4. 3D-FSE is possible (useful for 3D T2-weighted imaging in brain and
spine).
5. Contrast-enhanced magnetic resonance angiography (CE-MRA) images
can be acquired during the transit time of the gadolinium through the
artery or vein (first-pass technique) or through the center of k-space
(elliptical-centric technique)
6. Detection of very slow flow (blood or cerebrospinal fluid [CSF]) is
possible via phase contrast (PC) techniques.
7. PC MRA can now be performed in a reasonable acquisition time with
increased sensitivity to slow flow. (The duration of the flow-encoding
gradient can be reduced, allowing for a shorter TR.)
8. High-resolution imaging of CSF flow in small shunts is possible.
9. Ultra high-resolution MRA is possible (1024 × 1024 matrix over a 22-
cm FOV, resulting in 250 µm resolution!).
10. High-resolution (both spatial and temporal) dynamic MR techniques
are possible because, by shortening the TE and TR, a larger matrix can
be employed at a given FOV in the same acquisition time. Thus, both
spatial and temporal resolutions can be improved. One application is
dynamic MR study of breast cancer (turboFLASH sequence with TR = 7
msec, TE = 3 msec, 128 × 128 matrix, acquisition time = 7 × 128 = 896
msec ≅ 0.9 sec).
11. The minimum TE can be reduced, thus minimizing dephasing and
increasing T1 contrast. For example, high performance gradients can
provide high-dose gadolinium performance with the use of a single dose
of gadolinium, or, alternatively, provide single dose performance with
only half a dose of gadolinium.
Very fast sequential postgadolinium imaging can be performed for
evaluation of decreased perfusion leading to cerebral or myocardial
ischemia. Additionally, diffusion imaging is possible by the addition of
diffusion-sensitizing gradient pulses to any pulse sequence. The current
main clinical application is in early detection of cerebrovascular
ischemia.

Key Points
High performance gradients have revolutionized MR
imaging. In short, a higher magnetic strength allows
a shorter duration, thus reducing the cycle time. This
in turn reduces the minimum TE and TR, which
shortens the acquisition time. In summary, high
performance gradients allow many new and improved
features, including faster scanning (including EPI,
GRASE, 3D-FSE, CE MRA, etc.), improved spatial and
temporal resolution, smaller FOV, high-resolution
MRA (both TOF and PC) with improved visualization
of small vessels and slow flow, and improved imaging
of CSF flow, just to name a few.

Questions
30-1 High performance gradients require
(a) high gradient strength (Gmax)
(b) short rise time (tR)
 (c) both
(d) none

30-2 The slew rate is defined as

(a) tR/Gmax
(b) Gmax/tR
(c) γGmax/tR
(d) none of the above

30-3 Advantages of high performance gradients include all of the

following except
(a) faster scanning
(b) smaller FOV
(c) reduced chemical shift
(d) diffusion imaging

30-4 Two gradients with different profiles but the same area (Fig.
30-1) will have the same effects (i.e., phase changes) on
(a) stationary spins
(b) flowing spins
(c) both
(d) none
31
The Many Combinations of MRI

Introduction
Thus far we have studied a vast array of MRI sequences that seem
somewhat related to one another, but also, quite different. The pulse
sequences that we have looked at are the basic sequences used in MRI;
however, there are many more combinations of sequences that can be
performed and in this chapter we will look at how these interrelate.

Basic Building Blocks

The basic elements needed to build an MRI pulse sequence can be
simplified into four components: (i) prep pulse (optional), (ii) radio
frequency (RF) pulse, (iii) refocusing mechanism, and (iv) readout
(Table 31-1). This is obviously a fairly simplified approach and omits
certain items such as the phase-encode step, which is in every
sequence. It also puts in an artificial distinction between the gradient
refocusing mechanism in gradient-echo (GRE) sequences and the
frequency readout, which are being performed simultaneously.
Additionally, the refocusing 180° RF pulses in the fast spin-echo (FSE)
technique are important both for refocusing and for making multiple
readouts. Regardless of the few shortcomings, this approach is useful
for understanding the new sequences that the manufacturers are
offering on their magnets.
Prep Pulse (Optional). The prep pulse comes temporally before the
other three elements, but it is also the one component that is optional.
When a prep pulse is used in a sequence, it can actually have an
important part to play in the appearance of the image as the root pulse
sequence. There are three basic types of prep pulses: (i) 180° RF
inversion pulse, (ii) a fat saturation (or chemical saturation) pulse
(usually 90° RF pulse), and (iii) a magnetization transfer (MT) pulse.
All three of these pulses have different characteristics and properties,
with resultant different effects. We have discussed all these pulses in
prior chapters; however, it is important to note that any of these can
be used with any pulse sequence to achieve a desired effect on the
image.
For instance, a 180° RF pulse is used in both STIR and FLAIR imaging
with the only variables being the TI, TR, and TE (conventional inversion
recovery [IR] and fast IR pathways are shown in Table 31-2; a novel
application is seen in Fig. 31-1). The 180° RF pulses can also be used to
suppress flow such as in a double inversion recovery (DIR) (or black-
blood) technique that is useful in cardiovascular imaging. This
technique uses a nonslice selective 180° pulse followed by another 180°
pulse that is slice selective. This idea can be carried out further by
adding a third 180° pulse that nulls fat (known as a triple IR or fat-
saturated blackblood technique—really just a combination of STIR and
DIR) (Fig. 31-2). An inversion pulse is also useful in cardiac imaging for
identification of infarcted myocardium in delayed enhancement
imaging. In this technique, gadolinium is given and the heart is imaged
10 to 20 min later using an IR-GRE technique with an inversion

time of 200 to 300 msec to null normal myocardium (Table 31-3 and Fig.
31-3).
Figure 31-1. A: An SSFSE FLAIR image of the brain in an
uncooperative patient. There are corresponding SSFSE (B), T2 (C)
with marked motion, and later FLAIR (D) images when the patient
was more cooperative. There are periventricular white matter
ischemic changes and an old right parietal stroke. This example
demonstrates a novel approach for newer fast imaging sequences
that can be used in uncooperative patients.

Table 31-1 The Many Combinations of MR

Prep Pulse RF Refocusing Readout

Pulse

IR 90° 180° Single

(180° pulse) SE

GRE

Fat sat (chem Multiple

sat) <90° Gradient (segmented)

FSE

EPI

Multiple (single
MT shot)

EPI

SSFSE

Abbreviations: IR , inversion recovery; MT , magnetization

 transfer; SE , spin echo; GRE , gradiant echo; FSE , fast spin
echo; EPI , echo planar imaging; SSFSE , single-shot FSE ; RF ,
radio frequency.

Table 31-2 Inversion Recoverya

Prep Pulse RF Pulse Refocusing Readout

aFor abbreviations, see Table 31-1 footnote.

Figure 31-2. A: Short-axis DIR (double inversion recovery) or black-

 blood technique. B: The same slice using the TIR (triple inversion
recovery) technique—notice the nulling of the fat on the TIR versus
the DIR.

Table 31-3 IR GREa

Prep Pulse RF Pulse Refocusing Readout

aFor abbreviations, see Table 31-1 footnote. Pathway for

delayed enhancement imaging.

Furthermore, fat saturation (chemical saturation) is commonly used

with spin-echo, FSE, GRE, and echo planar imaging (EPI) sequences,
that is, any type of readout. (Table 31-4 lists some of the more common
combinations and Fig. 31-4 illustrates some of these.) Thus, we have
seen that the prep pulses can be added to any sequence, and any
combination of prep pulses could be performed prior to the RF pulse in
order to achieve a certain effect.
 Figure 31-3. A short-axis IR-GRE delayed enhancement image
showing transmural bright signal in the anterior wall and septum
diagnostic of infarcted myocardium (TR 7/TE 3/TI 150 msec).

RF Pulse. The RF pulse would be the start of the sequence if there

were no prep pulse. The RF pulse, as we have discussed previously,
induces resonance into the system and tips the longitudinal
magnetization into the transverse plane. A 90° pulse is used for the
spin-echo technique, whereas a partial flip angle pulse <90° is used for
GRE techniques. T1, T2, or proton density weighting is achieved by
altering the TR, TE, and flip angle (for the gradient technique). Again,
any of the above prep pulses can be used with any RF pulse.
Refocusing Mechanism. As we have discussed in prior chapters, the
free induction decay (FID) usually comes on too early and decays too
rapidly, keeping us from having enough time to encode the signal
spatially and read it out (with the exception of ultrashort TE techniques
currently in development). The solution to this problem is found in
refocusing the spins. This is done either with a 180° pulse, which
eliminates any external magnetic inhomogeneities and allows us to
measure the signal on the T2 curve, or with a gradient, which can be
done much faster but at the cost of measuring the signal from the
steeper T2* curve. A gradient refocusing mechanism could be useful
with any RF flip angle. However, the same cannot be said for a 180°
refocusing pulse, which is only useful for a 90° RF pulse, but not for a
partial flip angle RF pulse (see Chapter 20).
Readout. The readout is a collective concept that embodies both the
frequency-encode gradient (which is also the refocusing gradient in

GRE sequences) and any additional repetition of the readout (with its
accompanying different phase-encode steps) such as additional 180° RF
pulses in FSE or the alternating gradients in both the frequency and
phase directions in EPI. An overarching idea is that the readout
following a single RF pulse can either fill a single line of k-space,
multiple lines (segmented or multishot

filling of k-space), or all the lines of k-space (single-shot technique).

Advances in gradient strengths and computers make any of these
possibilities available for any technique.
Figure 31-4. Examples of chemical (spectral) fat saturation. A:
Postgadolinium CSE T1 image in a patient with diffuse metastasis of
the right lobe (TR 467/TE 14 msec). B: FSE T2 image in a patient
with siderotic nodules and cirrhosis (TR 4000/TE 84 msec). C:
HASTE image (SSFSE) in a patient with cirrhosis, ascites, and
multiple metastases (unusual in a cirrhotic patient; TR 1000/TE 83
msec). D: Postgadolinium SPGR in the same patient as (B) (TR
160/TE 4 msec).

Table 31-4 Fat Saturationa

Prep Pulse RF Pulse Refocusing Readout

 aFor abbreviations, see Table 31-1 footnote. Typical pathways

are shown, although any combination is possible.

