PHYSICS OF MEDICAL X-RAY IMAGING (1)Chapter 10

CHAPTER 10. DIGITAL SUBTRACTION ANGIOGRAPHY (DSA)

10.1 Introduction

In this chapter we will discuss a technique that is known as digital subtraction angiography but which is
also referred to as digital radiography, digital fluoroscopy, and photoelectronic imaging. We will use
these terms interchangeably although the term digital subtraction angiography refers specifically to
techniques which subtract two images that are obtained before and after contrast media is administered to
the patient for the purposes of studying blood vessels (angiography). The more general term, digital
radiography encompasses the use of all digital electronic techniques in x-ray imaging. According to some
writers this term also includes the use of x-ray computed tomography (CT), although digital radiography
in this chapter will refer only to those techniques in which digital electronics are used to acquire planar
rather than tomographic images. In addition, we will concentrate on systems that use image intensifiers
viewed by video cameras since these systems provide helpful illustrative examples.

10.2 System Design for a Digital Radiograph System

Figure 10.1. Digital Radiographic System Design A general system diagram for a digital radiographic system is
System Control and Operator
given in Figure 10-1. At the heart of this system is a digital
Interface
image processing system which acquires images from a video
camera and provides timing signals to both the x-ray generator
Computer Interface
and the image acquisition system to control the flow of data
X-Ray
Digital Image
from the x-ray source into the image processor.
Storage
Generator

Digital
Processing
The image acquisition process begins when timing signals,
X-Ray Tube and
Collimator delivered to the x-ray generator under computer control,
Digital-to-Analog
Conversion

initiates the production of x-rays which are transmitted

through the patient and received by the image intensifier. An
Image Logarithmic Look-
Intensifier Up Table
Analog Image

aperture, placed between the image intensifier and the video

Storage

Optics and
Aperture
Analog-to-Digital
Conversion camera, controls the amount of light delivered to the camera.
This manages the signal-to-noise ratio of the acquired image as
Image
Display

Video
Camera
Analog
Processing will be discussed later in this chapter. A video camera receives
the light image from the image intensifier and converts it to a
Timing

electronic video signal which is delivered to the image

Sync Stripper Signals
Circuit

processor in analog form. The image processor digitizes the

Solid Line Indicates Data Flow

Dashed Line Indicates System Control

image, stores it in memory, and makes it available in digital

form for subtraction with another image set acquired at a different time or at a different energy. The
basic components of the imaging system including the x-ray tube and x-ray generator, the image
intensifier, and the video camera are similar to but must be of higher quality than those used in
conventional fluoroscopy to ensure proper synchronization and match between analog and digital
components.

A common algorithm using digital radiographic systems is temporal subtraction (Figure 10-2). In this
technique, dynamic images of the patient are acquired at a rate of 1 exposure per second or more. A
contrast agent is injected into the patient either intravenously or intra-arterially. A second set of
dynamic images is acquired after the contrast agent flows into the area being imaged. The unopacified
PHYSICS OF MEDICAL X-RAY IMAGING (2)Chapter 10

images (no contrast) are subtracted from the opacified images IMAGE BEFORE IMAGE AFTER

with the subtraction process isolating the signal (which is

CONTRAST CONTRAST

present only in the opacification image), removing the static Head

anatomical structures that are common to both the opacified and

unopacified images. The elimination of background structures Carotid
arteries
makes the arteries visible in the subtraction image even when Neck

they are not visible or barely visible before subtraction.

TEMPORAL

10.3 Imaging System Components SUBTRACTION

CONTRAST Carotid
Arteries

10.3.1 Image Intensifier

The subtraction algorithm assumes that the patient's anatomy is

similar or identical in both the mask and contrast images. The
video camera, the x-ray tube, and the other system components
must be stable enough to ensure this equality so the anatomical SUBTRACTION IMAGE
structure can be subtracted. To preserve the contrast available Figure 10.2. Temporal Subtraction.
in the radiographic image, the image intensifier must have a high
contrast ratio and the analog-to-digital conversion should provide sufficient spatial sampling to preserve
the resolution of the image intensifier.

10.3.2 Light Aperture

Small Camera Aperture Large Camera Aperture

A light aperture (Figure 10-3), similar to those Video Video

found on single-lens reflex cameras, is placed Camera Video Camera the Same Video
Signal for a Specific Light Intensity
Camera

Input
behind the output phosphor of the image Camera Camera
intensifier to control the amount of light reaching Aperture Aperture

the video camera for a given exposure rate. A

small aperture requires a greater radiation exposure Image
Intensifier
Image
Intensifier
Figure 10.3.
to deliver an adequate light level to the video Image
camera, decreasing the effect of quantum noise and Intensifier
Aperature
producing a better overall signal-to-noise ratio in Patient
Adjusting.
Patient

the image. Conversely, a large camera aperture is

used when one wishes to minimize patient
exposure in cases where quantum noise does not
limit the diagnostic information in the image. High Patient Exposure Low Patient Exposure
(High Photon Statistical SNR) (Low Photon Statistical SNR)

10.3.3 Video Camera

One of the most critical components in the imaging chain for our example digital radiographic system is
the video camera. The basic function of the video camera is to produce an analog electronic signal that is
proportional to the amount of light received by the target of the camera.
PHYSICS OF MEDICAL X-RAY IMAGING (3)Chapter 10

Figure 10.4, Schematic Of Video Camera A schematic diagram of a video camera is given
Photoanode
(Video Target) in Figure 10-4. The photoactive element in the
+
+ Cathode
camera is the video target that changes in
+
+
electrical conductivity when exposed to light.
+
+ Scanning an electron beam across sequential
Electron beam is scanned
+
+across the video target lines of the video target, with the deposited
Light +
electronic charge passing through the target to
form an electrical current, creates the video
signal. Regions where the target has been
Video Amplifier Video Signal exposed to high light levels produce a high
(Measures Video
Target Current)
Output conductivity, and thus a large current. Regions
of the target that have been exposed to low
levels of light produce a lower conductivity and thus a smaller video camera current. The resulting signal
is a measure of the light level input to the video target. The information is read out serially as the
electron beam is swept over the target to generate an analog video signal. The video signal is time-
varying signal that encodes the two-dimensional light image at the target as a temporal record. Time
points in the video signal correspond to spatial locations within the light (and x-ray) image.

