Halse Centric Scan SPRITE MRI

Download as pdf or txt
Download as pdf or txt
You are on page 1of 11

Journal of Magnetic Resonance 165 (2003) 219–229

www.elsevier.com/locate/jmr

Centric scan SPRITE magnetic resonance imaging


Meghan Halse, David J. Goodyear, Bryce MacMillan, Pavol Szomolanyi,
David Matheson, and Bruce J. Balcom*
Department of Physics, MRI Centre, P.O. Box 4400, University of New Brunswick, Fredericton, NB, Canada E3B 5A3
Received 18 June 2003; revised 11 August 2003

Communicated by Joseph Ackerman

Abstract

Two rapid, pure phase encode, centric scan, Single Point Ramped Imaging with T1 -Enhancement (SPRITE) MRI methods are
described. Each retains the benefits of the standard SPRITE method, most notably the ability to image short T2 systems, while
increasing the sensitivity and generality of the technique. The Spiral-SPRITE method utilizes a modified Archimedean spiral k-space
trajectory. The Conical-SPRITE method utilizes a system of spirals mapped to conical surfaces to sample the k-space cube. The
sampled k-space points are naturally Cartesian grid points, eliminating the requirement of a re-gridding procedure prior to image
reconstruction. The effects of transient state behaviour on image resolution and signal/noise are explored.
Ó 2003 Elsevier Inc. All rights reserved.

Keywords: MRI; Materials; Short T2 ; SPRITE; Spiral trajectory

1. Introduction age acquisition at a greater speed and minimizes gradient


vibration arising from impulsive Lorentz forces. The
Single point imaging (SPI) MRI methods have pro- SPRITE sequence enables the introduction of T1 contrast
ven to be a useful imaging modality for studies of short through partial saturation of magnetization components
relaxation time systems [1–4]. Unlike frequency encod- which have longitudinal relaxation times that are longer
ing, where the time evolution of magnetization is mea- then the gradient switching time [6]. A wide variety of
sured, SPI is a pure phase encode technique. Images magnetization preparation techniques may be employed
acquired using SPI methods are free from artifacts due to control image contrast or permit relaxation time
to B0 inhomogeneity, chemical shift, and susceptibility mapping. For example, Mastikhin et al. [7] have shown
variations [5]. In SPI radio frequency (RF) pulses are that SPRITE with centric half k-space sampling may be
applied in the presence of the phase encode gradient, G, used to obtain images with T1 and T1q weightings
and a single point is acquired during the free induction through magnetization preparation. Additionally, Beyea
decay (FID) at a time tp commonly referred to as the et al. [3] and Axelson et al. [8] have used SPI methods for
encoding time. The acquisition is regarded as inefficient relaxation time mapping of short T2 nuclei.
since only one k-space point is acquired for each gra- SPI sequences rely on non-selective broadband RF
dient switch on/off. Any attempt to increase acquisition pulses. In order to obtain homogeneous excitation of the
speed can result in excessive gradient vibration due to spin system, the duration of the pulses are limited by the
rapid gradient switching. maximum spectral width of the object (Gmax  sample
The SPRITE (Single Point Ramped Imaging with T1 length). In SPI this restriction calls for small RF flip
Enhancement) sequence [6], a development of SPI, em- angles and therefore, in some cases, requires extensive
ploys ramped phase encode gradients which permits im- signal averaging to obtain acceptable S/N. This may be
acceptable for systems where longitudinal relaxation
*
Corresponding author. Fax: +1-506-453-4581. times are sufficiently short. However, for materials
E-mail address: bjb@unb.ca (B.J. Balcom). with long T1 s, extensive averaging leads to prolonged

1090-7807/$ - see front matter Ó 2003 Elsevier Inc. All rights reserved.
doi:10.1016/j.jmr.2003.08.004
220 M. Halse et al. / Journal of Magnetic Resonance 165 (2003) 219–229

acquisition times. Increasing acquisition speed results in must be constant [14]. Nyquist sampling requires the in-
excessive saturation of longitudinal magnetization which terval along the radial direction of k-space, for M spiral
leads to low image intensity. Using a centric sampling interleaves and a given image field of view, FOV, to be
acquisition scheme, the z-magnetization could be sam- M
pled more efficiently, permitting high quality image ac- Dkr ¼ ; ð2Þ
FOV
quisition which is impossible for SPI. Mastikhin et al.
[7] have shown that SPRITE with half k-space centric where the FOV can be defined as
order sampling may be used to image long T1 materials, 2MNr Dr ¼ FOV ð3Þ
with acceptable image acquisition times and S/N. and Dr is the nominal pixel resolution of the image
Since the development of SPRITE, the original [14,15]. Using the notation of Ahn et al. [15] we define Nr
methodology, we have considered alternate k-space as the number of rotations of the spiral trajectory and Nh
trajectories to increase sensitivity (S/N and acquisition is defined as the number of sampled points on one ro-
speed) of the technique. In this paper we outline two tation of the spiral. The angular sampling interval along
new pure phase encode centric sampling MRI acquisi- the k-space trajectory can be written as
tion strategies. Spiral-SPRITE [9] employs pure phase
encode, centric sampling along a modified Archimedean 2p
Dh ¼ : ð4Þ
spiral trajectory. The Spiral-SPRITE technique has been Nh
the subject of a recent conference publication [10] as- The radial and angular k-space sampling intervals are
sociated with the 5th International Meeting on Recent depicted in Fig. 1. For the single point approach to
Advances in MR Applications to Porous Media. Coni- spiral scan imaging, we use a formulation which is
cal-SPRITE [11], a 3D centric sampling technique, em- similar to that used for frequency encoding methods. We
ploys pure phase encoding with a system of spiral consider the construction of one spiral assuming that
trajectories mapped to conical surfaces, to sample k- multiple interleaved spiral trajectories, if desired, can be
space. These techniques have several natural advanta- determined through a rotation of coordinates. We allow
ges: simplified image contrast, increased S/N, reduced for the possibility of spiral interleaves in the variable M.
acquisition times, and reduced gradient duty cycle, In this work M is always 1.
which are the subject of this communication. In Cartesian coordinates the kx and ky components of
Spiral k-space sampling has become a common pure the spiral trajectory can be written as
frequency encode methodology in biomedical imaging.
kx ¼ afc ðhÞ;
The nested cones trajectory has been suggested in con- ð5Þ
junction with frequency encoding by Nishimura and ky ¼ afs ðhÞ;
Irarrazabal [12]. Frequency encode spiral trajectories where fc ðhÞ and fs ðhÞ are functions defined by
require re-gridding of the experimental k-space data
points prior to image reconstruction [13]. This data
processing step is avoided in Spiral-SPRITE and Coni-
cal-SPRITE since the acquired k-space data points quite
naturally fall on Cartesian grid points.