Figure 31-5. b0 (A) and diffusion-weighted (B) spin-echo EPI

images in a patient with complete infarction of the left basal
ganglia from a large MCA CVA that involved the lenticulostriate
branches.

For instance, we have already discussed spin-echo sequences that fill a

single line of k-space versus the FSE technique that fills multiple lines
of k-space versus the ½ NEX singleshot technique (HASTE or SSFSE) that
fills all the lines of k-space from a single RF pulse or shot (Table 31-5
and Fig. 31-4A-C). From Table 31-5, you can also see that in typical
spinecho imaging any prep pulse and any readout (single, segmented,
or single shot) can be performed. For even faster spin-echo imaging,
EPI readouts could be performed, usually in a single-shot technique
such as in diffusionweighted imaging (Table 31-6 and Fig. 31-5).

Table 31-5 Spin Echoa

Prep Pulse RF Pulse Refocusing Readout

aFor abbreviations, see Table 31-1 footnote. Dashed lines

represent optional pulses.

Table 31-6 Spin-Echo EPIa

Prep Pulse RF Pulse Refocusing Readout

 aFor abbreviations, see Table 31-1 footnote. Solid lines

represent typical diffusion-weighted image, and the dashed

lines represent optional pathways. Note that EPI images are
usually acquired with fat saturation due to the marked
chemical shift artifacts without it.

Most gradient sequences fill a single line of k-space per RF; however,
newer techniques are being implemented such as an FGRET (fast
gradient-echo train) that is useful in cardiac perfusion imaging (Table
31-7 and Fig. 31-6). This is a GRE sequence with a segmented filling of
k-space (four lines per RF pulse).

Table 31-7 GRE-EPIa

Prep Pulse RF Pulse Refocusing Readout

aFor abbreviations, see Table 31-1 footnote. Solid lines

represent GRE -EPI for FGR ET sequence, and dashed lines

represent other possible combinations.
You can now more easily appreciate that MRI sequences (especially the
newer applications) are usually a mix and match of the basic elements
in a novel combination for a specific purpose.

Figure 31-6. Short-axis dynamic postgadolinium perfusion study

shows normal enhancement of the myocardium (TR 6/TE 1.5
msec).
 Key Points
1 Each pulse sequence can be simplistically viewed
as a prep pulse (optional), RF pulse, refocusing
mechanism, and readout.
2 There are three main prep pulses: (i) 180° RF
pulse (inversion recovery), (ii) fat saturation (or
chemical sat) pulse, and (iii) magnetization transfer
(MT) pulse.
3 The RF pulse is either a 90° RF pulse or a partial
flip angle (<90°) pulse.
4 The refocusing mechanism is either by 180° RF
pulse or a gradient.
5 The readout can either fill a single line of k-space,
multiple lines of k-space (segmented), or all the lines
of k-space (single shot) following the RF pulse.
6 Any of the above can be mixed and matched in any
combination that is desirable for either image
characteristics or speed with the exception of a
partial flip RF pulse followed by a 180° refocusing
pulse.

Questions
31-1 T/F A prep pulse is required in every MRI pulse sequence.

31-2 T/F A 180° prep pulse can be followed by a GRE sequence.

31-3 T/F A 180° refocusing pulse is useful in a GRE sequence.

31-4 Which of the following sequences are not useful?

 (a) MT, 90° RF, 180° refocusing pulse, single readout
(b) fat sat, 90° RF, 180° refocusing pulse, multiple readout
(segmented)
(c) 180° inversion pulse, <90° RF, gradient refocusing, single
readout
(d) no prep pulse, 90° RF, 180° refocusing pulse, multiple
readout (singleshot EPI)
(e) none of the above; they are all useful

31-5 T/F Multiple readouts of the singleshot variety can be based

either on an FSE technique or on an EPI technique.
Appendix A: Suggested Readings
1. American College of Radiology. Glossary of MR Terms. 5th ed.
Reston, VA: ACR; 2005.

2. Bradley WG. Optimizing lesion contrast without using contrast

agents. J Magn Reson Imaging. 1999; 10:442-449.

3. Bradley WG, Waluch V, Lai K, et al. The appearance of rapidly

flowing blood on magnetic resonance images. Am J Roentgenol.
1984;143:1167-1174.

4. Bushberg JT, Seibert JA, Leidholdt EM, et al. The Essential Physics
of Medical Imaging. 2nd ed. Philadelphia: Lippincott Williams &
Wilkins; 2002.

5. Chun Y, Schmiedl UP, Weinberger E, et al. Threedimensional fast

spin-echo imaging: pulse sequence and in vivo image evaluation. J
Magn Reson Imaging. 1993;3:894-899.

6. Edelman RR, Wielopolski P, Schmitt F, et al. Echoplanar MR

imaging. Radiology. 1994;192:600-612.

7. Erickson SJ, Cox IH, Hyde JS, et al. Effect of tendon orientation
on MR imaging signal intensity: a manifestation of the “magic angle”
phenomenon. Radiology. 1991;181:389-392.

8. Hashemi RH, Bradley WG, Chen D-Y, et al. Suspected multiple

sclerosis: MR imaging with a thin-section fast FLAIR pulse sequence.
Radiology. 1995;196:505-510.

9. Henkelman RM, Hardy PA, Bishop JE, et al. Why fat is bright in
RARE and fast spin-echo imaging. J Magn Reson Imaging.
1992;2:533-540.
10. Hennig J, Nauerth A, Friedburg H. RARE imaging: a fast imaging
method for clinical MR. Magn Reson Med. 1986;3:823-833.

11. Huda W, Slone R. Review of Radiologic Physics. 2nd ed.

Philadelphia: Lippincott Williams & Wilkins; 2003.

12. Kapelov SR, Teresi LM, Bradley WG, et al. Bone contusions of the
knee: increased lesion detection with fast spin-echo MR imaging
with spectroscopic fat saturation. Radiology. 1993;189:901-904.

13. Oppenheim AV, Willsky AS, Nawab S. Signals and Systems. (2nd
Ed.). New Jersey: Prentice-Hall; 1996.

14. Pierpaoli C, Jezzard P, Basser PJ, et al. Diffusion tensor MR

imaging of the human brain. Radiology. 1996;201:637-648.

15. Prince MR, Grist TM, Debatin JF. 3D Contrast MR Angiography.

Berlin: Springer; 1997.

16. Prince MR, Narasimham DL, Stanley JC, et al. Breath-hold

gadolinium-enhanced MR angiography of the abdominal aorta and its
major branches. Radiology. 1995;197:785-792.

17. Stark DD, Bradley WG, eds. Magnetic Resonance Imaging. Vol. 1-
3. 3rd ed. St Louis: Mosby; 1999.

18. Ulug AM, Moore DF, Bojko AS, et al. Clinical use of diffusion
tensor imaging for diseases causing neuronal and axonal damage.
Am J Neuroradiol. 1999;20:1044-1048.

19. Saremi F, Grizzard JD, Kim RJ. Optimizing cardiac MR imaging:

practical remedies for artifacts. Radiographics. 2008;28:1161-1187.
20. Lotz J, Meier C, Leppert A, et al. Cardiovascular flow
measurement with phase-contrast MR imaging: basic facts and
implementation. Radiographics. 2002;22:651-671.

21. Scott AD, Keegan J, Firmin DN. Motion in cardiovascular MR

imaging. Radiology. 2009;250: 331-351.

22. Simonetti OP, Kim RJ, Fieno DS, et al. An improved MR imaging
technique for the visualization of myocardial infarction. Radiology.
2001; 218:215-223.

23. Foo TK, Bernstein MA, Aisen AM, et al. Improved ejection
fraction and flow velocity estimates with use of view sharing and
uniform repetition time excitation with fast cardiac techniques.
Radiology. 1995;195:471.

24. Srichai MB, Lim RP, Wong S, et al. Cardiovascular applications of

phase-contrast MRI. Am J Roentgenol. 2009;192:662-675.

25. Huber AM, Schoenberg SO, Hayes C, et al. Phasesensitive

inversion-recovery MR imaging in the detection of myocardial
infarction. Radiology. 2005;237:854-860.

26. Sievers B, Addo M, Kirchberg S, et al. Impact of the ECG gating

method on ventricular volumes and ejection fractions assessed by
cardiovascular magnetic resonance imaging. J Cardiovasc Magn
Reson. 2005;7:441-446.

27. Mukherji SK. Clinical Applications of MR Spectroscopy. Wiley-

Liss: New York; 1998.

28. Salibi N, Brown MA. Clinical MR Spectroscopy: First Principles.

Wiley-Liss: New York; 1998.
29. Danielsen ER, Ross B. Magnetic Resonance Spectroscopy
Diagnosis of Neurological Diseases. Marcel Dekker: New York; 1999.

