Medical Imaging Systems
Medical Imaging Systems
Medical Imaging Systems
xiii
arSystems 8
ition of Linearity 8
uperposition Integral. Delta Function. and Impulse Response 9
Invariance 11
ification 11
imensional Fourier Transforms 13
istence Conditions 13
tory 1
dical Imaging Modalities 2
smission of Electromagnetic Energy 3
ing with Acoustic Energy 4
parison of X-Rays and Ultrasound 5
lear Medicine 6
viii Contents
Fourier Transform Relationships 15
Linearity 15
Magnification 15
Shift 15
Convolution 15
Cross Correlation 16
Separability 16
Frequently Occurring Functions and Their Transforms 17
Sampling 18
Elementary Probability 20
Problems 21
3 Physics of Projection Radiography 23
Contents
6 Noise Considerations in Radio h grap y andFluoroscopy 75
Resolution Considerations of the SNR 80
Recorder Statistics 80
Fluoroscopy 83
Image Intensifier 84
Additive Noise 87
SNR of the Line Integral 88
Scatter 90
Scatter Analysis 91
Scatter Reduction Through Separation 94
Scatter Reduction Using Grid Structures 96
Linear Detector Arrays 101
Problems 102
ix
7 Tomography 106
145
Detectors 146
Considerations with Gamma Ray C
149 - ameras 147
Imaging Structures 153
Collimator 157
163
Systems 164
Source Imaging 166
168
Motion Tomography 107
Circular Motion 110
Multiple-Radiography Tomography' T .
C d d S . omosynthesls 111
o e ource Tomography 112
Computerized Tomography 113
Reconstruction Mathematics It .
D
. - eratlve Approaches 114
Irect Reconstruction Methods F .
E
- oUrier Transform Ap h
xamples Using Fourier Transform A proac 117
A' pproach 122
Iternate Direct Reconstruction-Back P . .
Filtered Back-Projection R . rOJectlon 123
. econstructlon System 125
Convolution-Back Projection 127
Reconstruction of the Attenuation Coefficient 130
Modalities 131
in Computerized Tomography 138
142
Screen-Film Systems 64
Critical-Angle Considerations 69
Energy Spectrum Considerations 70
Alternative Approaches to Recorder Systems 70
Overall System Response 72
Problems 73
5 Recorder Resolution Considerations 63
Parallel Geometry 23
Attenuation Relationship 24
Source Spectrum 25
The Attenuation Coefficient 27
Analytic Expressions for the Attenuation Coefficient 33
Problems 34
Point-Source Geometry 36
Depth-Dependent Magnification 39
Examples of Point-Source Geometry 40
Extended Sources 44
Analysis of Imaging Using Planar Sources 46
Alternate Analysis Using Planar Objects 48
Effects of Source Size 49
Simplifying Relationships Using Solid Objects 49
Nonparallel Source Distributions 53
Effects of Object Motion 56
Representative Source Configurations 58
Problems 59
4 Source Considerations in Radiographic Imaging 36
245
Index 251
:xi Contents
Topics in Medicallmaging 225
agnetic Resonance 225
btraction Radiography 233
Imaging 235
Acquisition and Processing Systems in Th D' .
240 ree- Imenslonal Radiography
Ultrasonic Reconstruction Systems 240
of Radiographic Parameters 242
StimUlatEld positron Emission 243
Basic Reflection Imaging 174
Attenuation Correction 176
The A Scan 176
The M Mode 178
Cross-Sectional Imaging or B Mode 178
Diffraction Formulation 181
Steady-State Approximations to the Diffraction Formulation 183
Fresnel Approximation 185
Fraunhofer Approximation 188
Acoustic Focusing 190
Wideband Diffraction 191
Ultrasonic Characteristics of Tissue 195
Attenuation 195
Velocity 196
Reflectivity 197
Compound Scan for Specular Interfaces 199
Noise Considerations 200
Speckle Noise 201
Problems 202
9 Basic Ultrasonic Imaging 173
:x: Contents
10 Ultrasonic Imaging Using Arrays 204
Imaging Arrays 205
Limitation of Imaging Arrays 208
Electronic Deflection and Focusing 208
Linear Array 208
Linear Array with Deflection 212
Linear Array with Focusing 214
Wideband Responses of a Linear Array 216
Two-Dimensional Array Systems 218
Rectangular Array 218
Concentric Ring Array 220
Annular Ring Array Transmitter 221
Theta Array 222
Problems 224
7
Introduction
"{>f'o,nl"frn,p.(j primarily with the creation of images of structures within
object. Although the object studied will be the human body,
ge developed will be applicable to a variety of nonmedical applica-
.' nondestructive testing.
an body consists of tissues and organs which are primarily water,
es, with water being the dominant constituent. A wide variety of
are present, such as iodine in the thyroid, tellurium in the liver,
e blood. These elements playa minor role in medical imaging.
to change, especially with the advent of computerized tomog-
and air, however, dominate the ability, or lack of it, to
with various types of radiation so as to create images.
made to provide a complete chronology of medical imaging.
oversimplified fashion, we attempt to highlight the role of the
and engineer in some historical context.