Figure 10.5. Interlaced Video Scanning The video target can be read out in one of two
Odd Video Field (1/60 sec)
262.5 video lines
Even Video Field (1/60 sec)
262.5 video lines different ways. In video cameras used in the
broadcast industry, the electron beam is scanned
across the target in 262 1/2 passes across the area of
the target. The resulting 262 1/2 lines form an image
of the target, which is called a video field (Figure 10-
5). The video field is produced once every 1/60 of a
second. During the next field, the electron beam scans
along lines located between two lines of the previous
field. Therefore, the second field is acquired with
lines interlaced between those of the first field. The
two fields are called even and odd fields with each
field comprising 1/60 of a second, and two fields make
one frame acquired in 1/30 of a second that is
comprised of 525 lines. This scan choice was chosen
Video Frame (1/30 sec) by the broadcast industry to reduce bandwidth during
Formed from Two Interlaced Video Fields and Is
Composed of 525 Video Lines transmission while avoiding flicker in the viewed
video image. However, interlaced scanning is not ideal for digital radiography. The basic problem of the
interlaced mode is that the video fields are read out continuously. However, most video targets have a
certain amount of lag so that even when they are exposed to a constant light level, it takes several video
fields before the output signal is stable. Thus, in digital radiographic systems, after the x-ray beam is
initiated, lag in the video signal produces images obtained in this early phase to be unstable. These early
fields must be discarded, although this clearly is undesirable since it underutilizes the x-ray exposure
delivered to the patient (Figure 10.6). This is partly due to the fact that parts of a field will receive more
exposure than others following the onset of an x-ray exposure.
PHYSICS OF MEDICAL X-RAY IMAGING (4)Chapter 10

Progressive mode scanning resolves the problems with wasted exposure (Figure 10.7). When this mode
is utilized, an image is stored in the target during a short x-ray exposure and completely read out while
the x-ray beam is off. This approach eliminates the wasted x-ray exposure that is needed to bring frame
to a readable level using continuous x-ray exposure and interlaced scanning. It does this by not mixing
the processes of storage and readout from the video camera target.

There are other aspects of the video Figure 10.6. Continuous Exposure Interlaced Scan
camera that are important in a digital
radiographic system. First, the size of Continuous X-
Wasted X-Ray Exposure

the video signal should be directly Ray

Exposures
proportional to the x-ray fluence
delivered to the input phosphor of the Interlaced Video
Fields
image intensifier. Second, the video
camera must have low lag. This means Video Video

that an image acquired at one point in

time by the video camera does not persist
on its target for more than one readout Figure 10.7. Pulsed Exposure with Progressive
period. This is especially important Scan

where rapidly moving objects, such as the Pulsed X-Ray

heart, are being imaged by the digital
radiographic system. A video camera that Progressive
has the characteristics of good contrast
linearity as well as low lag is a plumbicon target camera since it features a camera tube with a lead oxide
(PbO2 ) vidicon target. Another benefit of plumbicon cameras is that they have excellent noise
characteristics in comparison to other types of video cameras, offering electronic signal-to-noise ratios in
the range of 1000:1 and as high as 2000:1.

10.4 Digital Image Processor

The block diagram for a typical digital image processor is shown in Figure 10-8 & 9. The digital
processor has basic functions illustrated in this diagram including (1) acquiring and digitizing the video
images, (2) storing the digital images in memory, (3) performing arithmetic operations (subtraction,
addition and constant multiplication) on the image data, (4) displaying the digital images on video
monitors, and (5) storing the image data on magnetic media or an optical disk. The image processor also
contains a microprocessor or system controller that controls the basic operations of the image processor,
the x-ray generator, and other components, coordinating and controlling the operation of the digital
imaging system.
PHYSICS OF MEDICAL X-RAY IMAGING (5)Chapter 10

Figure 10.8. Digital Image Processing System VCR TV Monitor

TV Monitor

Analog

Register
Image

DAC
Low-Pass
Log
ADC

TV Camera Image
Camera Aperture
Register
Image

Digital Image Data

PatientTu
beIntensif
ierMemor
yMemory
Antialiasi
ng
X-Ray
FilterMag
netic Data Processing
CD/DVD
Tapeor Digital Magnetic Disk
Optical Digital tape
DiskComp
uterImage

Figure 10.9. DSA System Components

Image Intensifier:
CsI input phosphor - Excellent x-ray absorption
Titanium Window - High contrast ratio (low light scattering)
Spatial Resolution - Can be relaxed for digital imaging

Variable Light Aperture:

Small aperture - Greater patient exposure for improving SNR
Large aperture - Lower patient exposure at lower SNR

Video Camera:
Camera type: Vidicon with lead oxide (PbO2) target ("Plumbicon")
Camera response: Output signal proportional to light input signal
Temporal characteristics: low lag
Signal-to-Noise Ratio: 1000:1 to 2000:1
Scanning Modes: Progressive scanning for best dose utilization
Interlaced scanning also used in some cases