30. Majos C, Aguilero C, Alonso J, et al. Proton MR spectroscopy

improves discrimination between tumor and pseudotumoral lesion in
solid brain masses. Am J Neuroradiol. 2009;30:544-551.
Appendix B
Answers

Chapter 1
1-1 (a) See Figure 1-17
(b) See Figure 1-18
1-2 ei(x+y) = eixeiy. So cos (x + y) + isin (x + y) = (cos x + isin x)(cos y +
isin y) = (cos x · cos y − sin x · sin y) + i(cos x · sin i + sin x · cos y)
1-3 sinc(0) = sin(0)/0 = limd/dx (sin x/x) xrarr;0
= cosx/1 = cos0/1 = 1/1 = 1 (x = 0)
1-4 (a) e−1 = 0.37 (b) e−2 = 0.14
(c) e−1 = 0.37
1-5 d/dt(Ae−t/T) = A (−1/T)e−t/T, which is −A/T at t = 0. This is the
slope of the line tangent to the curve at t = 0 that crosses the t-axis at
t = T.
1-6 ln (ex) = x = ln 8 = ln 23 = 3ln 2 = 3 × 0.693 = 2.079
1-7 (a) 0 (b) 0.5 (c) 1
(d) 0 (e) 1 (f) 0.5
(g) 0 (h) −1

Chapter 2
2-1 (a) 14.9 MHz
(b) 21.3 MHz
(c) 42.6 MHz
(d) 63.9 MHz
(e) 85.2 MHz
(f) 127.8 MHz
2-2 T
2-3 F (only mobile protons)
2-4 F
2-5 T
2-6 F (speed of light)
2-7 T
2-8 F
2-9 F
2-10 T
2-11 T

Chapter 3
3-1 T
3-2 F
3-3 T
3-4 T
3-5 T
3-6 F (twice)
3-7 T

Chapter 4
4-1 (a) T (b) F (c) F (d) T
4-2 b
4-3 d
4-4 F (by T2*)
4-5 c
4-6 (a) i (b) ii

Chapter 5
5-1 (a) 1.56 (b) 90 msec
(c) 0.72 and 1.05
(d) 1.28, 50 msec, 0.88 and 1.28
(e) 2.10
5-2 H2O/fat = 0.25 and 1.63; CSF/GM = 0.41 and 1.40
5-3 b
5-4 c
5-5 (a) N(H)e−TE/T2 (i.e., ideal T2W)
(b) N(H)(1 − e−TR/T1) (i.e., ideal T1W)
(c) N(H)(i.e., ideal PDW).
5-6 (a) ii (b) i (c) iii (d) iv

Chapter 6
6-1 T
6-2 (a) iii (b) i (c) iv
(d) i (e) ii
6-3 T
6-4 (a) i (b) ii (c) iv

Chapter 7
7-1 Setting 1 − 2e−t/T1 = 0, we get e−t/T1 = 1/2 so −t/T1 = ln (1/2) =
−0.693 so t = 0.693 T1
7-2 SαN(H) (1 − 2e−TI/T1) (1 − e−TR/T1) and N(H) (1 − 2e−TI/T1 − e−TR/T1 +
2e−(TR+TI)/T1) ≅ N(H) (1 − 2e−TI/T1 − e−TR/T1 + 2e−TR/T1) = N(H) (1 − 2e−TI/T1
+ e−TR/T1)
7-3 (a) ii (b) i
7-4 T

Chapter 8
8-1 (a) N(1 − e−TR/T1)e−TE1/T2 and N(1 − e−TR/T1)e−TE2/T2
(b) N(1 − e−TR/T1) e−TE1/T2*
(c) 0.61 and 0.37
8-2 (a) i (b) iii (c) ii
8-3 F (not that due to spin-spin interactions)

Chapter 9
9-1 T
9-2 (a) F (b) F
9-3 T
9-4 T

Chapter 10
10-1 (a) 4.7 mT/m (b) 1 mm
10-2 f
10-3 T
10-4 T

Chapter 11
11-1 (a) i (b) iii (c) ii
11-2 T
11-3 360°/128 = 2.8°
11-4 (i) d (ii) a
11-5 T
Chapter 12
12-1 b
12-2 c
12-3 T

12-4 F(1/ )
12-5 d
12-6 F

Chapter 13
13-1 T
13-2 F (phase-encoding gradient)
13-3 T
13-4 T
13-5 T
13-6 F
13-7 F (there is conjugate symmetry)
13-8 F

Chapter 14
14-1 a, d, e
14-2 (a) 512 sec = 8 min, 32 sec
(b) 5120 sec = 85 min, 20 sec = 1 hr, 25 min, 20 sec!
14-3 b

Chapter 15
15-1 (a) 15 cm (reduces the minimum FOV)
15-2 b
15-3 61°
15-4 d
15-5 F (increases)
15-6 b
15-7 a

Chapter 16
16-1 T
16-2 (a) T (b) F (cycles/cm)
16-3 T
16-4 d
16-5 (a) T (b) T
16-6 T

Chapter 17
17-1 (a) 384 sec = 6 min, 24 sec
(b) 3840 sec = 64 min = 1 hr, 4 min
(c) 6 min, 24 sec
17-2 10 slices

17-3 (a) SNR is increased by

(b) Chemical shift is doubled.
(c) Coverage is reduced since Ts = Nx/BW is doubled
17-4 c
17-5 b
17-6 a

17-7 g
17-8 f
17-9 d
17-10 c
17-11 c
17-12 e
17-13 a
17-14 e
17-15 f
17-16 d
17-17 (a) ii (b) i

Chapter 18
18-1 e
18-2 (a) In terms of number of pixels

0.2 T 0.5 T 1.0 T 1.5 T

50 kHz 0.15 0.38 0.76 1.15

10 kHz 0.76 1.91 3.82 5.73

4 kHz 1.91 4.77 9.54 14.31

(b) In terms of mm

0.2 T 0.5 T 1.0 T 1.5 T

50 kHz 0.14 0.36 0.72 1.07

10 kHz 0.72 1.79 3.58 5.37

4 kHz 1.79 4.48 8.95 13.43

(c) The narrower the BW or the stronger the main magnetic field, the
greater the chemical shift.
18-3 (a) 51.2 (b) 256/51.2 = 5
(c) fewer ghosts
18-4 b
18-5 d
18-6 F (the other way around)
18-7 a
18-8 T
18-9 d
18-10 (a) 48 pixels or 7.5 cm
(b) 2 ghosts
18-11 c
18-12 c
18-13 d
18-14 F (55°)
18-15 a
18-16 c
18-17 e
18-18 T

Chapter 19
19-1 d
19-2 (a) 34 min, 8 sec
(b) 4 min, 16 sec
19-3 F (decreased coverage)
19-4 d
19-5 d
19-6 c
19-7 4 min, 16 sec
19-8 T

Chapter 20
20-1 T
20-2 T
20-3 T
20-4 T
20-5 T
20-6 F (more T1W)
20-7 (a) 15.4 sec
(b) 230.4 sec = 3 min, 50 sec
20-8 F (the opposite)
20-9 T
20-10 b
20-11 T

Chapter 21
21-1 d
21-2 f
21-3 T
21-4 T

Chapter 22
22-1 T
22-2 F
22-3 d
22-4 F
22-5 F
22-6 T
22-7 b

Chapter 23
23-1 e
23-2 d
23-3 a
23-4 T
23-5 T
23-6 F

Chapter 24
24-1 b
24-2 b
24-3 c

24-4 F
24-5 T

Chapter 25
25-1 b
25-2 a
25-3 F
25-4 T
25-5 T
25-6 e
25-7 e
25-8 T

Chapter 26
26-1 c
26-2 F (distal to a stenosis)
26-3 T
26-4 F
26-5 d
26-6 d
26-7 T
26-8 F
26-9 F
26-10 T
26-11 a
26-12 c
26-13 c
26-14 F
26-15 T
26-16 c

Chapter 27
27-1 d
27-2 b
27-3 (a) i (b) i (c) ii (d) ii
27-4 F
27-5 e
27-6 T
27-7 T
27-8 (a) ii (b) i
(c) v (d) iii
(e) iv
27-9 T
27-10 c

Chapter 28
28-1 d
28-2 a
28-3 b
28-4 c
28-5 d
28-6 c
28-7 a

Chapter 29
29-1 b
29-2 a
29-3 c
29-4 c
29-5 d
29-6 F: also elevated in MS and ADEM.
29-7 F: since Cho has a shorter T1 than NAA, the Gd shortens the T1
relatively more for NAA, raising its signal.
29-8 e

Chapter 30
30-1 c
30-2 b
30-3 c
30-4 a

Chapter 31
31-1 F
31-2 T
31-3 F
31-4 e
31-5 T
Appendix C

Abbreviations
α, θ, φ
Alpha, theta, phi. Symbols used to designate an angle
(e.g., flip angle)
Γ
γ. Gyromagnetic ratio (in T/MHz)
Μ
µ. Micron (10-6 m)
Σ
σ. Standard deviation
Τ
τ. Symbol used to designate time
ω, ω0
Omega. Angular (Larmor) frequency (in radians/sec)
2D
Two dimensional
2DFT
Two-dimensional Fourier transform
3D
Three dimensional
3DFT
Three-dimensional Fourier transform
A2D, ADC
Analog-to-digital converter or conversion
ASSET
Array spatial and sensitivity encoding technique
B0
Main external magnetic field
B1
Magnetic field associated with the RF pulse
b-FFE
balanced fast field echo
b-SSFP
balanced steady-state free precession
BW
Bandwidth
CNR
Contrast-to-noise ratio
CSE
Conventional spin echo
DE
Driven equilibrium
EPI
Echo planar imaging
ESP
Echo spacing
ET
Echo train
ETL
Echo train length
FC
Flow compensation
FFE
fast field echo
FFT
Fast Fourier transform
FGR
Fast GRASS
FID
Free induction decay
FIESTA
Fast imaging employing steady-state acquisition
FISP
Fast imaging with steady-state precession
FLAIR
Fluid-attenuated inversion recovery
FLASH
Fast low-angle shot
FOV
Field of view
FSE
Fast spin echo
FSPGR
Fast SPGR
FT
Fourier transform
Gx
Frequency-encoding gradient
Gy
Phase-encoding gradient
Gz
Slice-select gradient
GE
Gradient echo
GM
Gray matter
GMN
Gradient moment nulling
GMR
Gradient-motion rephasing
GRASE
Gradient and spin echo
GRASS
Gradient recalled acquisition in the steady state
GRE
Gradient echo or gradient-recalled echo
HASTE
Half-Fourier acquired single-shot turbo spin echo,
Siemens
Hz
Hertz (1 cycle/sec)
IPAT
Integrated parallel acquisition techniques
IR
Inversion recovery
kHz
Kilohertz
M0
Initial longitudinal magnetization
MAST
Motion artifact suppression technique
MEMP
Multi-echo multiplanar
MHz
Megahertz
mm
Millimeter (10-3 m)
MOTSA
Multiple overlapping thin-slab acquisition
MPGR
Multiplanar GRASS
MP-RAGE
Magnetization-prepared, rapid acquisition gradient echo
MR
Magnetic resonance
MRA
Magnetic resonance angiography
MRI
Magnetic resonance imaging
msec
Milliseconds
MT
Magnetization transfer
MTC
Magnetization transfer contrast
Mxy
Transverse magnetization in the x-y plane
Mz
Longitudinal magnetization
NEX
Number of excitations
nm
Nanometer (10-9 m)
NMR
Nuclear magnetic resonance
NPW
No phase wrap
NFW
No frequency wrap
NSA
Number of signal averages
Nx
Number of frequency-encoding steps in the x direction
Ny
Number of phase-encoding steps in the y direction
Nz
Number of phase-encoding steps in slice-select
direction (in 3D imaging)
PC
Phase contrast
PD
Proton density