2. Theory

2.1. The spiral-SPRITE trajectory

Spiral-SPRITE is the foundation for Conical-


SPRITE; therefore, we shall first present the Spiral-
SPRITE trajectory and introduce a common notation
for these centric scan techniques.
The traditional Archimedean spiral k-space trajectory
in polar coordinates is given by
kr ¼ ahk^; ð1Þ
where h is the polar angle measured with respect to the kx -
axis, a is a constant which defines the rate of change of the
Fig. 1. An illustration of Spiral k-space sampling. The sampling in-
spiral trajectory and along the radial direction in k-space, terval Dkr is the radial distance between consecutive rotations of the
and k^ is the unit vector directed along kr . For constant spiral trajectory. Dh is the angular spacing between consecutively
radial sampling the derivative of jkr j with respect to h sampled k-space points, as defined by Eq. (4).
M. Halse et al. / Journal of Magnetic Resonance 165 (2003) 219–229 221

h Using Eqs. (9) and (11) we find that the k-space coor-
fc ðhÞ ¼ cos h; dinates of the spiral trajectory after n pulses can be
hmax
ð6Þ written as
h
fs ðhÞ ¼ sin h;
hmax M n1 2pðn  1Þ
kx ¼ 2p cosð Þ;
where hmax is the maximum angle of the trajectory which FOV Nh Nh
ð12Þ
is dependent on the number of spiral rotations, Nr , of M n1 2pðn  1Þ
ky ¼ 2p sinð Þ:
the spiral trajectory and the range of h is such that FOV Nh Nh
0 6 h 6 hmax ; hmax ¼ 2pNr : ð7Þ We project the spiral coordinates onto a square
The constant, hmax , has been introduced so that the equispaced W  W Cartesian grid where W is typically a
sum of the squares of the functions is equal to unity when power of 2 so the FFT may be employed for rapid image
evaluated at the last point on the spiral trajectory (i.e., at reconstruction. The range of kx and ky are defined by
hmax ). This formulation leads to the determination of the jkx j 6 kmax and jky j 6 kmax . We round (kx , ky ) to nearest
constant a. At the extremity of k-space we find neighbour grid points (kx ; ky ) given by
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi c
kmax ¼ kx2 þ ky2 ¼ a: ð8Þ kx ¼ Gx tp ;
2p ð13Þ
c
However, kmax ¼ Nr Dkr , which is just the number of ky ¼ Gy tp ;
2p
spiral rotations multiplied by the k-space step in the
where c is the nuclear gyromagnetic ratio, Gx and Gy are
radial direction. Using Eqs. (2)–(8), the spiral k-space
magnetic field gradient amplitudes, and tp the encoding
coordinates can be further written as
time. The magnetic field gradient amplitudes used for the
M h spiral scan technique are then calculated using Eq. (12).
kx ¼ cos h;
FOV 2p ð9Þ We now consider how one would choose Nr and the
M h Nh given that the final image will be on a square equi-
ky ¼ sin h:
FOV 2p spaced grid. Since we are assuming that the image has a
For traditional spiral imaging techniques, the vari- final dimension of W  W with an equivalent field of
able h would be time dependent. It typically depends on view in each dimension, then the nominal resolution
hardware constraints such as gradient amplifier instan- along the x- or y-directions of the image is simply the
taneous peak current, slew rate, and RMS current, FOV divided by the number of points along one di-
which determine the rate at which the spiral trajectory is mension of the grid. Using Eq. (3) the approximate
traversed during data acquisition [14]. The constraints number of spiral rotations for a given number of spiral
are important in conventional spiral imaging where interleaves is
spiral k-space trajectories permit acquisition of fre- W
quency encoded signals with long inherent T2 s using a Nr ¼ : ð14Þ
2M
minimal number of RF excitations. In spiral single point In order to determine Nh we consider the following ar-
imaging, h is also time dependent. However in Spiral- gument. The spiral trajectory, including interleaves, lies
SPRITE the speed at which we scan k-space is not de- within a circle with radius kmax centered on the origin.
termined by the gradient amplifier slew rate. Instead the The ratio of the circular area to the area of the square k-
rate is governed by the magnetic field gradient rise time, space grid is
or how fast phase encode gradients can be switched
2
from one level to another, and the repetition time be- pkmax 3
 :
2
ð15Þ
tween gradient switches. At an arbitrary time t after the ð2kmax Þ 4
initial RF pulse the number of pulses can be written
  The total number of spiral points, including the points
t from spiral interleaving, is M  N and would be con-
n¼ þ 1; ð10Þ
TR tained within the circular area. The ratio of the total
number of spiral points to the total grid points is ap-
which is defined in terms of a nearest integer function. proximately equal to the ratio of the areas presented in
Here TR is the repetition time between RF pulses. Each Eq. (15). For an equally spaced grid the approximate
k-space point on the spiral trajectory corresponds to a number of grid points contained within the circular area
single RF excitation of the sample with detection of a is therefore
chosen FID point. The angle spanned after n pulses is
given by 3
MN ¼ W 2 ð16Þ
  4
t 2p and since N ¼ Nr Nh is the total number of spiral points
h¼ ¼ ðn  1ÞDh: ð11Þ
TR Nh we find that the number of points per spiral rotation
222 M. Halse et al. / Journal of Magnetic Resonance 165 (2003) 219–229