10.4.1 Analog-to-Digital Conversion

PHYSICS OF MEDICAL X-RAY IMAGING (6)Chapter 10

We will assume that an analog video image has been acquired by the x-ray system, image intensifier, and
video camera (Figure 10.8). The analog signal is delivered to the image processor that provides some
preprocessing to adjust the amplitude and level of the video signal to satisfy the input specifications of
the analog-to-digital converter. The step size of the analog-to-digital converter must be selected so that it
does not introduce additional noise to the image signal after digitization and generally is chosen to
approximately equal the standard deviation of the electronic noise. Because the signal-to-noise ratio of
most video cameras is approximately 1000:1, the entire range of the video signal must be covered by
approximately 1000 digitization of quantization steps, corresponding to an analog-to-digital converter
with 10 bits (1024 steps) to cover this analog signal range. The ADC controls conversion of the
temporal analog video signals to sequence of digital numbers. The controlling microprocessor uses
software to format this sequence of digital numbers such that they can be used to index image locations
using a row, column scheme. The spacing between columns is determined by the sampling rate of the
ADC which together with video scan rate determine the number of columns for the digitized image. The
number of rows is set by the number of scan lines provided per video frame.

Many images are acquired in a 512 x 512-pixel matrix although some systems use a 1024 x 1024 matrix.
The image matrix controls the sampling rate of the analog to digital converter. For example, if a 512 x
512 image matrix is used to digitize an image that is acquired over 1/30 of a second; the sample period of
the analog-to-digital converter equals 1/30 of a second divided by 5122. This period is approximately
100 nanoseconds, corresponding to a sampling frequency of about 10 megahertz. This sampling
frequency limits the bandwidth of the system to approximately 5 megahertz and the analog
preprocessing must include a low-pass filter to avoid aliasing thereby limiting the spatial frequency of
the incoming video signal to 5 megahertz or less in this example.

10.4.2 Logarithmic Transformation

Following digitization, the image data are logarithmically transformed, meaning that the pixel values are
replaced by their logarithm. The logarithmic transformation is required to remove stationary anatomical
structure during image subtraction. The logarithmic transformation can be performed on the analog signal
prior to digitization with a logarithmic amplifier (i.e. a specialized operational amplifier). However, most
imaging systems currently perform the logarithmic transformation following the analog-to-digital
converter with a digital look-up table that simply replaces a digital value with a new value proportional
to its logarithm.

10.4.3 Image Memory and Integration Feedback Loop

After logarithmic transformation and digitization of the incoming video signal, the image is stored in one
memory of the image processor. Each pixel in the digital image is represented by a digital number having
a minimum of 10 bits corresponding to the digitization range of the image-processor’s analog-to-digital
converter.

Often more than one image is added ("integrated") to reduce noise and improve the SNR of the image.
This averaging is provided by a feedback loop in which the incoming image is added to the contents of
the previously stored image on a pixel-by-pixel basis. If the image processor has a 10 bit analog-digital
converter, the image memories must have at least 4 more bits available corresponding to the range of 16
–32 bits per pixel. This allows multiple images of the same area to be averaged or added to reduce the
PHYSICS OF MEDICAL X-RAY IMAGING (7)Chapter 10

noise in the acquired images. Virtually all image processors have more than one memory plane and many
have 3 image memories. This requirement is obvious in the case of digital subtraction angiography where
the mask image is acquired in one memory and then subtracted from opacification image acquired in a
second memory. Where a sequence of images is acquired, they all can be stored in digital memory in the
image processor. However, this is very expensive and a more likely technique is to subtract the images,
enhance the contrast signal and store the resulting subtraction image either on a high speed magnetic or
optical disk.

10.4.4 Image Subtraction

In digital subtraction angiography, two images are acquired. The first is the "mask" image, which is
obtained before contrast media is injected into the patient. The second is the "opacification" image,
which follows injection of the contrast media and is obtained when the contrast bolus reaches the artery
to be imaged.

The mask and opacification images can be modeled mathematically by assuming that the patient has a
thickness xt and a linear attenuation coefficient of µt. Before contrast media is injected into the patient,
the photon fluence delivered to the image intensifier is

Im = I0 e − µt x t (10-1)

The contrast media then is injected to opacify the artery. If the artery has thickness xI (where xI << xt)
and has a linear attenuation of µI, the image intensifier receives a photon fluence of

−( µ t x t + µ I x I )
I I = I0 e (10-2)

If α is the conversion factor which relates the amplitude of the video signal to the photon fluence
received by the image intensifier, the mask and opacification image signals produced by the video camera
are

Im = αI 0 e −µ t x t (10-3)

− (µ t xt + µ I x I )
I I = αI0 e (10-4)

We now will use equations 10-3 and 10-4 to demonstrate the difference between subtraction of the
images without logarithmic transformation (linear subtraction) and subtraction of the images following
logarithmic transformation (logarithmic subtraction).

10.4.5 Linear subtraction

Some of the early investigators of digital subtraction techniques used a linear subtraction algorithm to
isolate the opacification signal. In linear subtraction, the opacification image is subtracted from the mask
PHYSICS OF MEDICAL X-RAY IMAGING (8)Chapter 10

image without logarithmic transformation. If Slin is the subtraction image, then linear subtraction
produces an image having the form

S lin = I m − I I = αI 0 e − (µt xt ) − αI 0 e − (µt xt + µ I xI ) = αI 0 e − (µt xt ) [1 − e − (µ I xI ) ] (10-5)

If we assume a small iodination signal such that µIxI<< 1, then

− ( µt xt )
Slin = α ( µI x I )I0 e (10-6)

This shows that, using linear subtraction, the thickness of the iodine (xI ) is modulated by the patient
thickness xt . Linear subtraction produces images that retain the unwanted patient anatomy still
superimposed on the desired opacified arterial image.