PDW
Proton density weighting
PDWI
Proton density-weighted image
pixel
Picture element
POMP
Phase offset multiplanar
pm
Picometer (10-12 m)
ppm
Parts per million
PS
Pulse sequence
PSD
Pulse sequence diagram
PSIF
Opposite of FISP
RARE
Rapid acquisition with relaxation enhancement
Re
Reynolds number
RF
Radio frequency
SE
Spin echo
SENSE
Sensitivity encoding
SMASH
Simultaneous acquisition of spatial harmonics
SNR, S/N
Signal-to-noise ratio
SPGR
Spoiled GRASS
SSFP
Steady-state free precession
SSFSE
Single-shot fast spin echo, GE
STIR
Short TI (or tau) inversion recovery
T1
T1 relaxation time, longitudinal relaxation time, and
spin-lattice relaxation time
T1W
T1 weighting, T1 weighted
T1WI
T1-weighted image
T2
T2 relaxation time, transverse relaxation time, and
spin-spin relaxation time
T2*
T2* relaxation time
T2W
T2 weighting, T2 weighted
T2WI
T2-weighted image
T2*W
T2* weighting, T2* weighted
T2*WI
T2*-weighted image
T
Tesla; period of a periodic signal
TE
Echo delay time (time to echo)
TI
Time to inversion (inversion time)
T0
“Overhead” (dead) time in the pulse cycle
TOF
Time of flight
TONE
Tilted optimized nonsaturating excitation
TR
Repetition time (time of repetition)
tR
Rise time
Ts
Sampling time
TSE
Turbo spin echo
Turbo
Siemens' and Philips' prefix to denote a fast scanning
mode
VB
Variable bandwidth
VENC
Velocity encoding
VEMP
Variable echo multiplanar
voxel
Volume element
WM
White matter
A
Acceleration compensation, 308
Acoustic schwannoma, 231
Acquisition time, 164 165 166 180 181 See also Scan time
Active time, 124 125 127
ADC. See Analog-to-digital conversion (ADC); Analog-to-digital
converter (ADC)
Algebraic manipulation, 173
Aliasing, 131 132 133 134 135 136 132 185 186 187 188
as an analogy, 136 137 138 138
defined, 136
prevention of, 187 188 188
undersampling and, 138 139
Amplitude
of signals, 109 110 151
Analog-to-digital conversion (ADC), 130 defined, 123
Analog-to- digital converter (ADC), 257
Angle
magnitude and, 5 6
of vector, 6 6
Angular frequency (ω), 6 17 26 See also Larmor frequency
Anisotropy, 201
Anisotropy index, 263
Apparent diffusion coefficient, 264
Applied magnetic field
induced and, relationship between, 20
Arc, 3
Array spatial and sensitivity encoding technique (ASSET), 279
Arrhythmia reject window, 329
Artifact(s), 185 186 187 188 189 190 191 192 193 194 195 196
197 198 199 200 201 202 203 204 205 206 207 208 209 210 211
212
central, 203 204 204
chemical shift, 188 189 190 191 192 193 194 195 190 191 192
193 262
CSF flow and, 200 200 212 212
 data errors and, 212
in echo planar imaging (EPI) technique, 261 262
external magnetic field, 205 206 207
free induction decay (FID), 204 204
geometric distortion, 105
ghost, 198 199 198 199 208
gradient-related, 207 211 212
image processing, 185 186 187 188 189 190 191 192 193 194 195
196
magic angle, 200 201 202
magnetic susceptibility, 206 207 208 209 210
motion, 147 197 198 199 200 201 198 199 200 201
N/2 ghost, 261
overshoot, 104 See also Ripples effect
partial volume, 196 197
patient-related, 196 197 198 199 200 201
RF-related, 201 202 203 204 205
feedthrough zipper, 205 205
zipper, 203 204 205 204
stimulated echo, 204 205
susceptibility, 261 262
truncation, 189 195 196 196 197
undershoot, 104 See also Ripples effect
zipper, 203 204 205 204
feedthrough, 205 205
Asymmetric field of view (FOV), 270 271 271
Audio frequency, 104 105
Autocalibrating reconstruction for Cartesian sampling (ARC), 279
Auto RF (radio frequency) pulse, 38
Auto shimming, 206
B
Back projections, 111 112 113 113
advantage of, 112
disadvantage of, 112
Band-pass filters (BPF), 272
Bandwidth (BW), 92 99 104 105 106 106 129 130 130
center frequency of, 104 105 106 106
decreasing, and chemical shift artifact, 192 193 194
field of view (FOV) and, 167 168 169 171
Gaussian RF pulse, 104 105
measurement of, 99 100 100
narrower frequency, 101
receiver, 140
and sampling interval (ΔTs), 139 140 141 142 143
signal-to-noise ratio (SNR) and, 177 178
transmission, 140
Bar magnet, 18 18
Bellows device, 327 328 329
b-FFE (balanced fast field echo, Philips), 330
Bi-lobed gradients, 235 235 236
Binary number, 130
Bipolar flow-encoding gradient, 312
bit(s), 130
Black-blood MRA, 323 See also Magnetic resonance angiography (MRA)
fat-saturated, 356
Blood flow, 295 295
normal appearance of, 295 295
Blood motion, 331
Body coil(s), 27
Bounce point artifact, 195
Boundary effect, 195 195
BPF. See Band-pass filters (BPF)
Brain. See also White matter
hematoma in, and tissue contrast, 70 71
hemorrhage in, and tissue contrast, 66 70 72 73
lesions of
 tissue contrast and, 61 61
suppression of. See Tissue suppression
tissue contrast in, 60 61
TE (echo delay time), 63 64 63 64
TR (repition time) and, 63 64 63 64
Brownian motion, 177
BW. See Bandwidth (BW)
Byte, 130
C
Cardiac gating, 300
Cardiac motion, 329 330 331
Cardiac MRI, 327
cine gating imaging, 337 338 339 340
cine imaging, 335 336 337
clinical applications, 340
compensation, 327 328 329 330 331
delayed enhancement, 335
GRE and true FISP gating, 335
k-space filling strategies, 331
late gadolinium enhancement, 335
motion effects, 327 328 329 330 331
pulse sequences, 340
static imaging, 332 333 334
Carrier frequency, 130 151
Center frequency, 105
of RF pulses, 129 130 130
Central artifacts, 203 204 204
Cerebrospinal fluid (CSF), 60 294
flow of
artifact caused by, 200 200
imaging of, 156 156
proton density factor for, 62 62 62
recovery and decay curves for TR (repition time) and, 63 64 63 64
suppression of. See Tissue suppression
T1 characteristics, 60 60 62
T2 characteristics of, 60 61 62
Charged nuclei, 19
Chemical saturation pulse, 356
Chemical shift artifact(s), 188 189 190 191 192 193 194 195 190
191 192 193
in EPI, 262
Chemical shift effect of the second kind, 194 195 194
Chemical (Spectral) presaturation, 274 276 288 289 289
advantages of, 288
 disadvantages of, 289
CNR. See Contrast-to-noise ratio (CNR)
Cobalt (Co), 21
Cocurrent flow, 303 304
Coil(s)
body, 27
gradient, 26 27 28 98
frequency-encoding (readout) (Gx), 28
phase-encoding (Gy), 28
slice-select (Gz), 28
head, 26
phased-array, 28
quadrature, 28
radio frequency (RF), 26
shim, 28
solenoid, 28
surface, 27
transmit/receive, 27
Combined flow phenomena, 303
Complementary SPAtial Modulation Magnetization (CSPAMM), 337
Complex numbers, 5 153 154 155 156 157 158 159 160
conjugate (Hermitian) symmetry, 153 154
fractional echo, 155 155
magnitude (modulus) and phase image, 155 156 157 156
NEX
½, 153 155 155
¼, 155 155
real and imaginary images, 155 156
Composite signals
sampling of, 141 142 143 142
Conjugate (Hermitian) symmetry, 153 154
Contiguous slice option, 273 273
Contiguous slices, 203 203
Contrast averaging, 226 227 228
Contrast enhanced MRA, 315 354
advantages, 316
disadvantages, 316
Contrast-to-noise ratio (CNR), 287
Conventional spin echo (CSE), 217 218 257
scan time for, 217
Convolution, 131 133
Cosine, 3 4 5
cosine data
data space with, 152 153 153
Cosine function, fourier transform (FT) of, 90 91 92 93 99 100 109
110 134 134
Co-tangent, 3
Countercurrent flow, 303 304
Coverage, 128 129
parameters of, 176 181 182 181
Creatine (Cr), 345
Cross-talk, 101 101 201 202 203 202 203
minimize, 101 102
Crusher gradient(s), 163 163 230
CSE. See Conventional spin echo (CSE)
CSF. See Cerebrospinal fluid (CSF)
Curves
Gaussian, 101 101 104
“Custom-made” techniques, 176
Cycle(s), 6 7
D
Data errors, artifact caused by, 212
Data space, 117 118 119 120 119 145 146 145 171 See also K-space
components of, 152 153 153
definition of, 171
Fourier transform of, 121 121
in time domain, 120
trajectories, 120
Dead time, 127 127 164 165
Decay constant, 9 10
Decay/recovery curves, TR (repetition time) and, 63 64 63 64
Delayed enhancement imaging, 255 356
Delta function, 131 133
Demodulation, 105 106
Deoxyhemoglobin
tissue contrast and, 70
Dephasing, 41 42 43 83 125 146 163 163
intravoxel, 298
odd-echo, 298
in phase spins, 41
90° RF pulse turned off, 41
spin-spin interaction(s), 42 43 83
DFT. See Digital Fourier transform (DFT)
2DFT. See Two-dimensional fourier transform (2DFT)
Diamagnetic substances, 21
magnetic susceptibility in, 21
in MRI, 21
opposite direction in, magnetic field in, 21
reduced magnetic field in, 21
Diamagnetism, 21 206 209
Diastolic pseudogating, 300
Dielectric effects
artifacts caused by, 212 212
Diffusion coefficient, apparent, 264
Diffusion imaging
in echo planar imaging (EPI) technique, 262
Diffusion tensor imaging (DTI)