2 Introduction
TRANSMISSION OF ELECTROMAGNETIC ENERGY
I
Iii
I I
I
I I
I
--.-j DIAGNOSTIC X-RAY
I SPECTRUM
I
I
I
I
I
/
100m 10m 1.0m 10cm 1.0cm 1.0 A 0.1 A 0.01 A 0.0001 A .,
+ t 0.001 A
2 X 10-
22
3.7 X 10-
44
FREQ;Y AND PHOTON ENERGY
FIG. 1.1 Transmission of EM waves through 25 em of soft tissue.
It is instructive to the entire electromagnetic spectrum, from de to
cosmic ray.s, a regIOn to do suitable imaging of the interior regions. The
relative SUItabIlIty be evaluated based on two parameters, resolution and
attenuation. obtam a useful image, the radiation must have a wavelength
under 1.0 em m the body for resolution considerations. In addition, the radiation
should be reasonably attenuated when passing through the body. If it is too
highly attenuated, transmission measurements become all but impossible because
of noise. If it is almost completely transmitted without attenuation, the measure-
ment cannot be made with sufficient accuracy to be meaningful. The attenuation
can be due to absorption or multiple scatter.
Figure 1.1 is an attempt to illustrate the relatively small region of the
electromagnetic which is suitable for imaging of the body. In the
long-wavelength regIon at the left, we see immediately that we have excessive
attenuation at all but the very long wavelengths where the resolution would
TRANSMISSION
10
0
10-1
'"
10-2
\
10-3
10-4
10-5
\
10-6
10-7
10-8
10-9
10-10 1. The measurement of the transmission of x-rays through the body
2. The measurement of the reflection of ultrasonic waves
through the body
3. The measurement of gamma rays emitted by radioactive
which have been selectively deposited in the body
Present medical imaging systems in clinical use consist of three basic
We consider each of these briefly.
The earliest use of techniques of this type goes back to the discovery of
x-rays by William K. R0entgen in 1895. Many of the major systems' contribu.
tions to radiography, such as intensifying screens, tomography (imaging of a
specific plane), and the rotating anode tube, came within the next 10 to 20 years.
Thus most of the efforts in radiography, since the 1930's, have been toward
improving components rather than systems.
It is interesting to note that during the latter period, profound improvements
in internal visualization of disease processes were achieved by the creativity of
clinicians rather than physical scientists and engineers. A variety of procedures
were developed for selectively opacifying the regions of interest. These included
intravenous, catheter, and orally administered dyes. Thus the radiologist, faced
with the limited performance of the instrumentation, devised a variety of
procedures, often invasive to the body, to facilitate visualization of otherWise
invisible organs.
Beginning in the 1950's, and reaching a peak in the mid-1970's, we entered
a revolutionary era in diagnostic instrumentation systems. New systems
conceived of and developed for noninvasively visualizing the anatomy
disease processes. Here the physical scientist and engineer have played
dominant roles, with clinicians being hard-pressed to keep up with the ImlmeIllseil
amount of new and exciting data.
This revolution began with nuclear medicine and ultrasound, which
serious imaging limitations, provided noninvasive visualization of
processes which were otherwise umwailable. The new era reached its peak
the introduction of computerized tomography in the early 1970's. Here
cross-sectional images were obtained that rivaled the information obtained
exploratory surgery. These instruments rapidly proliferated and became
standard technique for a wide variety of procedures.
This revolutionary process is continuing, not only with profound Imi)r01ie:[.
ments in these systems, but also with initial studies of newer speculative >111,"-""111<1_
modalities. These include the use of microwaves and nuclear magnetic
MEDICAL IMAGING MODALITIES
Comparison of X-Rays and Ultrasound 5
Ibcity of propagation to calculate the depth. Unfortunately, in the
the. fiii,nbvend where soft tissue imaging is suitable, air exhibits excessive
uency a , , l' 'bl
freg(?sC!\; This is not a problem in x-rays, where air attenuatIOn IS neg Igi e.
attenua .... regions of the anatomy, primarily the lungs, cannot be studied by
Th
US cettalD . h "t d b
c. 'i . cd Fortunately the entire thoraCIC or c est regIOn IS no covere y
ltraS
Oun
.' d h d'
U . c..Th.. I'S an opening in the front part of the left lung calle t e car lac
1
gs ere , b '
un 'Th' window allows ultrasonic studies of the heart, WhICh are ecommg
notch. IS.
. easingly Important.
mcr .