10.4.6 Logarithmic subtraction

In comparison to linear subtraction, logarithmic subtraction does not retain stationary anatomical
structure that may obscure the small signal contributed by the opacified artery. The mask and
opacification image data are subtracted after they are digitized and logarithmically transformed.
Mathematically, the logarithmic subtraction image Slog is

Slog = ln( I m ) − ln( I I ) = [− µt xt ] − [− µt xt − µ I xI ] = µ I xI (10-7)

so that the resulting logarithmic subtraction signal1 is related to the iodine signal and is not affected by
the patient thickness or the anatomy on which the opacified artery is superimposed. However, the noise
level in subtraction images is always higher.

10.4.7 Image Display and Archival Memory

After processing, the digital images are delivered to a digital-to-analog converter that generates an analog
video signal fed to a video monitor. The image data then can be displayed for examination by the
radiologist.

The analog video signal also can be stored using a video cassette recorder or analog video disk recorder for
archival purposes. The dynamic range of older technology (i.e. analog tape or disk recorders) was
approximately 200:1, in comparison to the 1000:1 dynamic range of the video camera, which we
previously discussed. However, newer digital video recordings preserve the dynamic range of the digital
radiographic images.

1 In x-ray CT, image density is measured in terms of "Hounsfield units", abbreviated HU, and named after one of the fathers
of CT, Godfrey Hounsfield. Similarly, your author proposes that DSA images be measured in terms of "Mistretta units",
abbreviated MU, named after one of the fathers of DSA, Chuck Mistretta. Further, we propose that the Mistretta unit, MU,
be pronounced "moo" rather than "mew". This will avoid confusion with the Greek letter µ having the same spelling. This
pronunciation also commemorates the fact that much of the early work in DSA was conducted at the University of
Wisconsin, in the midst of America's Great Dairyland.
PHYSICS OF MEDICAL X-RAY IMAGING (9)Chapter 10

For absolute fidelity of the image data however, digital image storage is the preferred method. These
devices (including magnetic disk, magnetic tape, and optical disk), record the image data in digital form
which virtually avoids the possibility noise being added by the recording media. Digital image storage is
necessary when images are recorded before subtraction, and is useful when extensive processing (image
integration or iodine quantification) will be performed. The principal disadvantages of digital storage are
that it is considerably more expensive than analog storage, and that one requires a digital image processor
for displaying the image.

10.5 Noise in Digital Subtraction Angiography

In Chapter 8 we discussed how quantum statistical noise σq, electronic noise σe , and digital quantization
noise σΔ all contribute to system noise, and if we design our digital subtraction angiography system
correctly, quantization noise is negligible.

By way of summary, if a radiographic signal is composed of N photons, then the uncertainty (i.e.
standard deviation) in that signal is N , since photon generation and attenuation behave according to
Poisson statistics. Assuming that the camera produces a maximum video output of Vmax at Nmax
photons, the video signal VN corresponding to N photons is given by the proportionality relationship

VN N V
= or VN = max N (10-8)
Vmax N max N max

Where both Nmax and Vmax are constants for any one system configuration. Therefore, the uncertainty in
the video signal due to quantum statistical sources is
€
V  V 
σ q =  max σ N =  max  N (10-9)
 N max   N max 

The electronic noise contributed by the video camera typically is characterized in terms of the camera's
dynamic range, defined to be the ratio of the peak video signal Vmax divided by the standard deviation
(σe ) of the video signal. If D€is the dynamic range of a video camera, then the standard deviation σe of
the electronic noise from the camera is given by

Vmax
σe = (10-10)
D

Finally, the quantization error or quantization noise is the error introduced into that analog signal when it
is digitized. If Δ is the width of the quantization step (i.e. the interval associated with the least
€
significant bit of the analog-to-digital converter) where all analog values from µI-Δ/2 to µI+Δ/2 are
equally likely and are converted into the value µI by the analog-to-digital converter, then the variance of
the quantization error is
PHYSICS OF MEDICAL X-RAY IMAGING (10) Chapter 10

Δ2
2
σ =
Δ (10-11)
12

10.5.1 System Noise in Digital Subtraction Angiography

€
The system noise variance in digital subtraction angiography is obtained by adding the noise variance
from each system component, assuming that these noise contributions are independent. In this
calculation, we assume that the image system consists of a video camera viewing the output phosphor of
the image intensifier, and that the quantization error contributed by the analog-to-digital converter is
negligible.
The camera output contains a component Vq
VIDEO CHARACTERISTIC CURVE
proportional to the exposure to the input
Video Saturation Level phosphor as well as a time-varying term Ve
V
max arising from the "dark current" of the system.

V = Vq+ Ve (10-12)

The uncertainty in the video output can be

calculated using propagation of errors
N
max
Number of Photons σ v2 = σ q2 + σ e2 (10-13)

showing that the contributions from the video

VIDEO DYNAMIC RANGE
dark current σe and quantum statistical
€
σ sources σq add in quadrature to give the total
noise σv in the video system.