in echo planar imaging (EPI) technique, 262 263 263
diffusion-weighted imaging, 361
Digital Fourier transform (DFT), 160
Dipolar Hamiltonian, 200
Dipole-dipole interactions, 20 21
Direction, of signal, 153
Discitis, tissue contrast in, 68
Dixon effect, 242 243 244 243
Double inversion recovery (DIR), 332 356
DTI. See Diffusion tensor imaging (DTI)
2D-TOF MRA, 310
3D-TOF MRA, 310
Dysprosium (Dy), 21 206
E
Echo(es), 85 86 108 123 124
asymmetric, 87
delay time. See TE (time to echo)
fractional, 155 155
symmetric, 87 87
Echo planar imaging (EPI), 50 145 354
advantages and disadvantages of, 263 264
artifacts in, 261 262
basic idea in, 257
blipped, 258 259
clinical applications of, 265 266
contrast in, 260 261 261
for diffusion imaging, 262
for diffusion tensor imaging (DTI), 262 263 263
functional, 262 263 264 265
gradient echoes in, 257
k-space trajectory in, 260 260
multishot, 257 259 259
for perfusion imaging, 263 264 265
root pulsing sequence, 260 261 261
scan time in, 260
segmental, 259
sequence, 359
single-shot, 257 258 258
types of, 257 258 259
Echo sampling period (ESP), 260
Echo spacing (ESP), 218 218 219
Echo train length (ETL), 180 218 219 220 221 222 223 220 221 222
223
Eddy currents
artifacts caused by, 207 211
Edema
inversion recovery (IR) imaging of, 78 79 79
magnitude reconstruction, 79 80 81
proton density factor for, 62 62
 vasogenic, and tissue contrast, 61 61
Effective TE (TEeff), 219 221
Eigenvalues, 263 264
Eigenvectors, 263 264
diffusion, 265
principal, 263
Electric field, of electromagnetic waves, 16 17
Electromagnetic field, 18 19 20
Electromagnetic spectrum
frequency ranges in, 16 18
windows of, 16 17 18 18
Electromagnetic wave(s)
angular frequency of, 17
characteristics of, 16 17
components of, 16 17 17
electric field of, 16 17
electromagnetic windows and, 18
energy of, 18 18
frequency ranges in, examples of, 18
linear frequency of, 17
magnetic field of, 17
Maxwell's wave theory and, 16 17
RF pulse, 31
speed of light of, 16
Electromagnetic windows, 18
Elliptical-centric technique, 354
Endometrioma, tissue contrast and, 73
Energy levels, of atomic nucleus, 19
Entry phenomenon, 300
EPI. See Echoplanar imaging ( EPI)
ESP. See Echo spacing (ESP)
ETL. See Echo train length (ETL)
Euler's equation, 11 12 12
Even-echo rephasing, 298 299 300
effect, 226
Even function, 152
Fourier transform (FT) of, 94 94
Excitation
definition of, 165
Exponential functions (ex), 8 9 8 9 10
properties of, 13
External magnetic field artifact(s), 205 206 207
External magnetic field (B0), 21 23 24
inhomogeneity, and dephasing, 42 43 83
in Larmor equation, 26 26
with RF pulse transmission, 31 31 32 33
F
Faster imaging, 331
Fast FLAIR (fluid-attenuated inversion recovery), 284 285 286 287
288 286
Fast Fourier transform (FFT), 160
Fast gradient echo train. See FGRET (fast gradient echo train)
Fast imaging with steady-state precession. See FISP (fast imaging with
steady-state precession)
Fast IR, 231
Fast low-angle shot (FLASH), 247 248 249 250 251
characteristics of, 256 256
disadvantages of, 251
multiplanar (MPSPGR), 252
tissue contrast, 248
Turbo, 252 253
Fast spin echo (FSE), 145 257 269 354 356
gating, 332 333
and HASTE, 332
Fast spin-echo (FSE) imaging, 180 218 219 220 221 222 223 224 225
226 227 228 229 230 231
advantages of, 226 227
disadvantages of, 227 228
echo train length (ETL), 218 219 220 221 222 223 220 221 222
223
with higher bandwidths (BWs), 228 230
and magnetic susceptibility, 228
magnetization transfer in, 228
multi-echo, 224 225 226 227 228
slice coverage in, 223 224 223
three-dimensional (3D), 230 231 230 231
trade-offs of, 223 224
Fat
suppression, 79 80 81
clinical applications of, 290 291
T1 characteristics of, 58 59 65 68 73
T2 characteristics of, 57 59 60 65 69
Fat-saturated blackblood technique, 356
Fat saturation pulse, 228
Feedthrough zipper artifacts, 205 205
Felix Bloch, 18
Ferromagnetic substances
cobalt (Co), 21
iron (Fe), 21
magnetic susceptibility in, 21
in MRI, 21
nickel (Ni), 21
permanent magnetism in, 21
Ferromagnetism, 21 206 207 210
FFT. See Fast Fourier transform (FFT)
FGRET (fast gradient echo train), 361 362 362
FID. See Free induction decay (FID)
Field of view (FOV), 112 167 168 169 167 171 172 173
bandwidth and, 167 168 169 171
gradient strength and, 167 168 169 171
increasing, and prevention of aliasing, 188 188
FIESTA (fast imaging employing steady-state acquisition), 330
First-pass technique, 354
FISP(fast imaging with steady-state precession), 247
characteristics of, 256 256
multiplanar, 252
Turbo, 252 253
FLAIR (fluid-attenuated inversion recovery), 284 286
fast, 284
FLASH. See Fast low-angle shot (FLASH)
Flip angle, 34
partial, 233
partial techniques, 233
Flip(ping), 34
90°, 35 36 37
180°, 37 38
partial, 38 38
physical properties behind, 34
Flow
 blood, 295 295
combined, 303
concurrent, 303 304
countercurrent, 303 304
laminar, 294 295
plug, 294 295
second order, 308
turbulent, 294 295 298
vortex, 294 295
Flow compensation techniques, 200
Flow encoding, 311
Flow phenomena, 294 295 296 297 298 299 300 301 302 303 304
305 306 307 308
combined, 303
Flow-related enhancement (FRE), 300 301 302 303
Flow (stream) separation, 294 295
Fluid-attenuated inversion recovery (FLAIR), 182
Formication, 264
Fourier series, 94 95 96
Fourier transform (FT), 11 90 91 96 99 See also Fast Fourier
transform (FFT)
of complex signal, 92 93
of cosine function, 90 91 92 93
of cosine function, 134 134
for cosine wave, 99 100
of data space, 121 121
digital (DFT), 160
equation for, 90
of even function, 94 94
versus Fourier series, 94 95 96 95
imaginary images, 156
of k-space, 171 173 174
of odd function, 94 94
real images, 156
of RF pulses, 130 130
of sinc function, 91 92 92 130 131 131 150 150
of sine wave, 92 93 94 93 94
 of spikes, 131 133
of truncated sinc function, 105
FOV. See Field of view (FOV)
Fractional anisotropy, 263
Fractional echo, 155 155 269 270 270
Fractional NEX(number of excitations), 268 269 269
Fractional RF, 270 270
FRE. See Flow-related enhancement (FRE)
Free induction decay (FID), 44 45 46 46 49 49 50 51 52 51 52 74
75 75 83 84 85 86 97 163 359
artifacts caused by, 204 204
measurement of, 235 236 236
Frequency, 6 See also Angular frequency (ω)
angular (ω), 6
linear (f), 6
maximum, 167 168
phase and, relationship between, 174
Frequency domain, 101
Frequency-domain parallel imaging technique. See GRAPPA
Frequency encoding, 108 109 110 111 112 113
Frequency-encoding (Gx) gradient, 108 109 124 124 162 163 163
164 167 167 168
Frequency-encoding (readout) (Gx) gradient coils, 28
Frequency oversampling, 188
Frequency range, 168
3D-fresh blood imaging (FBI), 323 324
FSE. See Fast spin echo (FSE)
FSE imaging. See Fast spin-echo (FSE) imaging
FT. See Fourier transform (FT)
Full bandwidth, 178
Full echo train, 224 225 225
Function, mathematical, 6
Functional imaging, 339 340
G
Gadolinium (Gd), 21 206
Gauss, 23
Gaussian curve, 101 101 104 See also Gaussian RF pulse
Gaussian distribution, 177
Gaussian RF pulse, 103
bandwidth, 104 105
slice-select gradient, 105 106
Generalized autocalibrating partially parallel acquisition (GRAPPA), 279
Geometric distortion artifacts, 105 211 211 212
g-factor, 282 283
Ghost artifact(s), 198 199 198 199 208
Gibbs phenomenon. See Truncation artifacts
Glioblastoma multiforme (GBM), 342
Glx, 345
GMN. See Gradient moment nulling (GMN)
Gradient(s), 49 97 105
bi-lobed, 235 235 236
frequency-encoding (Gx), 108 109 162 163 164 167 168
high performance, 228 229 230
linear, 105 123 124
nonlinear, 105
nonlinearities in, artifacts caused by, 207 211 211
slice-select (Gz), 162 163 163
slice-selective (Gz), 123
slope of, slice thickness and, 98 98
Gradient and spin echo. See GRASE (gradient and spin echo)
Gradient coil(s), 26 27 28 49 98
frequency-encoding (readout) (Gx), 28
phase-encoding (Gy), 28
slice-select (Gz), 28
Gradient echo (GRE) imaging, 38
advantages and disadvantages of, 245
chemical shift artifact of second kind (Dixon effect), 242 243 244
243
 fast scanning techniques, 247
DE-prepared, 255
IR-prepared, 254 255 255 255
multiplanar, 253
nomenclature, 247 248
tissue-prepared, 254 255 256
for flow imaging, 256
and magnetic susceptibility, 237 237 238 239
multiplanar variant of, 247 252
PSD diagram for, 239
scan time for, 236
signal-to-noise ratio in, 242
slice excitation and, 236
three-dimensional, 244 245
advantages of, 244 245 245
PSD diagram for, 244 244
scan time for, 244