. ' tructive to compare the two modalities which are capable of probing the
It IS IDS .'" h' h h 'd
d x-rays and ultrasound, One Important dlstmctlOn, w IC .. as receIve
bo debate, is that of toxicity, Although diagnostic x-ray levels have
conSI 'd bi d . d' ,
b
considerably reduced over the years, there are conSI era e ata m Icatmg
een b'l' f d' h
a small damaging effect which can increase .the proba Ilty 0 Iseases suc as
er
leukemia and eye cataracts. The eXIstence of a damage threshold con-
canc , ' , .
tinues to be a source of controversy. The preponderance of data at thIS
ears to indicate an absence of any toxic effects at the presently used dIag-
apPstic levels of ultrasound. These levels are well below those which produce
no " d
measurable temperature changes or cavitation. This apparent tOXICIty a vantage
for ultrasound has made its use more and more popular in potentially sensitive
regions such as the pregnant abdomen and the eyes.
Over and above toxicity, ultrasound and x-rays have a number of other
profound differences in their imaging characteristics. Ultrasonic waves, in
travel at about 1.5 X 10
3
m/sec, while x-rays have the free-space velOCity of
electromagnetic waves of 3 x 10
8
m/sec. This difference essentially means that
pulse-echo or time-of-flight techniques are relatively straightforward in ultra-
sound and extremely difficult in x-ray systems. The propagation time in 1 cm of
wateris 6.7 fJ,sec using ultrasound and 33 psec using x-rays. Thus state-of-the-
art electronic techniques can quite readily resolve different depths using ultra-
sound but cannot begin to accomplish range gating using pulsed x-ray sources.
In addition, the velocity of propagation in the x-ray region is essentially
independent of the materials used. Thus the refractive index of all structures is
unity. The only mechanisms of interaction are absorption and scattering. \his
lack of refractive properties has both desirable and undesirable
It is desirable in that the transmitted radiation travels in straight lines thro-hgh
the body and is undistorted by different tissue types and shapes. The undesirable
aspect is that lenses cannot be constructed, so that selective imaging of specific
planes is difficult. Ultrasound, on the other hand, has a wide variation in
refractive indices of materials. Thus lens effects and focusing structures can be
readily obtained. These same refractive effects cause some distortion within the
COMPARISON OF X-RAYS AND ULTRASOUND
Introduction
Having defined the x-ray region as the only suitable part of the
spectrum for studying the body, we now investigate the potential of aC()USI:ICI.
radiation. The velocity of propagation of sound in water, and in most
tissues, is about 1.5 x 10
3
m/sec. Thus, as with electromagnetic radiation,
resolution criteria eliminate wavelengths longer than about 1.0 cm. We
concentrate on the frequency spectrum well above 0.15 MHz.
The attenuation coefficient in body tissues varies approximately
tional to the acoustic frequency at about 1.5 db/cm/MHz. Thus
attenuation rapidly becomes a problem at high frequencies. For the
parts of the body, as in abdominal imaging, frequencies above 5 MHz are
never used and values of 1.0 to 3.0 MHz are common. For the imaging
shorter path lengths, as occurs in studies of the eye or other superficial
frequencies as high as 20 MHz can be used.
In ultrasound, unlike with x-rays, reflection images are produced,
render the system unusable. Here the transmission through soft tissue can
approximated as exp (-201/ J...), where I is the path length.
In the intermediate regions of the spectrum corresponding to the infrare
optical, and ultraviolet regions, we again have excessive attenuation due to bot
absorption and scatter at the myriads of tissue interfaces. This excessive atten
tion continues, as shown, into the soft x-ray regions.
Between 0.5 and 10-
2
A, corresponding to photon energies of about 25 ke
to 1.0 Mev, the attenuation is at reasonable levels with a wavelength far shorte
than the resolution of interest. This is highly desirable since it ensures tha
diffraction will not in any way distort the imaging system and the rays w'
travel in straight lines. This is clearly the suitable region for imaging and repre
sents the widely used diagnostic x-ray spectrum.
At shorter wavelengths, with the energy per photon hv getting increasingl
higher, the attenuation becomes smaller until the body becomes relativel
transparent and it ceases to be a useful measurement. Also, at these shorte
wavelengths, the total energy consists of relatively few quanta, resulting in poo
counting statistics and a noisy image.
Before concluding our look at the electromagnetic spectrum we shaul
point out that the wavelengths in the microwave region represent wavelengths'
space. The dielectric constant of water in the microwave region is about 8
resulting in a refractive index of about 9. Thus if the body and a microwa
source-detector system are immersed in water, the wavelengths are reduced
about an order of magnitude. In this case the attenuation of a 1.0-cm wave i
water is no longer prohibitive and a marginal imaging system can be considere
Experiments in this regard have been made [Larsen and Jacobi, 1978].