which would be equivalent to the number of points such that the overall trajectory inscribes a sphere in the
contained in the circular area is k-space cube. The constants an and bn for each of the Nc
3 W2 nested cones are given by
Nh ¼ : ð17Þ R tan /n
4 MNr an ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ;
This provides an approximate value for the number of 1  tan2 /n
ð19Þ
points per spiral rotation. tan /n
bn ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ;
The Spiral-SPRITE trajectory is similar to SPRITE 1  tan2 /n
phase encoding which consists of a set of equally in-
cremented discrete gradient steps [6]. However, for where the pitch angle for each cone is
Spiral-SPRITE the Gx and Gy gradients are sinusoidally ðn  1Þp
/n ¼ ; n ¼ 1; 2; . . . ; Nc
ramped through a set of discrete gradient steps as il- ðNc  1Þ
lustrated in the pulse sequence presented in Fig. 2. The
and the radius is a constant which describes the largest
trajectory is reminiscent of the pure frequency encoding
sphere which can be inscribed within the k-space cube
square spiral trajectory presented by Meyer et al. [16].
kmax
R¼ :
2.2. The conical-SPRITE trajectory 2
For illustrative purposes, a system of three nested
The spiral trajectory is a natural 2D k-space sampling cones is shown in Fig. 4.
method. The extension to 3D requires an additional Maintaining the notation used above for Spiral-
phase encode gradient in the third dimension. The re- SPRITE, the Cartesian coordinates for the Conical-
sulting Ôstack of spiralsÕ trajectory does not begin sam- SPRITE trajectory are
pling in the centre of the k-space cube for all spirals and h
is therefore not a true 3D centric scanning technique. kx ¼ an cos h;
hmax
We may avoid this difficulty by generating a spiral tra-
jectory which is mapped onto a conical surface to ex- h
ky ¼ an sin h;
plore the third dimension of k-space. (Fig. 3) This is hmax
accomplished by extending the Archimedean spiral with h
kz ¼ bn :
the addition of a h dependent component in the z-di- hmax
mension
k ¼ ahk^r þ bhk^z : ð18Þ
This cone can be described by a pitch angle, /, and a
radius, R. The pitch angle is defined by the inverse
tangent of the ratio of the constants, a and b. The radius
is given by the square root of the sum of the squares of a
and b. A nested system of these cones can be defined

Fig. 2. The Spiral-SPRITE pulse sequence. Each gradient step corre-


sponds to a sample excitation using a broadband RF pulse. The gra- Fig. 3. A single spiral mapped to a conical surface. This conical tra-
dients are sinusoidally ramped through a discrete set of gradient jectory can be classified by a radius, R, defined as the length of the side
amplitudes for each spiral interleave. This diagram is illustrative of a of the cone and a pitch angle, /, defined as the angle between the
single spiral interleave. central kz -axis and the surface of the cone.
M. Halse et al. / Journal of Magnetic Resonance 165 (2003) 219–229 223

Fig. 5. The 3D Conical-SPRITE pulse sequence. Gx and Gy are sinu-


Fig. 4. A system of three conical trajectories nested to form the top half soidally ramped while Gz is linearly ramped through a discrete set of
of a sphere in k-space. Three cones are chosen for ease of illustration. gradient amplitudes for each nested cone. The first point on each cone
In practice 39 cones are implemented for 3D Conical-SPRITE. is acquired in the absence of any gradients. This diagram is illustrative
of the first few dozen points on a single conical trajectory.
Using Eq. (11) we can write the Cartesian coordinates
The maximum number of points in a single conical
after n pulses
  acquisition is limited by gradient duty cycle and longi-
ðn  1Þ 2pðN  1Þ tudinal magnetization evolution considerations.
k x ¼ an cos ;
Nr Nh Nh The pulse sequence diagram for Conical-SPRITE is
 