V V Substituting the expressions derived above,

max
Dynamic Range =
max
σ we obtain the noise in the video signal in
terms of the number of photons (N), i.e.
number of photons per pixel, the dynamic
Time range (D) of the video camera, the maximum
Figure 10-11. In an ideal video camera, the video signal video level (Vmax), and the maximum number
output is proportional to the level of light exposure up to a
maximum saturation level (Vmax). The video dynamic range is of photons (Nmax) corresponding to the
defined as the maximum video signal divided by the root maximum video level (Figure 10.11)
mean square of the noise in the video signal.

2
V  V 2
2
σ =  max  N +  max 
v (10-14)
 N max   D 

Using Eq. 10-14 the signal to noise ratio (SNR) at 100% contrast for a digital image is
€
PHYSICS OF MEDICAL X-RAY IMAGING (11) Chapter 10

V
V V Vmax
SNR = = = (10-15)
σ 2
Vmax 2
Vmax N 1
2
N + 2
+ 2
N max D2 N max D

Since the analog signal is proportional to the number of photons (N) that we associate with a pixel area
in the digital image
€
V N
= (10-16)
Vmax N max

we can express the signal to noise ratio in terms of the number of photons (per pixel) N at the input
phosphor of the image intensifier (if 100% are absorbed) as
€
N
N max N
SNR = = (10-17)
N 1 N 2

2
+ 2 N + max
N max D D2

We now will investigate this relationship in several important special cases.

€
Effect of Electronic Noise
Assume that opacified artery Case 1: High Dynamic Range Systems
provides 50% attenuation
Image
Intensifier
Assume electronic noise (Ideal system)
σ = 1 mV
1V
If the system dynamic
Opacified
Artery
range D is very large, then
N = N max the system signal-to-noise
Incident X-Ray Beam ratio (equation 10-17)
reduces to
Vmax = 1 volt 100 mV
Dynamic Range = 1000 = 1
mV
σ=1 mV
SNR = N (10-18)
N = 1 N max
10

in which case electronic

10 mV
noise is negligible and the
DQE of the system is
Radiograph of unity. Thus, when the
Opacified Artery N = 1 N max
100 dynamic range is very large,
the noise is contributed
Figure 10-12. The best signal-to-noise ratio in a subtraction image entirely by photon
is obtained when the image data are obtained at the highest possible
video signal level. Because video noise is approximately constant statistics and is Poisson
in amplitude, the signal-to-noise ratio diminishes as the video distributed.
signal decreases.
PHYSICS OF MEDICAL X-RAY IMAGING (12) Chapter 10

Case 2: High signal levels (N = Nmax)

If we are operating at a high signal level where the number of photons used to generate the
image is near the maximum value for the system, then

N max 1
SNR = = (10-19)
2
N 1 1
N max + max
2
+ 2
D N max D

For example assume an exposure of X = 16 mR, a 0.5 mm x 0.5 mm pixel area, E= 60 keV,
and a dynamic range of D =1000. At 60 keV, the mass energy absorption coefficient of
air is €

 µen  cm2
  =0.0289 (10-20)
 ρ  air gm

so that the photon fluence at 60 keV corresponding to an exposure of X = 16 mR (from

Eq 9-21) is
€
photons
Φ =5.03x10 8 (10-21)
cm 2

Over a 0.5 mm x 0.5 mm pixel area, the number of photons is

Nmax = (5.03€x 108 photons/cm2) (0.05cm)2 = 1.26 X 106 photons (10-22)

Therefore the system signal-to-noise ratio (i.e. the ratio of the maximum signal Vmax
divided by the standard deviation of the system) is

1 1
SNR = = = 747 (10-23)
1 1 1 1
+ 2 +
N max D 1.26x106 1x106

At this point we ask whether this level of signal-to-noise is adequate to see a small low-
contrast object in a noisy image. This question can be answered approximately by using
€
the Rose model which relates SNR, contrast, size and fluence. The Rose model states that

k 2 = C 2ΦA = C 2 N (10-24)

where k = 5 (a constant specifying the SNR at which the low contrast signal is visible)
and
€
N = Φ A = number of photons used to image object of area A
PHYSICS OF MEDICAL X-RAY IMAGING (13) Chapter 10

C = contrast level of signal = ΔN/N

N
ϕ = photon fluence in background region =
A
A = area of object

ΔN = difference in number of photons used to image the object and a background

region of equal area.

Therefore, from the Rose model

 ΔN  2 2 2
k 2 N = C2 N 2 =   N = (ΔN ) (10-25)
 N 

Let N = σ2 be the variance of the noise. Therefore, an alternative way to express the Rose
model is to state that

k σ = ΔN (10-26)

Intuitively, this states that the difference in the background and object’s signals must be k
times the standard deviation of the noise. From our derived value of the signal-to-noise
ratio at 100% contrast,

SNR =N/σ (10-27)

and rearranging to follow the format of (10-26) we have

SNR σ = N (10-28)

Now dividing equation 10-26 by equation 10-28 and substituting values for k from the
Rose Model equation and SNR from equation 10-23 yields

C = ΔN/N = k/SNR = 5/747 = 0.67 % (10-29)

Thus the operating level where the image intensifier is irradiated at its maximum exposure
level, the contrast of a minimally perceptible target is

C = ΔN/N ≈ 1% (10-30)

Note that this low contrast detectability limit is determined as the ratio of SNR (or k) from the
rose model to the SNR produced by the imaging system.