IMAGING WITH ACOUSTIC ENERGY
6 Introduction
body. Fortunately, most tissues have comparable propagation velocities, with
bone and regions containing air being the primary exceptions.
Diffraction effects occur where the object of interest has structure compara_
ble to the wavelength. In x-rays, having wavelengths less than 1 A, these are
nonexistent. However, in ultrasound the wavelength of approximately 0.5 rom
can cause diffraction effects in tissue. To achieve their desired resolution proper_
ties ultrasonic imaging systems are required to operate close to the diffraction
limit, whereas x-ray systems do not begin to approach this limitation.
These various properties of the radiation determine which clinical procedures
they are most suitable for. For example, pulse-echo ultrasound is useful in
visualizing the internal structure of the abdomen because of the distinct reflec_
tions which are received at the interfaces between organs and from various
lesions within the organs. Radiological studies of the abdomen, using x-rays,
show little of the internal organs because of their comparable transmission and
close packing. Special procedures are often used to apply radiopaque dyes
selectively to visualize specific organs. The thorax or chest, on the other hand'
essentially unavailable to ultrasound because of the air in the lungs. '
raphy, however, achieves significant visualization in the chest because of
considerable differences in attenuation of air, soft tissue, and bone. Many
examples exist where the relative capabilities of the two modalities cOlmpJerrlenti,1
each other.
NUCLEAR MEDICINE
In nuclear medicine radioactive materials are administered into the body a
are selectively taken up in a manner designed to indicate disease. The gam
rays emitted from these materials must be high enough in energy to escape
body without excessive attenuation. Thus higher energy gamma-ray emitters a
usually desirable, the only exception being that they are more difficult to im
and to detect efficiently. The energy range 25 kev to 1.0 Mev used in nuel
medicine is comparable to that of radiography, although somewhat higher
the average.
Nuclear medicine has a number of interesting features which make it v
useful in diagnosis. Very small concentrations of materials are required
visualization, unlike the significant amounts of radiopaque dyes required
transmission radiography. Noninvasive intravenous administration of mate'
is used as compared to invasive catheterization, which is often used in radi
raphy. In general, nuclear medicine images look poor in that they hav.e lat
resolutions of about 1 em and are noisy because of the limited number ofp
tons. However, the images have the desirable characteristic of directly indica
pathology or disease processes. In many studies they are taken up only
Nuclear Medicine 7
diseased regions. Radiography, on. the other hand, exhibits high resolution and
relative freedom from nOIse. However, radiographic images essentiall
display anatomy, so th.at disease processes are often distinguished by
of the normal anatomIcal features.
2
Linear Systems
The Superposition Integral, Delta Function, and Impulse Response 9
function. This powerful concept allows us to decompose the image,
system on the individual parts with the system function, and then sum to
P
erate .
o . the desired output Image.
must be emphasized that S in equation (2. I) is a linear system operator
ddoes not apply, in general, to nonlinear systems. For example, consider a
an rnear system, such as photographic film, which exhibits 8. saturation value
non lnd which a further increase in input intensity results in no change in the
beyo ded density. Equation (2.1) does not apply since we can have two input
:eco:sities, II and 1
2
, each of which do not reach the saturation value, whereas
jDte d h' 1 H . h 1' f
th 'r weighted sum excee s t IS va ue. owever, smce t e meanty property 0
(2e;) provides a variety of powerful techniques, we often approximate nonlinear
into a linear form to aid in our understanding, despite the errors in the
sy proxirnation. Linearization techniques of this type are often employed in
:rectronics, where piecewise linear models are used to represent nonlinear-device
characteristics.
THE SUPERPOSITION INTEGRAL,
DELTA FUNCTION, AND IMPULSE RESPONSE
Here the delta function at Xl = Yl = 1] has sifted out the particular value
(2.2)
(2.3)
(2.4)
ff c5(x, y)dxdy = 1.
rx,2 exp [-nrx,2(x
2
+y2)] = c5(x, y).
. decompose our input function gl(X
1
, Y1) into an array of these twq-
dimenSIOnal delta functions by using the sifting property of the delta function
as given by ,
The linearity property expressed in equation (2.1) enables us to express the
response of any linear system to an input function in an elegant and convenient
manner which provides significant physical insight. We first decompose our
input function into elementary functions. We then find the response to each ele-
mentary function and sum them to find the output function.