ðn  1Þ 2pðN  1Þ depicted in Fig. 5. The x and y gradients vary sinusoi-
k y ¼ an sin ; ð20Þ dally, reminiscent of the Spiral-SPRITE sequence, while
Nr Nh Nh
ðn  1Þ the z gradient is a stepped linear ramp.
k z ¼ bn : In both Spiral-SPRITE and Conical-SPRITE, the
Nr Nh
rounding of points to the equispaced square or cubic
As with the spiral trajectory, we project the conical grid results in duplication of some of the Cartesian grid
coordinates onto a cubic equispaced W  W  W points, especially near the centre of k-space where
Cartesian grid where W is typically a power of 2 so the sampling is denser. Duplicate points are not acquired,
FFT may be employed for rapid image reconstruction. with consequent modifications to the gradient wave-
The range of kx , ky , and kz are defined by jkx j 6 kmax , form, except in the case of duplicate points which are
jky j 6 kmax , and jkz j 6 kmax . The magnetic field gradient acquired after the same number of RF pulses on adja-
amplitudes used for the Conical-SPRITE scan technique cent cones or spiral interleaves. In this case the duplicate
are then calculated using Eqs. (19) and (20). points have the same longitudinal magnetization satu-
The parameters Nc , Nr , and Nh are chosen with con- ration weighting and are signal averaged.
sideration given to the previous knowledge that the
trajectory will be projected onto a cubic equispaced grid
and to the desired values for the total number of points 3. Experimental
in the final trajectory, and the maximum number of
points acquired along a single conical trajectory. The Conical-SPRITE sequence was implemented on
The total number of points acquired by this acquisi- a Nalorac (Martinez, CA) 2.4 T 32 cm i.d. horizontal
tion strategy is proportional to the volume of the sphere bore superconducting magnet with a water cooled
inscribed within the k-space cube. The ratio of the vol- 7.5 cm i.d. gradient set (maximum gradient strength
ume within the sphere defined by the nested cones tra- 100 G/cm) driven by Techron (Elkhart, IN) 8710 am-
jectory to the volume of the entire k-space cube is plifiers. The 1 H RF probe was a 1 H free 32 rung
 4 3 quadrature birdcage coil (Morris Instruments, Ottawa).
3
pR 1
3
 : The 19 F RF probe was a homemade 8 rung birdcage
ð2RÞ 2
quadrature coil. The probes were driven by a 2 kW
Therefore the total number of points acquired by the AMT (Brea, CA) 3445 RF amplifier. The console was a
conical SPRITE method has an upper bound of ap- Tecmag (Houston, TX) Apollo. All experiments were
proximately 1=2W 3 . performed at ambient temperature.
224 M. Halse et al. / Journal of Magnetic Resonance 165 (2003) 219–229

The Cartesian components of the generic Spiral- where the observed MR signal is relatively small. The
SPRITE and Conical-SPRITE gradient waveforms, observed image intensity at any point in the image is
with gradient values between the arbitrary limits  100, related to the equilibrium longitudinal magnetization
were calculated and loaded onto the Apollo console. through
The generic waveform provided a template which could  
 1  eTR =T1 
tp =T2 
S ¼ M0 e  ð21Þ
be linearly scaled in amplitude by the console according  1  cos heTR =T1  sin h;
to the specific imaging parameters, tp and FOV, for a
given experiment. The template conical k-space trajec- where the exponential factor containing T2 accounts for
tory, used for all the Conical-SPRITE experiments, was magnetization de-phasing after the application of the
constructed using IDL version 5.5 (Research Systems, RF pulse [6,17]. This particular aspect of SPRITE has
Boulder, CO). The number of nested cones was chosen been investigated experimentally and analytically by
to be Nc ¼ 39, the number of k-space rotations for each Mastikhin et al. [7] and Prado et al. [4]. The approach to
conc was Nr ¼ 40. The total number of points acquired steady state through application of multiple RF pulses
for the 3D 643 images was 61 690 while the maximum has been described by Vlaardingerbroek and den
number of points on a single k-space cone was 2242. Boer [18].
The 19 F sample was a Teflon sleeve with a diameter of The centric scan SPRITE techniques may be em-
38 mm and a length of 89 mm. The relaxation times of ployed to utilize longitudinal magnetization more
the sample were T1 ¼ 150 ms and T2  ¼ 35 ls. The Tef- efficiently, especially for materials which exhibit long
lon sleeve image was acquired with tp ¼ 30 ls, spin–lattice relaxation times (on the order of hundreds
FOV ¼ 8 cm  8 cm  20 cm, TR ¼ 2 ms, flip angle ¼ 8° of milliseconds). For TR  5T1 the successive applica-
in 16 averages. The total image acquisition time for the tion of RF pulses leads to saturation of z-magnetization.
3D 643 image was 40 min. However, in the case of a centric scan SPRITE tech-
The 1 H resolution phantom was a crosslinked cis- nique there is no z-magnetization saturation of the
polybutadiene disk. The dimensions of the disk are central k-space point. Therefore for a centric scan
presented in Fig. 7. The relaxation times for this sample SPRITE experiment the local image intensity at any
were T1 ¼ 218 ms and T2 ¼ 330 ls. Images of the reso- point in the image is related to the equilibrium longi-
lution phantom were acquired with tp ¼ 100 ls, tudinal magnetization through
FOV ¼ 6 cm  6 cm  6 cm, TR ¼ 2 ms, 4 averages and 
S ¼ M0 etp =T2 sin h: ð22Þ
flip angles of 4.6°, 6.5°, 9.1°, and 13.1°. The total
acquisition time for each 3D 643 image was 12 min Fig. 6 shows a 3D rendered Conical-SPRITE image
and 40 s. of a Teflon sleeve with T2 ¼ 35 ls and T1 ¼ 150 ms. This
The second 1 H imaging sample was a 3 cm long water image demonstrates the viability of the use of Conical-
saturated concrete sample with a diameter of 2 cm. The SPRITE for acquiring high quality images, in reason-
longitudinal relaxation time of the water within the able acquisition times, of samples with short T2 and
concrete was approximately 1 ms with a T2 of approxi- longer T1 (hundreds of milliseconds). The S/N for this
mately 150 ls. The water saturated concrete images were image was 10.5. If a similar image was acquired with
acquired with tp ¼ 80 ls, FOV ¼ 6 cm  6 cm  6 cm, flip standard SPRITE the S/N would be reduced by 42% due
angle ¼ 34° in a single scan. The first 3D 643 image was to longitudinal magnetization saturation effects.
acquired with TR ¼ 2 ms in 2 min and 18 s. The second
3D 643 image was acquired with TR ¼ 250 ls in 33 s.
S/N for all images was calculated as a ratio of the
mean image intensity of a region of interest within the
sample area compared to the mean image intensity from
a similar region of the background.
Point spread functions were numerically simulated
using IDL version 5.5 (Research Systems, Boulder, CO).