Case 3:

If N = 1/10 Nmax, then

PHYSICS OF MEDICAL X-RAY IMAGING (14) Chapter 10

N max
N 10
SNR = = (10-31)
N2 2
N max N max
N + max + 2
D2 10 D

With Nmax= 1.26 x106 photons and D = 1000 as before, we have SNR = 96.3. From
€ SNR = 96.3, an observer would be able to see a minimal contrast
equation 10-29, when
level of

C = ΔN/N = k/SNR = 5/96.3 = 5.2 % (10-32)

Case 4:

If N = 1/100 Nmax, then

N max
N 100
SNR = = (10-33)
N2 2
N max N max
N + max + 2
D2 100 D

With Nmax= 1.26 x106 photons and D =1000, we have SNR = 9.96 in which case only
contrast levels greater than
€
C = ΔN/N = k/SNR = 5/9.96 = 50.2 % (10-34)

are visible in the image. At this point, the SNR of the image is severely limited by the
electronic noise, not by quantum statistics.
PHYSICS OF MEDICAL X-RAY IMAGING (15) Chapter 10

10.6 Methods to Improve the Noise Characteristics of Digitally Subtracted Radiographs

10.6.1 Removal of Bright Spots

Figure 10-13. Bright Spots in Digital Subtraction Angiography
From the above examples, it is obvious that
digital radiographs obtained with image Transmissive regions of the patient create "bright spots" which force other regions to be imaged at lower
intensifier systems will be severely limited signal levels where video noise degrades the image

by electronic noise if the image is not "Bright Spot"

obtained with the maximum number of

photons Nmax. What is not obvious is that
we often are forced into this undesirable
situation since the image contains regions
with both high and low x-ray transmission. Opacified
High transmission is seen at the edge of the Artery

patient or in body regions containing air Position

(lungs or bowel gas). When these regions of
high x-ray transmission are imaged at
maximum video levels, patient regions with By eliminating the bright spot with proper bolusing, all regions of the image can be imaged at maximal
signal levels to achieve the maximum video SNR

lower transmission are imaged at lower

video levels where the data are
compromised by electronic noise. A
common technique to reduce this problem
is to place bags of saline over the high Opacified
transmission areas “bright areas” or to place Artery
Saline Bag
pieces of aluminum in the x-ray beam to Position
decrease the exposure to regions where the
patient is transmissive (Figure 10-13).
Ideally, the input exposure field can be tailored to deliver a nonuniform exposure field to the patient so
that the exposure field reaching the image intensifier is uniform. In this case, all regions of the patient are
imaged closer to maximum exposure levels, to obtain the highest possible signal-to-noise levels in all
areas of the image.

10.6.2 Role of the Video Camera Aperture

Another method to improve the signal-to-noise ratio in digital subtraction angiography is to increase the
exposure delivered to the patient, decreasing the noise contribution from quantum statistical sources.
However, because a specific light level delivered to the camera target will produce a maximum video
response, the x-ray exposure cannot be increased indefinitely without making other adjustments in the
system to insure that this maximum light level is not exceeded. The video camera aperture has a
fundamental role in this respect to control the level of quantum noise in the digitally subtracted
angiogram. Because the aperture is located between the output phosphor of the image intensifier and the
input optics of the video camera, decreasing the aperture diameter also decreases the amount of light
reaching the camera target and lowers the camera response for a given x-ray exposure level.
Correspondingly, the x-ray exposure level must be increased while the aperture diameter is decreased to
maintain a constant video signal level. When the camera aperture is decreased, more x-ray photons are
PHYSICS OF MEDICAL X-RAY IMAGING (16) Chapter 10

used to acquire the image at the quantum sink (i.e. the input phosphor of the image intensifier).
Therefore, the overall signal-to-noise ratio of the video signal increases (assuming that the x-ray exposure
is adjusted to maintain a maximal video signal in the patient image). This reduces the degree of quantum
statistical noise and improves the overall noise characteristics of the image.

It is important to stress the complementary roles of the x-ray exposure level and the video camera
aperture. If the x-ray exposure level is increased without adjusting the aperture, then the increased light
output of the image intensifier can drive the video camera into saturation, producing a useless signal that
saturates at its maximum level. Similarly, decreasing the camera aperture, while holding x-ray exposure
fixed will reduce the light delivered to the video camera, resulting in a smaller video signal. The quantum
noise component scales with the video signal, but the SNR of the video signal will be reduced due to the
fixed level of electronic noise in the video system. Thus, the camera aperture must be adjusted to
provide a video signal near the maximum level to avoid electronic noise, and this signal should correspond
to the region of interest in the body.

10.6.3 Image Integration

A final way to improve system SNR is to use various processing schemes that add (or "integrate")
images together either before or after subtraction to average out the noise contained in the digital
radiographic images. The simplest way this can be performed is by means of frame integration where
two or more frames of the image are added together, in an attempt to reduce both quantum statistical and
electronic noise contributions in the final digital radiograph. If M frames are added together, where all
the frames are nearly identical except for their random noise content, and if σ represents the noise in each
image, then the noise in the integrated image increases by M1/2σ while the signal increases by M.
Therefore, the SNR improves by M1/2. Frame integration has the advantage of reducing both the effects
of photon statistical as well as electronic noise sources. By comparison increasing radiation exposure
per frame only improves on quantum noise effects. However, frame integration has the serious
disadvantage that it is more prone to motion artifacts since a longer period of time is used to acquire a
single integrated image.

10.7 Spatial Resolution in Digital Subtraction Angiography

There are several factors to consider concerning spatial resolution in digital subtraction angiography. The
first is the digital matrix format (512x512 or 1024x1024) used to acquire the image data. The second is
the spatial resolutionof the image intensifier. The third is the degree of geometric unsharpness due to
focal spot size. As we discussed in previous lectures, there is a fundamental trade-off between the
increase in object detail that can be seen due to image magnification and the loss in object detail due to
increased geometric unsharpness. The rendition of object detail increases with greater magnification due
to the fixed resolution in the image intensifier and digital image matrix. On the other hand, increasing
geometric unsharpness (magnification of focal spot) degrades spatial resolution in the object with greater
object magnification.