The most convenient elementary function for our decomposition is the delta
function <5(x, y). The two-dimensional delta function has infinitesimal width in
alI dimensions and an integrated volume of unity as given by
The delta function can be expressed as the limit of a two-dimensional function
whose volume is unity. One example is the Gaussian, as given by
Many of the phenomena found in medical imaging systems exhibit /in
behavior patterns. For example, in a nuclear medicine wh
the intensity of the emitting sources double, the resultant Image mtenslty
double. Also, if we record an image intensity due to a first source and t
another due to a second source, the image intensity due to both sources ac
simultaneously is the sum of the individual image intensities. These two p
erties, scaling and superposition, define a linear system. This can be expre
formally as
Sfal
1
(x, y) + bI
2
(x, y)} = aSf!l(x, y)} + bSf!2(X, y)}
where S is the system operator, a and b are constants, and II and 12 are the
input functions. The functions are shown as two-dimensional since we are dea
with images.
The system operator Sin imaging systems is typically a blurring fune
that smears or softens the original image. As will be discussed subsequent
can be a convolution operation with the point-spread function of sys
Thus equation (2.1) is stating that the weighted sum of two blurred Irna
equal to the weighted sum of the two images which are then blurred by
DEFINITION OF LINEARITY
11
(2.11)
Magnification
oose h(xz, Y2; 1]). Knowing this response for all input points enables us
rescaPlculate the output 12(x2,Y2) for any input-source distribution II(XI'Yl) by
to
using (2.9).
In many linear systems we the. added simplification that the impulse
response is .for all Tn case the impulse response
merely shifts Its pOSitIOn for dIfferent mput pomts, but does not change its
functional behavior. Such systems are said to be space invariant. The impulse
response or point-spread function thus becomes dependent solely on the
difference between the output coordinates and the position of the impulse as
given by
SPACE INVARIANCE
h(x
2
, Yz; 1]) = h(x
z
- - 1]). (2.10)
Thus, referring to Fig. 2.1, the system is space invariant if the image of the
point source has the same functional form, but only translates, as the point
source is moved around the Xl' Yl plane.
The output function under these conditions becomes
A magnified image, in the general sense, is space variant since the impulse
is not dependent solely on the difference between input and output
coordmates. This is seen in the pinhole imaging system shown in Fig. 2.2. With
This is the two-dimensional convolution function, which is often abbreviated as
gz = gl ** h (2.12)
where the dual asterisks indicate a two-dimensional convolution. This convolu-
tional relationship clearly indicates the "blurring" of the gl input function by
the impulse response h. It is a very convenient form since its Fourier transform
is a simple product relationship whereby the transform of the output function
is the transform of the input function multiplied by the transfer function H,
the transform of. the impulse response. This provides an elegant relationship
between the two-dimensional spatial frequencies of the input and output.
. In medical imaging systems the impulse response h varies gradually for
mput coordinates. In these it becomes convenient to define regions in
which this variation is negligible. These space-invariant or isoplanatic regions
can be analyzed using the convenient convolutional form and thus have the
benefits of having a transfer function in the Fourier transform domain.
MAGNIFICATION
(2.
(2.
Linear Systems
Within the integral we have the system response to a two-dimensional del
function. This is known as the impulse response or point-spread function and
given by
which is the output function at X
2
, Y2 due to an impulse or two-dimensio
delta function at Xl = = 1]. Substituting (2.8) into equation (2.7),
we obtain the superposition integral, which is the fundamental concept that
derive from the linearity property (2.1). This enables us to characterize an out
function completely in terms of its response to impulses. Once we kn
h(x
2
, Y2; 1]) for all input coordinates, we can find the output due to any in
function gI'
As a simple example, consider the elementary imaging system of Fig.
Here a planar array of sources fl(X
I
, YI) are separated from a surface where
output intensity f
2
(x
2
, Y2) is recorded. As can be seen, the output is blurred
to the spreading of the radiation over the distance between planes. An imp
source at X I = YI = 1], thus results in a diffuse blur having an imp
FIG. 2.1 Elementary imaging system.
Using the linearity property of (2.1) we structure 1]) as a weighting fact
on each delta function so that the output becomes
of g 1 at that point. The output function g2(X2, Y2) is given by the system Operat
S as
g2(X
2
, Y2) = S{gJx1 , YI)}'
Using the delta function decomposition, we obtain the output function as
glx
2
, Y2) = S {f} 1])b(Xl - - .
10
12 Linear Systems
Two-Dimensional Fourier Transforms 13
(2.18)
I
(2.17)
g(x, y) = g:-l{G} = ff G(u, '1.') exp [i2n(ux + vy)]dudv.
G(u, v) = g:fg(x, y)} = ff g(x, y) exp [-i2n(ux + vy)]dxdy
Conditions
i,';.',.
.Before using the Fourier transform we must look at the required mathe-
matlcaLconditions on the fu'nction g(x, y) for its Fourier transform to exist.