4. Results/discussion

4.1. Centric scan signal equation


Fig. 6. A 19 F Conical-SPRITE image of a Teflon sleeve acquired with a
The original SPRITE technique is a longitudinal
tp of 30 ls, FOV of 8 cm  8 cm  20 cm, TR of 2 ms, flip angle of 8°
steady-state imaging method [6,7]. A dynamic approach with 16 averages. The total image acquisition time for the 3D 643
to equilibrium is achieved by applying low flip angle RF image was 40 min. The relaxation times of this sample are T1 ¼ 150 ms
pulses while initially sampling at the edge of k-space and T2 ¼ 35 ls. The image S/N was 10.5.
M. Halse et al. / Journal of Magnetic Resonance 165 (2003) 219–229 225

13.1°) leads to an increased S/N, according to the sin h


term in the signal equation.
The centric scan SPRITE techniques are naturally
density weighted. The signal equation, Eq. (22), sug-
gests that the available transverse magnetization at any
given point in the image is proportional to the equi-
librium magnetization, and thus the nuclei density,
reduced only by the transverse magnetization de-
phasing term, controlled by the choice of tp =T2 , and
the sine of the flip angle. The relationship between the
signal intensity and the sine of the flip angle can be
verified from the proton resolution phantom images in
Fig. 8. The signal to noise in the first image, acquired
with a flip angle of 4.6°, was calculated to be 25. The
Fig. 7. A schematic diagram of the resolution phantom employed in S/N calculated from the last image, acquired with a flip
Fig. 8. angle of 13.1°, was 78. The ratio of the sine of the flip
angles of these images is 2.8. The ratio of the S/N of
these images is 3.1. This demonstrates the validity of
The signal equation, Eq. (22), was further explored the relationship between signal and flip angle as pre-
through the acquisition of Conical-SPRITE images of a sented by the signal equation (Eq. (22)). The degra-
1
H resolution phantom at four different RF flip angles. dation of image resolution with increasing flip angle,
2D slices from these 3D images are presented in Fig. 8. which is also apparent in Fig. 8, will be discussed later
Increasing the applied RF flip angle (4.6°, 6.5°, 9.1°, in this manuscript.

Fig. 8. 2D slices from 3D 1 H Conical-SPRITE images of a crosslinked cis-polybutadiene resolution phantom. Images were acquired with a tp of
100 ls, an isotropic FOV of 6 cm  6 cm  6 cm, TR of 2 ms with 4 averages. The total acquisition time for each 3D 643 image was 12 min and 40 s. The
flip angles employed were (a) 4.6°, (b) 6.5°, (c) 9.5°, and (d) 13.1°.
226 M. Halse et al. / Journal of Magnetic Resonance 165 (2003) 219–229