If an image is recorded with an object magnification of M and if the focal spot has width “a”, the width
of the focal spot when projected onto the detector is (M-1)a. This width is equal to that of a structure
PHYSICS OF MEDICAL X-RAY IMAGING (17) Chapter 10

(M − 1)a
in the object plane having a width of . We therefore can characterize the cut-off frequency us
M
(the frequency suppressed by unsharpness) of focal spot blurring (i.e. geometric unsharpness) by the
inverse of the resolution width

M
us = (10-35)
(M − 1)a

d
Similarly, the resolution width of d for the image intensifier is in the object plane, giving a cut-off
M
frequency from the image intensifier of
M
uII = (10-36)
d

Finally, if the detector width is D and the image is digitized into an N x N matrix, the Nyquist frequency
imposes a limit on the spatial frequencies that can be faithfully reproduced in the image. For an object
magnification of M, the limiting spatial frequency udig imposed by the digtization process is:

N /2
udig = (10-37)
D/M

The trade-off between the decrease in detector unsharpness and the increase in geometric unsharpness
with increasing object magnification is best seen by graphing the cut-off frequencies for focal-spot
blurring and for detector response as a function of object magnification. As shown in Figure 10.14, the
curve labeled "source" shows how increasing magnification produces the resolution loss (i.e. decreasing
the cut-off spatial frequency) due to geometric unsharpness.

A focal spot size of approximately a = 1 mm is common in digital subtraction angiography,

corresponding to a cut-off frequency of

M
us = mm−1 (10-38)
(M −1)

We will consider two different components of detector resolution, from the image intensifier and from
the digital image matrix. The image intensifier (II) has a spatial resolution of about d = 0.2 mm giving a
cut-off frequency of €
uII = 5M mm-1 (10-39)

If we assume that a 512 x 512 image matrix is used to digitize an inscribed circular field of view with a
diameter of 23 cm (approximately 9 inches), then the pixel width is

N / 2 256 cycles
= FOV
= 1.11 cycles
mm (10-40)
D / M 230 FOV
mm
PHYSICS OF MEDICAL X-RAY IMAGING (18) Chapter 10

GEOMETRIC UNSHARPNESS
Source
This yields a cut-off spatial frequency of Point Width
Source =a

udig= 1.11 M mm-1 (10-41) Object

Plane
Image Intensifier Projected source
Resolution = d width = a(M-1)
In this example the digital image matrix
rather than the image intensifier limits the
spatial resolution of the detector a low CUTOFF FREQUENCIES IN OBJECT PLANE
magnification. However, at higher
u II = M u = M u = MN
magnification, geometric unsharpness d ( source (M-1)a ( dig 2D
becomes more problematic. The M
-
M
SYSTEM RESOLUTION vs. OBJECT-
intersection between the cut-off spatial 1
MAGNIFICATION 1
) )
frequency curve for the digital image a Image
a Intensifier
Source
matrix and that for focal spot blurring
Slope = 1/d
indicates the magnification at which the Optimal
Magnification
resolution of the imaging system is Slope = N/2D digitization
optimum.
Asymptotically
Approaches 1
a

Object Magnification

Figure 10.14. The spatial resolution of a DSA image can be limited

by geometric unsharpness from the x-ray focal spot, by image
intensifier resolution, or by aliasing from the digital image matrix.
The limitations of image intensifier resolution and aliasing decrease,
while that from geometric unsharpness increases with magnification.
PHYSICS OF MEDICAL X-RAY IMAGING (19) Chapter 10

CHAPTER 10: HOMEWORK PROBLEMS

1. You are visiting a hospital that has just acquired a digital subtraction angiography (DSA) system.
You watch as the radiologist sets the kVp and mA to obtain an exposure level that gives a
maximum video signal output. The radiologist then looks at the settings on the x-ray console and
decides to decrease the patient exposure and tolerate the accompanying increase in noise by
decreasing the x-ray tube current (mA) to 25% of its original value. You immediately remember
that it is important to adjust the video camera aperture for this new exposure level and alert the
radiologist of the problem. The radiologist thanks you for the information, readjusts the aperture
to achieve a maximum video signal output with the lower exposure level, and then continues the
examination.

For the following calculations assume that the dynamic range of the video camera is 1000:1.
Before the radiologist decreased the exposure or increased the camera aperture, a maximum video
signal was obtained at a setting of 70 kVp and 200 mA. At this setting for a 40-msec exposure,
the x-ray tube produces a photon fluence of 1.031 x 106 photons over a resolution area of 1 mm2.

a. When the initial exposure settings were established, the diameter of the aperture was 4
mm, corresponding to an f-stop of f/5.6. What is the correct diameter and f-stop of the
camera aperture after the patient exposure has been decreased to 25% of its original value?

b. How is the signal-to-noise ratio of the image affected if the patient exposure is reduced by
a factor of 4 but the camera aperture is not adjusted for the new exposure level?

c. What is the signal-to-noise ratio of the image if the patient exposure is reduced by a factor
of 4 and the camera aperture is adjusted to maintain a maximal video signal output?

d. Explain to the radiologist why the video camera aperture must be adjusted for the new
exposure level. Use your calculations from parts b and c of this problem to guide your
thought processes, but give the radiologist conceptual and intuitive (rather than
quantitative) explanations.