tissue contrast in, 238 239 240 241 242 239 240
Gradient-echo (GRE) pulse sequence, basic principles of, 233 234 235
236 237 238 239 240 241 242 243 244 245
gradient-echo (GRE) sequences, 356
Gradient moment nulling (GMN), 304
Gradient-recalled acquisition in the steady state (GRASS), 247 248 249
251 252 254
Gradient-recalled echo (GRE), 233 234 235 236 234 235 236 330
333 334
Gradient refocusing mechanism, 359
Gradient-related artifacts, 207 208 209 210 211
Gradient spoilers, 248 250
GRAPPA, 281 282
GRASE (gradient and spin echo), 354
GRASS. See Gradient-recalled acquisition in the steady state (GRASS)
Gray matter
proton density factor for, 62 62
T1 characteristics of, 60 60 62
T2 characteristics of, 61 61 62
GRE. See Gradient echo (GRE)
GRE imaging. See Gradient echo (GRE) imaging
Gyromagnetic ratio, 20 26
Gz. See Z gradient (Gz)
Gz gradient. See Slice-selective (Gz) gradient
H
Half-Fourier acquired single-shot turbo spin-echo (HASTE) sequences,
331
and fast spin echo, 332
Half NEX (1/2 NEX), 153 155 155
Hamiltonian, 201
dipolar, 201
HASTE. See Half-Fourier acquired single-shot turbo spin-echo (HASTE)
sequences
Head coils, 27
Hematoma, cerebral, and tissue contrast, 70 71
Hemorrhage
in brain, and tissue contrast, 70
in endometrioma, and tissue contrast, 73
Hemosiderin, 21 206
tissue contrast and, 72
Hermitian symmetry, 153 154
Hertz (Hz), 6
High performance gradient(s), 228 229 230 257
advantages of, 352 354
requirements for, 352
Hydration layer water, 58 59 58
Hydrogen proton
electromagnetic properties of, 18 19 19 20
natural motional frequency of, 57 58
I
Image
optimization of, scan parameters and, 176 177 178 179 180 181
182
real and imaginary components of, 155 156
Image construction
slice selection for, 97 98 99 100 101 102 103 104 105 106 107
Image contrast, parameters of, 176
Image processing
artifacts caused by, 185 186 187 188 189 190 191 192 193 194
195 196
defined, 123
Imaginary images
FT of, 156
Imaginary number(s), 5
Imaginary unit (i), 5
India Ink Etching, 195
Induced magnetic field
applied and, relationship between, 20
Integrated parallel acquisition technique (iPAT), 279
Interslice gap, 181 181
Intravoxel dephasing, 298
Inversion recovery (IR). See also STIR (short TI inversion recovery)
clinical applications of, 78 79 79
magnitude reconstruction, 79 80 80 81
null point in, 77 78 77
Inversion recovery (IR) pulse, 180 182
Inversion recovery (IR) techniques, 284 285 286 287 288
advantages of, 287
disadvantages of, 287 288
Inversion recovery pulse sequence, 76 77 78 79 76 77
Inversion time (TI), 78 79 78 79 80 182 See also STIR (short TI
inversion recovery)
IR. See Inversion recovery (IR)
Iron (Fe), ferromagnetism in, 21
J
“J-coupling,” 342
Jerk compensation, 308
K
Keyhole imaging, 226
Key points
of artifacts, 213
of data space, 161
of fast spin-echo, 231
of field of view (FOV), 169
of Fourier transform, 96
of gradient-echo (GRE), 246
of image construction, 106
of k-space, 174
of mathematical concepts, 13 14
of physical principles of magnetic resonance imaging, 29 30
of pulse sequence diagram (PSD), 166
of pulse sequences (saturation, partial saturation and inversion
recovery), 81
of radio frequency (RF) pulse, 38
of scan parameters and image optimization, 183
of signal processing, 143
of spatial encoding, 121
of spin echo (SE) pulse sequence, 88 89
of T1, T2, T2*, 46
of tissue contrast, 65
of TR, TE, and tissue contrast, 55
k-space, 117 118 119 126 127 145 146 145 171 172 173 174
analog version of. See Data space
center of, 147 148 147 148
conjugate (Hermitian) symmetry, 153 154
in conventional spin echo (CSE), 217 218
digitized, 140 141 141
edges of, 149 150 149 150
example of, 157 158 159 160 157 158 159
Fourier transform (FT) of, 171 173 174
and image, 171 172 173 172
image construction, 150 151 151
images of, 148 149 149
 properties of, 147 148 149 150 151 152 153
real and imaginary, FT of, 155 156
symmetry, 151 152 153 152 153
three-dimensional line drawing of, 151 152
unit interval in, 171 172 173 174 172
k-space averaging, 227
L
Laminar flow, 294 295
Lanthanide group, 21
Larmor equation, 26 168 174
derivation of, 36
linear frequency f in, 26
Larmor frequency (ω0), 36 49 57 98 290 See also Angular frequency
(ω)
magnetic field strength and, 98 99 98
Linear frequency (f), 6 17
in Larmor equation, 26
Linear gradient(s), 123 124
Lipids
and radiation necrosis, 348
Logarithms, 12
properties of, 13
Longitudinal magnetization (Mz), 36 37 38 233 234 235
after time interval TR, 52 52
recovery of, 41 41
RF pulse and, 40 41 40 41 48 48 52
saturation recovery and, 74
T2 relaxation time and, 41 41
Longitudinal relaxation time, 40 See also T1 relaxation time
Low bandwidth, 276
Lower frequency
cosine wave of, 136
Low-pass filters (LPF), 134 135 136 134 136 137 272
LPF. See Low-pass filters (LPF)
M
Magic angle artifacts, 200 201 202
Magnet(s)
design of, 23
field strength of, 23
permanent, 23
resistive, 23
superconducting, 23
types of, 23
Magnetic dipole moment (MDM), 20 20 23 24 24
in Larmor equation, 36
Magnetic field (B), 25 33
of electromagnetic waves, 17 17 18
external, 21 23 24 31 31 32 33
inhomogeneity, and dephasing, 42 43 83
Magnetic field intensity (H), 20
Magnetic field strength
Larmor frequency and, 98 99 98
Magnetic flux density (B), 20
Magnetic induction field (B), 20
Magnetic permeability (µ), 20
Magnetic resonance angiography (MRA)
black-blood, 323
contrast-enhanced, 315 354
advantages, 316
disadvantages, 316
phase contrast (PC), 311 312 313
advantages of, 313
disadvantages of, 314
2D- versus 3D-, 314
time of flight (TOF), 310
Magnetic resonance imaging (MRI)
definition of, 16
perfomance of, 21 22 23
physical principles of, 16 17 18 19 20 21 22 23 24 25 26 27 28
29
Magnetic susceptibility artifact(s), 206 207 208 209 210
Magnetic susceptibility (χ), 20 206 207 209 228
in diamagnetic substances, 21
in ferromagnetic substances, 21
and gradient-echo (GRE), 237 237 238 239
in paramagnetic substances, 21
Magnetism
diamagnetism, 21
ferromagnetism, 21
paramagnetism, 21
Magnetizability, measure of, 20
Magnetization. See also Net magnetization
longitudinal (Mz), 36 37 38 40 40 41 41 52 52 74 233 234 235
recovery of, 25
transverse (Mxy), 40 41 40 41 41 45 48 48
residual, in gradient echo (GRE) imaging, 238 239
Magnetization (cont.)
steady-state (Mss), in gradient echo (GRE) imaging, 238 239
Magnetization transfer (MT), 290 291 292
effect in FSE, 290
in fast spin-echo (FSE) imaging, 228
saturation, 322
Magnetization vector, 31 32 33 34 35 36 37 38 34 35 41 44 48 50
51 74 76 77 78 83 84 255
Magnitude
angle and, 5 6
of signals, 125 146 151 153
of the sine wave, 11
of vector, 6
Magnitude image (modulus), 6 155 156 157 156
Magnitude reconstruction, inversion recovery, 79 80 80 81
Mathematical concepts, of MRI physics, 3 4 5 6 7 8 9 10 11 12 13
Maximum frequency, 167 168
Maximum intensity projection (MIP), 319
disadvantages of, 319
Maxwell's wave theory, electromagnetic waves and, 16 17
MDM. See Magnetic dipole moment (MDM)
Meningioma
fast spin-echo (FSE) imaging of, 229
tissue contrast in, 67
Methemoglobin
in tissue contrast, 70 71
Microwaves, characteristics of, 16
Minimum intensity projection algorithm, 323
Mipping, 319
Mobile protons, magnetization and, 25
Modulation, of frequency, 105 106
Modulus. See Magnitude image (modulus)
Moiré fringe, 206 208
Motion
artifacts, 147 197 198 199 200 201 198 199 200 201
blood, 331
cardiac, 329 330 331
in cardiac imaging and solutions, 331
periodic, 197 198 199
random, 197 199 200 199 200
respiratory, 327 328 329
Motion-correction, 329
MOTSA, 321
MP-RAGE, 254
MRI. See Magnetic resonance imaging (MRI)
MRI pulse sequence, 356
MR signal
radio frequency (RF) and, 22
MT. See Magnetization transfer (MT)
Mucinous fluids, 59
Multi-echo fast spin-echo (FSE) imaging, 224 225 226 227 228
Multiple overlapping thin-slab acquisition. See MOTSA
Multislice acquisition, 127 127
Multivoxel techniques, 344 345
Myocardial tagging, 337
Myoinosotol, 345
N
Narrower frequency bandwidth
slice thickness and, 101 102
Natural logarithm, 12
Natural motional frequencies, of protons, 57
Navigator-echo gating, 328
Necrotic lymphoma, 349
Net magnetization, 24 24 25 25 31 32 34
Net magnetization vector (M0), 31 32 35
spiral motion of, 33
in x-y plane (Mxy), 35 35 37 38 38
Newtonian physics, 16
NEX. See Number of excitations (NEX)
NFW. See No frequency wrap (NFW)
N/2 ghost artifact(s), 261
Nickel (Ni), 21
NMR. See Nuclear magnetic resonance (NMR)