4.2. Rapidity of data acquisition acquired. Despite the uncertainty in true k-space coor-
dinates, which will degrade image resolution, a high
Image acquisition times are significantly decreased quality image is still acquired. The shorter TR of the
with Spiral-SPRITE and Conical-SPRITE compared to second image would reduce the image intensity, with a
conventional SPRITE imaging. This arises from two standard SPRITE acquisition, by approximately 40%
combined effects. First, the corners of k-space, where according to Eq. (21). The observed S/N of the two
signal intensity is typically low and where minimal images is however minimally different. This result up-
contributions to the image features are anticipated, are holds the centric scan signal equation (Eq. (22)) and il-
naturally omitted from the centric scan trajectories. This lustrates the increased sensitivity of the centric scan
reduces the number of acquired points, and conse- SPRITE techniques over the standard SPRITE method.
quently the acquisition time, to approximately 3=4 in the
case of Spiral-SPRITE (Eq. (15)) and 1=2 in the case of 4.3. Gradient duty cycle
Conical-SPRITE compared to the standard SPRITE
case. Second, the number of spiral interleaves or nested Our two new centric scan SPRITE acquisition strat-
cones is typically less than the number of analogous egies have a significantly reduced gradient duty cycle
lines acquired in conventional SPRITE imaging. This compared to analogous standard SPRITE experiments.
saves acquisition time on the delays between lines for The overall duty cycle is much less because less time is
magnetization recovery and gradient cooling. spent acquiring points at the extremities of k-space
In the case of samples with a short spin–lattice re- where gradient values are typically very high. Relative
laxation time (on the order of milliseconds), image ac- duty cycles for Conical-SPRITE and standard SPRITE
quisition times can be significantly reduced because TR experiments can be calculated by summing the gradient
for the experiment can be greatly reduced without sig- values, ranging from )100 to 100, within the template
nificant image resolution loss. The first 643 3D 1 H gradient waveform for each technique over the period of
Conical-SPRITE image of the water saturated concrete interest, normalizing the result by the total number of
sample (Fig. 9a) was acquired with TR ¼ 2 ms in 2 min gradient steps. The gradient duty cycle for a complete
and 18 s and had a S/N of 14. The second image (Fig. 9b) conventional SPRITE experiment is 50.8%. The gradi-
was acquired with a TR of 250 ls in 33 s and had a S/N of ent duty cycle for a complete Conical-SPRITE experi-
12. The reduction in total imaging time is not in pro- ment is 29.7% for Gx and Gy , and 33.8% for Gz . On a
portion to the TR reduction because of other delays in smaller time scale, within a single experiment, gradient
the pulse sequence, which have been maintained con- duty cycles for centric scan SPRITE techniques are
stant. Most notable is the time delay between sequential greatly reduced by the ramping of the gradients in all
cones. three dimensions. In a conventional SPRITE experi-
The TR in the second image, 250 ls, is on the order of ment, the primary phase encode gradient is a linear
the switching time of the gradients. This introduces an ramp while the two secondary phase encode gradients
uncertainty in the gradient values at the time of acqui- are maintained at a constant value throughout the ac-
sition, and consequently the value of the k-space point quisition of a single line in k-space. Standard SPRITE

Fig. 9. (a) A 2D slice from a 3D 1 H Conical-SPRITE image of a water saturated concrete sample acquired with a tp of 80 ls, an isotropic FOV of
6 cm  6 cm  6 cm, flip angle of 34°, and TR of 2 ms in a single scan. The total acquisition time of this 3D 643 image was 2 min and 18 s. The image
S/N was 14. (b) A 2D slice from a 3D 1 H Conical-SPRITE image of the same sample acquired with the same parameters as (a) except with a TR of
250 ls. The total acquisition time was 33 s. The image S/N was 12. The displayed slice in (b) is slightly offset from the slice in (a).
M. Halse et al. / Journal of Magnetic Resonance 165 (2003) 219–229 227

has a duty cycle of 50.8% for Gx , and 100.0% for Gy and


Mz;n ðTR Þ ¼ M0 ð1  wÞ cosn henTR =T1 þ M0 w; ð23Þ
Gz during the most demanding line acquired at the edge
of k-space. In Conical-SPRITE Gx and Gy are ramped where
sinusoidally while Gz is ramped in a stepped linear 1  eTR =T1
fashion. For the period of the most demanding conical w¼ : ð24Þ
1  cos heTR =T1
acquisition, the cone at approximately 45° to the kx ky -
Longitudinal magnetization depends on two terms.
plane, the duty cycle for Gx is 27.2%, Gy is 27.4%, and Gz
The first term represents the evolution of z-magnetiza-
is 46.6%. Thus for both the shorter timescale within a
tion during a SPRITE acquisition and depends on the
single experiment and for a timescale including an entire
number of applied RF pulses. This term influences im-
experiment, centric scan SPRITE techniques have a
age resolution while the second term is the steady-state
much lower gradient duty cycle than standard SPRITE
term and it affects image S/N [7]. The exponential term
techniques.
in Eq. (23) can be further written as
The gradient duty cycles considered above were cal-
culated based on the gradient strength or gradient cur- cosn henðTR =T1 Þ ¼ enðTR =T1app Þ ;
rent; however, the gradient duty cycle can also be 1 1 lnðcos hÞ ð25Þ
considered in terms of the gradient power deposition. ¼  :
T1app T1 TR
The power output is proportional to the square of the
current thus in the case of power output, the gradient The transient term in Eq. (25) is an exponential which
duty cycle reductions from standard SPRITE to Coni- depends on an apparent T1 which incorporates flip angle
cal-SPRITE will be even more significant. Imaging ex- effects and the natural lifetime T1 [7,18].
periments which cause significant gradient overheating Excluding the effects of diffusion during the centric
in standard SPRITE experiments are typically executed scan SPRITE acquisition the MR image resolution is
with minimal gradient heating in Conical-SPRITE ex- affected by a finite sampling function U ðtÞ and a func-
periments. tion RðtÞ which represents the effects of longitudinal
magnetization saturation through repetitive RF pulses.
4.4. Passive spoiling of transverse magnetization RðtÞ is dependent on the number of RF pulses and is
proportional to the transient term presented in Eq. (23).
Residual transverse magnetization due to previous The Fourier transforms of U ðtÞ and RðtÞ yield the
RF pulses may result in the formation of echoes and sampling point spread function (SPSF) and the evolu-
lead to image artifacts. In systems where T2 is less than tion point spread function (EPSF). The sampled image,
or comparable to the gradient step length (TR ) the including effects of finite sampling and relaxation be-
residual magnetization is efficiently suppressed by havior, is given by
‘‘passive spoiling’’ due to the phase encoding gradients Mðx; yÞ ¼ F fMðkx ; ky ÞU ðtÞRðtÞg; ð26Þ
[4,7]. In standard SPRITE active spoiling is usually re-
where F f g denotes the Fourier transform and Mðkx ; ky Þ
quired only when sampling around the center of k-space
is the sample magnetization in the absence of relaxation/
when cDGx Lx TR  2p and TR ¼ T2 [2]. The spoiling
saturation effects. The previous equation can also be
gradient strength in such cases has been discussed by
written as
Hennig [19]. The centric scan SPRITE strategies have an
inherent spoiling advantage due to sampling along spiral Mðx; yÞ ¼ F fMðk;x ky Þg  F fU ðtÞg  F fRðtÞg
like k-space trajectories. Each trajectory begins at the ¼ F fMðkx; ky Þg  SPSF  EPSF ð27Þ
origin of k-space, where Gx ¼ Gy ¼ Gz ¼ 0 G/cm. The
gradients increase in amplitude and are never all equal which explicitly states that the actual image is a con-
to zero as k-space is traversed to the extremity of volution with the points spread functions defined above.
k-space. Artifacts due to unwanted echoes have not yet We can explore the effects of the sampling function
been observed with our centric scan SPRITE methods. and the longitudinal magnetization relaxation function
through simulation of these point spread functions. A
4.5. Longitudinal magnetization evolution and image composite PSF, which is the convolution of the SPSF
blurring and the EPSF, was calculated for a single spiral acqui-
sition with a TR =T1 of 0.5% and a flip angle of 5°.
We now consider the effects of longitudinal magne- Fig. 10a depicts the product of sampling function, U ðtÞ,
tization evolution during a centric scan SPRITE exper- and the magnetization evolution function, RðtÞ, in k-
iment and return to the resolution effects of increasing space. The T1app decay, as described by Eq. (25), is
the RF flip angle observed in Fig. 8. Using the notation ÔwrappedÕ around the spiral trajectory, settling into a
of Mastikhin et al. [7] the available longitudinal mag- steady state, Eq. (24), away from the centre of k-space.
netization after n pulses, when magnetization prepara- The Fourier transform of this function is the composite
tion is not introduced, can be written as PSF and is shown in Fig. 10b.
228 M. Halse et al. / Journal of Magnetic Resonance 165 (2003) 219–229