2. You take a research position in an x-ray imaging laboratory at a major American university. Your
first development is a new video camera especially developed for pediatric coronary angiography.
Because the infant heart beats at a higher rate than that of the adult, the camera has a video frame
rate of 100 frames per second. Second, because you want to limit both the radiation exposure
and the amount of contrast media required for the angiographic study, the camera is designed with
a dynamic range of 4000 to 1. Finally, because the infant is small, the camera is designed with a
video signal bandwidth high enough to be compatible with a 2048 x 2048 digital image matrix.
You then begin to develop a digital image processor that is compatible with this new video
camera.

All of a sudden, your grandmother drops by with a batch of chocolate chip cookies. Grandma
has been very lonely lately since Grandpa bought a new Macintosh computer and discovered
PHYSICS OF MEDICAL X-RAY IMAGING (20) Chapter 10

"Internet browsing" so she starts asking lots and lots of questions. While munching on her
delicious cookies, you try to explain to Grandma what you are doing.

a. Grandma first asks you what a “video signal bandwidth” is, and why a higher “video
signal bandwidth” is needed if you want to obtain a 2048 x 2048 (rather than a 512 x 512)
image. What problems will you encounter if you have too low of a bandwidth? What
problems will you encounter if your camera has too high of a bandwidth? (By the way, if
you use the word “aliasing” in her presence, Grandma will make you wash your mouth
out with soap.)

b. Grandma then asks you what a "video dynamic range" is and why a higher "video
dynamic range" might let you decrease the radiation exposure and amount of contrast
media needed for the study.

As you are explaining these things to Grandma, a doctor who is walking through the
corridor outside the door of your laboratory distracts her. Feeling rather frisky because of
the lack of attention by Grandpa, Grandma decides to chase after this new target rather
than listen to you talk about medical imaging. As she runs out the door, she almost
knocks over Bruce Hasegawa who overheard your conversation, lost interest in an nurse
he was talking to in the hall, and has entered the room to join the conversation. Noting
the expression of intense curiosity on his face, you moan, wishing Grandma were asking
questions rather than Bruce. By now, Grandma has left you alone with Bruce who, to
your horror, is busily gobbling up the cookies and who also is starting to ask some
questions about your new invention.

c. How many bits are needed by the analog-to-digital converter to minimize quantization
errors in the digitized image data?

d. If you eventually want a 2048x2048 image at a rate of 100 video frames per second, what
is the digitization rate of the analog-to-digital converter for a digital image processor
compatible with this video camera? What is the maximum video bandwidth the camera
should deliver to be consistent with the sampling rate of the analog-to-digital converter?

e. The image intensifier has two different operating modes, one with a 6-inch diameter field-
of-view and other with a 9-inch diameter field-of-view. What spatial frequency (in
cycles/mm), as measured in the detector plane, will the 2048 x 2048 image matrix support
for each of the two operating modes?

f. Assume that your image intensifier has a point-spread function width of 200 microns and
that you operate the image intensifier with a 6-inch diameter field of view with an x-ray
tube having a 300 micron focal spot size. Determine the optimal magnification for
imaging the infant heart and the corresponding optimal cut-off spatial frequency for this
system? Is this magnification compatible for imaging a 2-inch diameter infant heart and
sufficient to see 400 micron diameter coronary arteries?
PHYSICS OF MEDICAL X-RAY IMAGING (21) Chapter 10

g. The infant thickness is equivalent to 15 cm of water and the effective energy of the x-ray
beam is 30 keV. What should the entrance exposure to the infant be if you want to see a
1% contrast level in a 400 micron diameter infant coronary artery with your video
camera?

h. Bruce comments that your calculations for parts c and d of this question show that you
need an analog-to-digital converter which is technically impossible to design and build at
this time. However, he tells you that the important issue in coronary imaging is one of
limiting motion unsharpness rather than one of frame rate. He would rather have a few
images each with minimal amounts of motion unsharpness rather than lots of images
acquired over the entire cardiac cycle. In other words, he prefers an imaging system with
a short acquisition time per frame, but does not need one with a video frame rate faster
than about 15 frames per second.

With this in mind, you start thinking about using a video camera with a progressive read-
out to be used in combination with a short x-ray exposure time. If you want to maintain
the optimal spatial resolution as defined by your calculation in part f of this question,
how short of an exposure time do you need if the infant's myocardium has a maximum
velocity of 5 cm/sec at the point of maximum cardiac contraction.

3. Let's think a little about digital look-up tables such as those utilized in digital subtraction
angiography for the logarithmic transformation. (We can offer you a hint that a digital look-up
table is built out of random access memory).

a. How might you design a digital look-up table in hardware? How would you load the
digital look-up tables with the transformed values? What circuit design would allow you
to deliver digital values and obtain the transformed values from the digital look-up table?
How much random access memory (RAM) is needed for a look-up table with a 10-bit
input and a 10-bit output?

b. If the look-up table has a 10-bit input, the largest value which will be delivered to the
look-up table will be 1023 (i.e. 210-1). However, the natural logarithm of 1023 is 6.93.
Describe how we can represent the logarithmically transformed values if we want a 10-bit
output from the look-up table. You should be able to give a specific mathematical
algorithm to make this possible. What effect will this have on the digitally subtracted
angiograms?

4. Compare the contrast resolution (at peak video signal) associated with a TV fluoroscopy system
with a dynamic range of 1000:1 for the following cases. Assume that the dynamic range refers to
a characteristic resolution element of (0.5 mm2).

a. Fluoroscopy at an exposure of 1 mR per image.

b. Digital radiography at an exposure of 1 R per image (neglect digitization noise).