No frequency wrap (NFW), 188 271 272 272
No phase wrap (NPW), 188 188
NPW. See No phase wrap (NPW)
NSA. See Number of signal averages (NSA)
Nuclear magnetic moment, 21
Nuclear magnetic resonance (NMR), 16
Nuclear paramagnetism, 21
Nucleus
energy states of, 19
even number of protons in, 19
MDM in, 20
odd number of protons in, 19
paired proton in, 19
Null Point, in inversion recovery, 77 78 77
Number of excitations (NEX), 141 165 177
½, 153 155 155
¼, 155 155
Number of phase-encoding steps (Ny ), 177
Number of signal averages (NSA), 165
Nutation (spiral motion), 33 35
Ny (number of phase-encoding steps), 177
Nyquist frequency, 178 271 354
Nyquist law. See Nyquist sampling theorem
Nyquist sampling theorem, 123 139 140 139 140
O
Oblique planes, 29
Odd-echo dephasing, 298
Odd function, 153
Odd function, Fourier transform (FT) of, 94 94
Osteomyelitis, tissue contrast in, 68
Out-of-phase scanning, 195
Overhead time (To), 125
Oversampling, 140 188
Overscanning, 156 188
Oxyhemoglobin, tissue contrast in, 70
P
Paired protons, 19
Parallel imaging
advantages, 281
autocalibration, 280 281
concept, 279
disadvantages, 281
prescan, 280
Paramagnetic substances
deoxyhemoglobin and, 21
dysprosium (Dy), 21
gadolinium (Gd), 21
hemosiderin, 21
increased magnetic field in, 21
magnetic susceptibility in, 21
methemoglobin and, 21
in MRI, 21
same direction, magnetic field in, 21
Paramagnetism, 21 206 209
nuclear, 21
Partial flip, 38 38
Partial flip angle techniques, 233
Partial saturation pulse sequence, 74 75 75
Partial volume artifacts, 196 197
Pelvis
fast spin-echo (FSE) imaging of, 229
Perfusion imaging, 337 339
in echo planar imaging (EPI) technique, 263 264 265
Periodically rotated overlapping parallel lines with enhanced
reconstruction (PROPELLER), 276 277
Periodic motion, 197 198 199
Period (T), 6 6
Permanent magnets, 23
Phase, 7 8 7
frequency and, relationship between, 174
Phase construction, 79
Phase contrast imaging, 330
Phase contrast (PC) MRA, 311 312 313
advantages of, 313
disadvantages of, 314
2D- versus 3D-, 314
Phased-array coils, 28
Phase difference, 4 163 164
Phase encoding, 113 114 115 116 117 118 119 120 114 115 116
117 118 123 124 125 126
defined, 125
motion artifact and, 198 199
symbol for, 123 125
TR (repetition time) and, 117 118 119 118 119
Phase-encoding (Gy) gradient, 113 114 115 116 117 118 119 113
114 115
coils, 28
Phase image, 155 156 157 156
Phase offset, 7 7
Phase-offset RF pulses, 272 273 273
Phase oversampling, 188
Phase-sensitive detection, 151
Phase shift, 7 8 115 116 117 158 158
Photography, 21 22
Pixel, 108
bandwidth of, 189 190
Pixel bin, 190 194
Pixel size
and chemical shift artifact, 190 191 194
Planck's constant (h), 36
Planck's law, 36
Plane of imaging, 28 29 29
oblique, 29
x, y, and z axes for, 28 28 29
Plug flow, 294 295
Postgadolinium imaging, 331 354
Precession, 25 26 26
frequency of, 33
 out of phase, 34
in phase, 34 36
RF pulse and, 31 32 33
Prep pulse, 356 359
Prescan, 38
PRESS (point-resolved spectroscopy) technique, 348
Propagation factor (C), of electromagnetic waves, 17 17
Proteinaceous material
T1 characteristics, 58 59 60 61 62 63 64 59 65
T2 characteristics, 57 59 60 65
Proteinaceous solutions, 58 59 60 61 62 63 64
Proton density, magnetization and, 25
Proton density factor: N(H), 61 62 62
Proton density-weighted image, 76 87 88 88
Proton magnetic resonance spectroscopy (MRS), 342
nuclear magnetic resonance (NMR) and, 342
Proton(s)
energy states of, 19 19 20 36
ensemble of, FID and, 49
even number of, 19
mobile, 25 51
natural motional frequencies of, 57
odd number of, 19
imaging in MR, used for, 20
paired, 19 20
precession of, 25 26 26 31 32 33 34
out of phase, 34
in phase, 34 36 37 40 40 41 43
unpaired, 20 20
wobble of, 25 26 26
Proton spins, direction of, 19
PR (projection reconstruction) acquisition, 316
PSD. See Pulse sequence diagram (PSD)
PSIF. See Steady-state free precession (SSFP)
Pulse sequence diagram (PSD), 74 162 163 164 165 166 310
for 3D gradient-echo (GRE) imaging, 244 244
in echo plannar imaging (EPI), 259 260
 for SE pulse sequence, 108 124
spin-echo (SE), 162 163 164 162 163 164 165
Pulse sequence(s), 74 75 76 77 78 79 80 81 83 84 85 86 87 88
See also Radio frequency (RF) pulses
definition of, 74
inversion recovery, 76 77 78 79 76 77
partial saturation, 74 75 75
saturation recovery, 75 76 76
spin-echo (SE), 83 84 85 86 87 88
Pythagorean theorem, 6
Q
Qp/Qs ratios (pulmonary to systemic flow ratio), 340
Quadrature coils, 28
Quantum mechanics, 16 19
Questions
on artifacts, 213 214
on data space, 161
on fast spin-echo, 232
on field of view (FOV), 169 170
on Fourier transform, 96
on gradient-echo (GRE), 246
on image construction, 107
on k-space, 175
on mathematical concepts, 14 15
on physical principles of magnetic resonance imaging, 30
on pulse sequence diagram (PSD), 166
on pulse sequences (saturation, partial saturation and inversion
recovery), 81 82
on radio frequency (RF) pulse, 39
on scan parameters and image optimization, 183 184
on signal processing, 144
on spatial encoding, 121 122
on spin echo (SE) pulse sequence, 89
on T1, T2, and T2*, 47
on tissue contrast, 73
on TR, TE, and tissue contrast, 55 56
R
Radiation necrosis
and lipids, 348
Radio frequency (RF), 11 18
Radio frequency (RF) pulses, 11 12 12 18 22 22 23 31 32 33 34 97
257 359 See also Flipping, partial
90°, 35 36 37 37 48 48 49 74 74 75
180°, 356
180°, 37 38 76 83
refocusing (rephasing), 83 85 87
auto, 38
center frequency of, 129 130 130
definition of, 31
frequency of, 33 34
FT of, 130 130
Gaussian. See Gaussian RF pulse
magnetic field generated by (B1), 32 33
magnetic resonance signal, 22 23
nonselective, 103
selective, 103
sinc, 103 103
Radio waves, characteristics of, 16
Random motion, 197 199 200 199 200
Rapid acquisition through parallel imaging design (RAPID), 279
Rapid acquisition with relaxation enhancement (RARE), 217 See also
Fast spin-echo (FSE) imaging
RARE. See Rapid acquisition with relaxation enhancement (RARE)
Rare earth elements, unpaired electrons in, 21
Readout, 359 360 361 362
Readout gradient. See Frequency-encoding (Gx) gradient
Real images, 155
FT of, 156
Received signal (S), 43 44 49 49 51 51
Receiver bandwidth, 140
Recovery curves. See Decay/recovery curves, TR (repetition time) and
Refocusing mechanism, 83 85 87 162 163 165 359
Relative anisotropy, 263
Relaxation, definition of, 40
Relaxation rate, 46
Relaxation time. See also T1 relaxation time; T2 relaxation time; T2*
relaxation time
longitudinal, 40
spin-lattice, 40
spin-spin, 42
transverse, 42
Repeated excitation radio frequency (RF) pulses, 320
Repetition time. See TR (repetition time)
Rephasing, 83 85 87 163
even-echo, 298 299 300
Resistive magnets, 23
Resolution
parameters of, 176
and signal-to-noise ratio (SNR), 179 180
Resonance, 33 34
Respiratory compensation (ROPE and COPE), 328 329
Respiratory motion, 327 328 329
Rewinder gradient, 230
RF. See Radio frequency (RF)
RF coil, 26
RF pulses. See Radio frequency (RF) pulses
RF (radio frequency) noise, 205 206
Right-hand rule, 44 44
Ring-down effect, 94 95
Ripples effect, 104
Root mean square (rms), 79
Root pulsing sequence, 260
Rotating frame of reference, 34 35 36 37
S
Sampling, 120 121 135 136 137 138 139 135 136 137 138 139
of composite signals, 141 142 143 142
Sampling interval (ΔTs), 171
Sampling time (Ts), 124 140 163
versus sampling interval, 139 140 140
Sat Pulses. See Spatial saturation pulses (Sat Pulses)
Saturated, 300
Saturation, 74 75
partial, 74 75
and prevention of aliasing, 188
Saturation effects, 320 321
Saturation recovery pulse sequence, 75 76 76
Scanners, recent advances in, 268 269 270 271 272 273 274 275
276 277
Scanning, recent advances in, 268 269 270 271 272 273 274 275
276 277
Scan parameters
primary, 176
secondary, 176
Scan time. See also Acquisition time
for conventional spin echo (CSE), 217
for gradient-echo (GRE), 236
SE. See Spin-echo (SE)
SE-EPI, 261 262 263
SE imaging. See Spin-echo (SE) imaging
Sensitivity encoding (SENSE), 279 281 283
SE pulse sequence. See Spin-echo (SE) pulse sequence
Sequential scanning, 236
SEW. See Slice-excitation wave (SEW)
Shared echo approach, 225 226 227 228 226 227
Shim coils, 23 28
Shimming coils (auto shimming), 206
Short TI inversion recovery (STIR), 81 81 182
Shot, 257
Signal, 6 7
definition of, 44
measurement of, 49
oscillating, 45
periodic, 6 7
received, 43 44 45 51 51
sinusoidal, 46 46 49
Signal intensity, 53 54
Signal loss, high velocity, 296 297 297