Fig. 10. (a) A numerical simulation of the product of the sampling function, UðtÞ, and the longitudinal magnetization evolution function, RðtÞ, in k-
space for a single spiral, TR =T1 ¼ 0:5% and flip angle ¼ 5°. The T1app decay of the longitudinal magnetization is ÔwrappedÕ around the spiral sampling
function, settling into a steady state displaced from k-space origin. The x and y coordinates are k-space points. (b) A numerical simulation of the
composite PSF in image space. This is the 2D FFT of the function in (a). The x and y coordinates are x and y pixels. The PSF has a Lorentzian-like
shape with additional low lying structure arising from the points omitted from the central region of k-space during the spiral trajectory. The width of
the Lorentzian peak indicates the extent of convolution blurring in an analogous imaging experiment.

Mastikhin et al. [7] demonstrated that for half The exponential filtering effect combined with the sam-
k-space centric scan SPRITE sampling the exponential pling methods in Spiral-SPRITE, and similarly in
decay of longitudinal magnetization leads to exponential Conical-SPRITE, leads to a more radially symmetric
filtering along the primary phase encode direction of composite PSF with some low lying structure due to the
k-space. The exponential filtering effect introduces a k-space points omitted during the spiral acquisition
convolution blurring to the image in a fashion similar to (Fig. 10b). The extent to which this filtering affects im-
the application of an exponential filter during image age resolution can be controlled by the choice of TR and
processing. The Fourier transform of the evolution the flip angle. The flip angle is a natural choice for the
function yields a point spread function that has a Lo- control variable. Fig. 11a shows the sampling function
rentzian shape and the convolution effect leads to blur- and longitudinal magnetization evolution function in k-
ring of the image along one direction in image space [7]. space for a single spiral with TR =T1 of 0.5% and a flip

Fig. 11. (a) A numerical simulation of the product of the sampling function, UðtÞ, and the longitudinal magnetization evolution function, RðtÞ, in k-
space for a single spiral, TR =T1 ¼ 0:5% and flip angle ¼ 15°. The T1app decay of the longitudinal magnetization is ÔwrappedÕ around the spiral sampling
function, settling into a reduced steady state displaced from the k-space origin. (b) A numerical simulation of the composite PSF in image space. This
is the 2D FFT of the function in (a). The base of the composite PSF is broader than the 5° flip angle case in Fig. 10b. This leads to increased
convolution blurring in the image.
M. Halse et al. / Journal of Magnetic Resonance 165 (2003) 219–229 229