The folloWing are the more important sufficient conditions on the function.
where;f is the Fourier transform operator, and u and v are the spatial frequencies
in the x and Y dimensions. Thus the two-dimensional function is being decom-
posed into a continuous array of grating-like functions having different period-
icities and angles. Each u, v point in the Fourier space corresponds to an
elementary "plane wave" type of function in object space. This complex expo-
nential function has lines of constant phase separated by (U
Z
+ VZ)-l/Z and at
an angle of tan-I (ujv) with the x axis.
into spatial frequencies u and v, having dimensions of
distance, provides a direct measure of the spatial spectrum and
bandwidth. The original function can be subject to spatial frequency filtering
degradations in the system. We then use the inverse Fourier
find the resultant object function as defined by
dent of the x I' Y1 coordinates, so that the aperture collects the same fraction,
pen f h . . f h .
independent o. t e o. t e pomt. If this assumption is not made, and the
I
Jectio
n
efficIency IS a functIOn of Xl and Yl' the system becomes space variant
co d . .
and cannot be structure In convolutIOnal form.
Electrical and communication engineers are very familiar with Fourier analysis
of various signals where a function of time is decomposed into an array of
complex exponentials of the form e
iW
!. One of the incentives for this decomposi-
tion is that the complex exponential is the eigenfunction of invariant linear
systems. That is, when a complex exponential signal is applied to an invariant
linear system, such as a network or filter, the same complex exponential will
appear in the output with various amplitude and phase weightings. Thus, with
the Fourier transform decomposition, the effect of filtering a signal is greatly
facilitated by operating in the frequency domain.
In imaging systems we use the two-dimensional Fourier transform G(u, v)
as defined by [Bracewell, 1965]
TWO_DIMENSIONAL FOURIER TRANSFORMS
(2.1
FIG. 2.2 Pinhole imaging system with magnification.
I
f-a---+-----
b
----.....\
IzCx
z
, Yz) = JJII - ,', Yz - 1'/')d,'d1'/'
= ** h(xz, Yz)
Using the substitution of (2.15), we have been able to use
form. Physically, we have convolved the appropriately magmfied mput I
with the impulse response so as to avoid the disparity between
systems. II(xzjM, yzj M) is the output image we would get using mfimtes
pinhole camera which experiences magnification. but no bl.urfIn
g
. The 1
factor is the loss of intensity due to the magnificatIOn of the Image.
The impulse response h(x
z
, yz) is a magnified .of the
function where the magnification factor is (a + b)ja. In addItIOn, 1m
response includes a collection efficiency term relating to fractIOn
intensity of each point in II is collected by the aperture or pmhole. In eq
(2.13) we effectively assume that the collection efficiency is a constant,
giving
the pinhole on axis, the impulse response by geometry becomes
h(x
z
, yz; " 1'/) == h(xz - M" Yz - M1'/)
where M = -bja. Thus the output intensity is given by
Iz(x
z
, Yz) ==. SS1
1
(" 1'/)h(xz - M', Yz -- M1'/)d'd1'/. (2.1
This equation can be restructured into the desirable convolutional form [Go
man, 1968] using the substitutions
t :-::: M, and 1'/' = MYJ
RELATIONSHIPS
ing relationships,
g:[g(x, y)} = G(u, v)
;f[h(x, y)} = H(u, v).
oblems to know the relationship between manipula-
ose of the frequency domain. For example, what is
spectrum if an object is magnified or shifted? The
iationships are given below without proof. The proofs
y consulting references [Bracewell, 1965; Goodman,
g:[ag -+ Ph} = ag:[g} + pg:[h} (2.22)
operation is linear, so that the transform of the weighted
.....h"'nc1<:: the weighted sum of their individual transforms.
g: {ff l1)h(x - y - = G(u, v)H(u, v)
Convolution
Shift
g:{g(x - a, y - b)} = G(u, v) exp [-i2n(ua + vb)] (2.24)
Translation of a function in object space introduces a linear phase shift in the
frequency domain.
(2
t
5)
f
The convolution of two functions in space is represented by simply multiplying
their frequency spectra. This relationship occurs very frequently in medical
imaging, where a spatial distribution g is blurred by an impulse function h.
The two-dimensional convolution operation is abbreviated as g ** h.
;f{g(ax, by)} = I IG( (2.23)
The stretching ofcoordinates in one domain results in a proportional contraction
in the other domain together with a constant weighting factor.
14 Linear Systems
Before becoming overly involved in these conditions, we must re
this book relates to an applied science, so that we are interested pri
studying the transforms of real physical phenomena. These will in
variety of two-dimensional source distributions or two-dimensional trans
functions. These physical distributions, by definition, have a transfor
as pointed out by Bracewell [1965], "... physical possibility is a valid su
condition for the existence of a transform." In our analysis, however
often use convenient idealized distributions to study various system p;o
These include the delta function, sinusoidal distribution, .dc or constant
and so on. Each of these useful functions violates one or more of the pre
listed sufficient conditions. It would be unfortunate, however, to a
these and lose the resultant insights. Instead, we use them in a limiting
so that the conditions remain satisfied.