Signal processing, 123 124 125 126 127 128 129 130 131 132 133
134 135 136 137 138 139 140 141 142 143 144
defined, 123
Signals
amplitude of, 109 110 151
composite, sampling of, 141 142 143 142
digitized sample, 146
direction of, 153
magnitude of, 125 146 151 153
Signal-to-noise ratio (SNR), 27 28 141 148 148 176 177 279 282 283
287
in 3D imaging, 178 179 180
parameters of, 176
resolution and, 179 180
Similarity transform, 263 263
Simultaneous acquisition of spatial harmonics (SMASH), 279
sinc function, 11 11 12
Fourier transform (FT) of, 92 93 94 93 94
FT of, 130 131 131 150 150
Sinc RF pulse, 103 103
sine data
data space with, 152 153 153
sine function, 94 153 154 155
Single-shot technique, 331 334
Single voxel techniques, 342 343 344
Sinusoidals, 3 4 4 5
exponentially decaying, 11 11 49 49
Slew rate (SR), 257 352
Slice excitation, gradient-echo (GRE) and, 236
Slice-excitation wave (SEW), 303
Slice(s)
contiguous, 128
gaps between, 128
interleaved, 128
number of (coverage), 128 129 223 224
sequential, 128
Slice-select gradient coils (Gz), 28
Slice-select gradient (Gz), 103 105 106 162 163 163
Slice selection, 128
Slice-selective (Gz) gradient, 123
Slice thickness, 100 100
cross-talk, 101 101
narrower frequency bandwidth and, 101 102
selection of, 97 97
slice-select gradient, 103
slope of the gradient and, 98 98
ways to change, 101 102 103
Slice-tracking, 329
Small bowel, tissue contrast in, 73
SNR. See Signal-to-noise ratio (SNR)
Solenoid coils, 28
Solids
T1 characteristics of, 58 59 65
T2 characteristics of, 57 59 60 65
Solid tumors, 349
SPAMM. See SPAtial Modulation of the Magnetization
Spatial encoding, 22 23 28 49 97
Spatial frequencies, 145
Spatial frequency domain, 120 173
Spatial frequency(ies), 173 174
SPAtial Modulation of the Magnetization, 337
Spatial presaturation, 289 290 290
Spatial saturation pulses (Sat Pulses), 274 275
Spectral density of noise, 177
Speed of light, of electromagnetic waves, 16
Spikes
FT of, 131 133
sequence of, 131 133
Spin(s), 18 19 20
interactions between, 42 43
out of phase, 34 42 43 83 84
out phase, 125 126
in phase, 34 41 43 83 84 125 126
Spin density, magnetization and, 25
Spin dephasing, 43 44
Spine, lumbosacral, tissue contrast in, 68 69
Spin-echo (SE) imaging, 83 See also Conventional spin echo (CSE); Fast
spin-echo (FSE) imaging
aliasing (wraparound) in, 185 186 186
contrast on, 182 182
Spin-echo (SE) pulse diagram, 83 84 85 86 87 84 85 86
Spin-echo (SE) pulse sequence, 83 84 85 86 87 88 146
asymmetric echoes in, 87
diagram of, 146
pulse sequence diagram (PSD) for, 108 124 162 163 164 162 163
164 165
symmetric echoes in, 87 87
Spin-lattice relaxation time. See T1 relaxation time
Spinning charged particle, 18 19 20 18 20 20
Spin quantum number (S), 19 20
Spin-spin interaction(s), 42 43 46 83
“spin-spin splitting,” 342
Spiral motion, of net magnetization vector (M0), 33
Split echo train, 225 225
Spoiled GRASS images (SPGR), 247 248 249 250 251
characteristics of, 256 256
disadvantages of, 251
Fast, 252 253
multiplanar (MPSPGR), 252
tissue contrast, 248
Spoiling
defined, 247 248
 RF (radio frequency), 248 249 250
Steady-state free precession (SSFP), 247 251 252
advantages of, 251
characteristics, 256 256
disadvantages of, 252
tissue contrast, 251
STIR. See Short TI inversion recovery (STIR)
STIR (Short TI Inversion Recovery), 284 285
Subtraction technique, 320
Superconducting magnets, 23
Superparamagnetic substances, 21
Superparamagnetism, 206
Suppression. See also Tissue suppression
Surface coil(s), 27 187 188
Susceptibility artifact(s)
in EPI, 261 262
T
Tangent, 3 6 6 9
T1 characteristics
of cerebrospinal fluid (CSF), 60 60 62 66 68
of fat, 58 59 65 68 69
of gray matter, 60 60 62
of proteinaceous material, 65
of proteinaceous solutions, 58 59 60 61 62 63 64 59 73
of solids, 58 59 65
of various tissues, 65
of water, 58 59
of white matter, 60 60 62 66
T1 weighting, 75 87 88 88
T2 characteristics
of cerebrospinal fluid (CSF), 60 61 62 66 68
of fat, 57 59 60 62
of gray matter, 61 61 62
of proteinaceous material, 57 59 60 65
of proteinaceous solution, 65 73
of solids, 57 59 60 65
of various tissues, 65
of water, 57 59 60
of white matter, 61 61 62 66
T2 weighting, 87 88 88
TE (echo delay time or time to echo), 51 52 53 52 85 86
in partial saturation, 74 75 75
saturation recovery, 75 76 76
and T1 effect, 60
and T2 effect, 60
and T2* effect, 54 54
TE (echo delay time or time to echo) (cont.)
and tissue contrast in, 63 64 63 64
lengthening of
cause of, 181 182 182
effects of, 181
reduction of
 effects of, 181
method for, 182
TEeff. See Effective TE (TEeff)
Tesla, 23
Three-dimensional (3D) acquisition, 273 274
Three-dimensional (3D) fast spin-echo (FSE) imaging, 230 231 230 231
Three-dimensional (3D) imaging
aliasing (wraparound) in, 186 187 188 187 188
signal-to-noise ratio (SNR) and, 178 179 180
Three-dimensional gradient-echo (GRE) imaging, 244 245
advantages of, 244 245 245
PSD diagram for, 244 244
scan time for, 244
TI. See Inversion time (TI)
Tilted optimized non-saturating excitation (TONE). See TONE (tilted
optimized nonsaturating excitation)
Time constant(s) (τ), 9 10
Time domain, 120
data space in, 120
Time-of-flight loss, 302 332
Time of flight (TOF) MR angiography, 256 290
Time-resolved imaging of contrast kinetics (TRICKS), 316
Time to echo. See TE (echo delay time)
Timing diagram. See Pulse sequence diagram (PSD)
Tissue contrast, 53 54 53
clinical application of, 57 58 59 60 61 62 63 64
in gradient-echo (GRE), 238 239 240 241 242 239 240
in spin-echo (SE) pulse imaging, 87 88 88 88
T1, 53 54 53 57 58 59 60 61 62 63
T2, 53 54 54
T2*, 54 54
Tissue preparation, 254
Tissue suppression. See also Fat, suppression of techniques for, 284
285 286 287 288 289 290 291 292
TI (time fo inversion). See Inversion time (TI)
To. See Overhead time (To)
TOF MRA, 310
TONE (tilted optimized nonsaturating excitation), 321 322 323 323
Toxoplasmosis, 349
Trade-off, 203
Transmission bandwidth, 140
Transmit bandwidth, 129 130
Transmit/receive coil(s), 27 44 45
Transverse magnetization (Mxy), 35 35 37 38 38
Transverse (Mxy) magnetization
with decay rate of T2, 41 41
received signal and, 45
residual, in gradient echo (GRE) imaging, 238 239
RF pulse and, 40 41 41 48 48
spiral-like decay of, 45
steady-state (Mss), in gradient echo (GRE) imaging, 238 239
T1 relaxation time and, 40 41 40 41
T2 relaxation time and, 41 41
T1 relaxation time, 24 25 40 41 41 43
T2 relaxation time, 41 42 43 41 43
T2* relaxation time, 46
Trigger delay, 330
Triple inversion recovery (IR), 356
“triplet,” 342
TR (repetition time), 50 51 50 62 128
decreased
effects of, 180 181
increased
effects of, 180
long, for multiplanar, 252
maximum number of slices, 129
in partial saturation, 74 75 75
and phase encoding, 117 118 119 118 119
saturation recovery, 75 76 76
and T1 effect, 60
and T2 effect, 60
and tissue contrast in brain, 63 64 63 64
True fast imaging with steady-state precession (True FISP), 334
“True” interleaving technique, 203
Truncation artifacts, 189 195 196 196 197
Ts. See Sampling time (Ts)
TSE. See Turbo spin echo (TSE)
Turbo spin echo (TSE), 217 217 See also Fast spin-echo (FSE) imaging
Turbulent flow, 294 295 298
Two-dimensional fourier transform (2DFT), 113
U
Uncooperative patient, 357
Undersampling, 188
and aliasing, 138 139
Unpaired electron
hemosiderin, at the end stage of hemorrhage, 21
rare earth elements and, 21
Unpaired protons, 20
Unsaturated, 300
V
Variance (σ2) of Gaussian distribution, 177
Vector(s), 4 5 5 12 12
Velocity encoding. See VENC
VENC, 312 313 340
Venetian blind, 321
Vertebral bodies
and chemical shift artifact, 191 191 192 193
View sharing, 337
Views per segment (VPS), 331
Visible light, characteristics of, 16
Vortex flow, 294 295
Voxel, 108
dimensions of, 177 177
size of, and signal-to-noise ratio (SNR), 177 178
Voxel volume, and signal-to-noise ratio (SNR), 177 178 178
W
Water
bound to hydrophilic macromolecule, 58
bulk phase, 58
clinical applications of suppression, 291 292
diamagnetism of, 21
hydration layer water, 58 58
hydrogen in, 20
T1 characteristics of, 58 59 65
T2 characteristics of, 58 59 65
Waveform(s), 6
narrow, 99 99
wide, 99 99
White matter
T1 characteristics of, 60 60 62 66
T2 characteristics of, 61 61 62 66
Wobble, 26
Wraparound. See also Aliasing
definition of, 186
X
X-ray, characteristics of, 16
X-ray imaging, 22 22
Z
z gradient (Gz), 103
Zipper artifact(s), 203 204 205 204
feedthrough, 205 205