angle of 15°. The corresponding composite PSF in im- sample of Fig. 6. We thank R.P. MacGregor for his
age space is shown in Fig. 11b. The width of the PSF is technical assistance.
noticeably increased which would result in increased
convolution blurring in the corresponding images.
The Conical-SPRITE trajectory is inherently 3D and
References
therefore presenting a sensible geometrical representa-
tion of the 4D composite PSF for this case is problem- [1] B.J. Balcom, SPRITE imaging of short relaxation time nuclei, in:
atic. However the features of the simulations presented P. Bl€
u mler, B. Bl€
umich, R. Botto, E. Fukushima (Eds.), Spatially
for the 2D spiral case are representative of those for the Resolved Magnetic Resonance, Wiley-VCH, Toronto, 1998, pp.
3D conical case. The Conical-SPRITE 1 H resolution 75–86.
phantom images presented in Fig. 8 demonstrate ex- [2] C.B. Kennedy, B.J. Balcom, I.V. Mastikhin, Three-dimensional
magnetic resonance imaging of rigid polymeric materials using
perimentally the convolution blurring effect of an in- single-point ramped imaging with T1 enhancement (SPRITE),
creased flip angle. The convolution blurring of the image Can. J. Chem. 76 (1998) 1753–1765.
in Fig. 8d, with a flip angle of 13.1°, is clearly greater [3] S.D. Beyea, B.J. Balcom, P.J. Prado, A.R. Cross, C.B. Kennedy,
than that of the image in Fig. 8a, with a flip angle of R.L. Armstrong, T.W. Bremner, Relaxation time mapping of
short T2 nuclei with single-point imaging methods, J. Magn.
4.6°. Further investigation of the features of the com-
Reson. 135 (1998) 156–164.
posite PSF and the segregation of the effects of the SPSF [4] P.J. Prado, B.J. Balcom, I.V. Mastikhin, A.R. Cross, R.L.
and the EPSF in order to accurately predict the con- Armstrong, A. Logan, Magnetic resonance imaging of gases: a
volution blurring effects for both the spiral and the single-point ramped imaging with T1 enhancement (SPRITE)
conical trajectory cases is the subject of ongoing study, J. Magn. Reson. 137 (1999) 324–332.
research. [5] S. Gravina, D.G. Cory, Sensitivity and resolution of constant-time
imaging, J. Magn. Reson. A 104 (1994) 53–61.
[6] B.J. Balcom, R.P. MacGregor, S.D. Beyea, D.P. Green, R.L.
Armstrong, T.W. Bremner, Single-point ramped imaging with T1
5. Conclusion enhancement (SPRITE), J. Magn. Reson. A 123 (1996) 131–134.
[7] I.V. Mastikhin, B.J. Balcom, P.J. Prado, C.B. Kennedy, SPRITE
The pure phase encode, centric scan, techniques MRI with prepared magnetization and centric k-space sampling,
J. Magn. Reson. 136 (1999) 159–168.
Spiral-SPRITE, and Conical-SPRITE are well suited to [8] D.E. Axelson, A. Kantzas, T. Eads, Single point 1 H magnetic
the study of a wide variety of short T2 systems. Both resonance imaging of rigid solids, Can. J. Appl. Spectrosc. 40
methods allow one to rationally control image intensity (1995) 16.
and spatial resolution. Both methods feature a simpli- [9] D.J. Goodyear, Development of single point spiral scan imaging:
fication of image contrast, and an increased sensitivity Spiral-SPRITE, M.Sc. Thesis, Univeristy of New Brunswick,
2000.
compared to standard SPRITE methods. The increase in [10] P. Szomolanyi, D. Goodyear, B. Balcom, D. Matheson, SPIRAL-
sensitivity is achieved through an increase in the raw SPRITE: a rapid single point MRI technique for application to
S/N and a simultaneous decrease in acquisition time. porous media, Magn. Reson. Imaging 19 (2001) 423–428.
Gradient heating concerns associated with conventional [11] M. Halse, Conical-SPRITE: a rapid, 3D MRI method for imaging
SPRITE imaging techniques are significantly reduced solid-like materials, B.Sc. Thesis, University of New Brunswick,
2002.
due to the reduced gradient duty cycles associated with [12] P. Irarrazabal, D. Nishimura, Efficient k-space trajectories for fast
the spiral and conical trajectories. Spiral and cones three-dimensional imaging, in: Proceedings of the SMR, Second
trajectories are readily implemented because the sam- Scientific Meeting and Exhibition, San Francisco, 1994, p. 30.
pled k-space data points fall on a Cartesian grid. This [13] J. Jackson, C. Meyer, D. Nishimura, A. Macovski, Selection of a
convolution function for Fourier inversion using gridding, IEEE
removes any need for re-gridding prior to image re-
Trans. Med. Imaging MI-10 (1991) 473–478.
construction. [14] K.F. King, T.K.F. Foo, C.R. Crawford, Optimized gradient
waveforms for spiral scanning, Magn. Reson. Med. 34 (1995) 156–
160.
Acknowledgments [15] C.B. Ahn, J.H. Kim, Z.H. Cho, High-speed spiral-scan echo
planar NMR imaging—I, IEEE Trans. Med. Imaging MI-5 (1)
(1986) 1–7.
B.J.B. thanks NSERC of Canada for operating and [16] C.H. Meyer, A. Macovski, Square spiral fast chemical shift
equipment grants in addition to a Steacie fellowship imaging, Magn. Reson. Imaging 5 (1987).
(2000–2002). B.J.B. also thanks the Canada Chairs [17] P.T. Callaghan, Principles of Nuclear Magnetic Resonance
program for a Research Chair in MRI of Materials Microscopy, Clarendon Press, Oxford, 1991.
(2002–2009). D.J.G. and M.H. thank NSERC of Can- [18] M.T. Vlaardingerbroek, J.R. den Boer, Magnetic Resonance
Imaging, Springer, New York, 1996.
ada for post-graduate scholarships. The authors grate- [19] J. Hennig, Echoes—how to generate, recognize, use or avoid
fully thank DuPont Canada for an Aid to Education them in MR-imaging sequences, Concepts Magn. Reson. 3 (1991)
award, 3M Canada for financial support, and the Teflon 125.

You might also like