One example of this limiting process is the representation of a two-d
sional delta function as the limit of a two-dimensional Gaussian funeti
given by
;f[a
2
exp (-a
2
nr
2
)} = exp (_
where p is the radial variable in Fourier transform space and p2 = u
2
Applying the limiting process to equation (2.20), we obtain
g:{<:5(x, y)} = exp ( _ = 1.
Thus the Fourier transform of the delta function is uniformly distribute
the two-dimensional frequency domain.
The limiting process has allowed us to stay within the required
matical conditions and yet evaluate the transform of a function that
our third condition. The particular function, <:5(x, y), is very useful,
transform. We can also use the same limiting procedure to evaluate the
form of a constant term of unity amplitude, a function that violates
condition. In this case we define the constant term as the limit of
Gaussian function, where a now approaches zero. Taking the limit
transform, we evaluate the Fourier transform of unity as <:5(u, v) a delta
in the frequency domain.
where r
2
= x
2
+ y2. The Fourier transform of this function (see Table
given by
1. The function is absolutely integrable over the entire domain.
2. The function has only a finite number of discontinuities and
number of maxima and minima in any finite region.
3. The function has no infinite discontinuities.
---------
16 Linear Systems
Frequently Occurring Functions and Their 'T' .1: .I ransJorms 17
Cross Correlation
transform becomes
G(p, ) = G(p) = 2n f'" rg ( )J (2 oRr 0 nrp)dr.
This Hankel transform of zero ord . f (2.35)
transform. er IS 0 ten referred to as the Fourier-Bessel
FREQUENTLY OCCURRING FUNCTI
AND THEIR TRANSFORMS ONS
We first present the transforms of a b
(Table 2.1). num er of well-known continuous functions
TABLE 2.1
(2.28)
so UJ q)h*(x + y + G(u, v)H*(U, v) (2.26)
g(x, y) ** hex, y) = g(x, y) ** h(-x, -y).
If we set g = h, we form an autocorrelation where
;rfg(X, y) ** g* (x, y)} = G(u, v)G*(u, v) = \G(u, v) p'.
This relationship is abbreviated using stars as g h*. It is closely related to
the convolution with the equivalence expressed by
G(p, cf = 2n anC- i)ne
i
n4> rglr)JnC
2nr
p)dr.
For the important case of circular symmetry, with no 8 variations,
g(r,8) = gR(r).
____ OVf'pnt for an. which is unity. The
Transform
comb (u)
comb (u) eomb (v)
J t (2np)
p
1 .
2" o(u) - _l_
2nu
sine (u) = sin nu
nu
sine (u) sine (v)
sine
2
(u)
sinc
2
(u) sine
2
(v)
1
Transform
1
2i [o(u -- 1) - o(u + 1)]
1
T[o(v - 1) + o(v + 1)]
o(u - 1 V _ 1)
.,;, .,;
exp (-np2)
o(u, v)
Function
Function
sin 2nx
cos 2ny
exp [in(x + y)]
exp (-nr
2
)
1
{
I IX I :::;;i-
= 0 otherwise
rect (y)
Ixl:::;;1
otherwise
xo, y - Yo) = o(r - ro)o(8 - ( 0 )
r
= O(x - n)
n=-o:)
comb (y)
{
I r:::;; 1
o otherwise
{
I x20
o otherwise
define a number of specialized functions a
wlttiQ.tft.d.eri'v::ltlf\n (Table 2.2). Some of th . nd presenttheir transform
ese are Illustrated in Fig. 2.3.
TABLE 2.2
geC8) = l: ane
inB
.
",=_00
If g(x, y) is separable in its rectangular coordinates as
g(x, y) = gx(x)gy(Y)
Separability
We then evaluate the transform of each angular harmonic by making use
relationship
;rtgir)einB} = (_Wein4>JCnlgir)1
where JC
n
l.1 is the Hankel transform of order n, which is given by
JC
n
lglr)1 = 2n rglr)Jn(2nrp)dr
where p and cf> are the Fourier transform variables in polar
the complete transform is given by
G(p, cf = ;rfgir)gi8)} = t anC_i)n
e
in4>JCJglr)1
n=-OO
;rfg(X, y)} = ;rxfgx};rytgy}
where ;r x and ;ry are one-dimensional Fourier transform operators. This
ship occurs often with simple objects and is a convenient simplification
evaluating the transforms.
A variety of interesting relationships result when the function g(x,
separable in polar coordinates, as given by
g(r, 8) = gir)gi
8
).
we can decompose the 8 variation into its angular harmonics as
then
Sampling 19
...-x
(2.42)
(2.41)
(2.39)
(2.38)
G(u, v) = G.(u, v)H(u, v).
)comb ( ; )g(x, y)] ** hex, y)
g(nX, mY)o(x _. nX, y - mY)
9h to restoring an appropriately bandlimited image from its
ier transform the samples, multiply the resultant spectrum by
and then inverse transform the result. An equivalent method
Jation functions as derived by taking the Fourier transform
. here
consists of a replicated array of the spectra G(u, v). If these spectra
verlap, the central one can be isolated through frequency-domain
will thus reproduce G(u, v). Since each spectra is separated by 1/X
filter for isolating the central replication is given by
H(u, v) = rect (uX) rect (v Y). (2.40)
rlap the maximum spatial frequencies present in the image are
i,s and 1/2 Yin the v axis. If the image is subject to this bandwidth
original spectrum can be restored using the filter of equation
(2.43)
,mY) sinc [lex - nX)] sinc (x - mY)} (2-j4)
he elegant in that it is a weighted sum of two-dim1en-
seach sample at x = nX, y = mY is used to weight a
ion centered at that point. The sum of these functions
indicated, it is appropriately bandlimited.
an be restated as the image having no frequency
X and 1/2 Y, half of the sampling frequencies 1/ X
These samples will completely represent g(x, y) if it is adequately bandlimited.
The spectrum of the samples g.(x, y), is given by
G.(u, v) = ;r[g.(x, y)} = XY comb (uX) comb (vY) ** G(u, v). (2.37)
The comb functions are each infinite arrays of delta functions separated by
1/ Xand 1/ Y, respectively. Since convolution with a delta function translates a
function, convolution with a comb provides an infinite array of replicated
functions. Thus equation (2.37) can be rewritten as
00 00 ( n m)
G.(u, v) = nI:oo G u - X' v - y
telationshiP
XYcomb (uX) comb (uY) = nX:oo o(U - ;, v - ;).
\ 1
0
-- x
___1 0 0.5
-0.5
I\(x)
______
-1.0
RECT x
d cial functions.
FIG. 2.3 Often encountere spe
SINC x
0 .
J,(21fP/P)
SAMPLING
. h se involving computer 0
. . erations, especIally to. b a finite array
In
many Imagmg op d'mensional functIon y 1
f uous two- 1 mplete Y
we represent a con m d which these samples co . g(x
W
e study conditions un er'
d
a two_dimensional ,
f n Consl er d It functlons
resultant
samp e . and Y in the y 1m
in the x dimensIon .
( Y
) is given by
g. x, (X) comb (L) g(x, y).
g.(x, y) = comb X Y
Linear Systems
18
20 Linear Systems
and 1/ Y. Any greater image frequency components will result in overlap of the
spectral islands. This overlap cannot be removed by filtering and results in
aliasing, where the higher-frequency components reappear at incorrect fre-
quencies.
ELEMENTARY PROBABILITY
A number of considerations in medical imaging involve simple probabilities or
stochastic considerations. This is especially true when studying the noise
properties of the signals or images. We first define the distribution function F(x)
of a random variable x, which is the probability P that the outcome X of the
event will be less than or equal to x as given by
Elementary Prohahility 21
uncertainty in this component th .
or e nOise po If
such as voltage or current its s wer. X represents an amp)'t d
. b ,quare represent th 1 U e
varIance ecomes the noise power. s e power and therefore .
In general th . Its
. ,esIgnal-to-noise ratio (SN .
to-nOIse power. For example in th I R) IS defined as the ratio of s' 1
. I (). ,eu trasound s t Igna -
a sIgna e t IS received from sound fl ys ems of Chapters 9 and 10
. (t) f h re ected from a . ,
nOIse n rom t e transducer ad' regIOn of interest Th
density as given by n amplIfier follows a Gaussian proba'bilit;
p(x) = (J' 1_ exp [-J... (X - X)zJ.
,.,/21& 2 (J' (2.51)
In the case of the noise signal net) i .
the resultant SNR given by , ts mean IS zero and its variance I'S 2
(J', WIth
(2.54)
PROBLEMS
aging system of th .
iu . e type Il1ustrated in Fi
. s R. USIng the geometry show g. 2.2 Uses a circuI'jlr
find the output spat' I fi n, and assuming a constapt
0
put
Spatial frequency r/eq(uency spectrum /z(u, v) in
m 1 u, v).
transform of zero order often
transform of the fioll '. called the Fourier-Bessel
Owmg.
h
2' h (J'; (2.52)
were, (J/I IS t e variance as shown in .
the x-ray and gamma-ray .equatIon (2.51).
of the energy in the form the noise is dominated by the d'
tu I a countable numb f Iscrete
. re e The probability density fun f er 0 recorded at
IS the POIsson function given by CIon descflbmg the number of
. k! (2.53)
IS the probability of exactly k h
pixel. equations the average is 2
lance (J'k = A.. Smce the numbe f h ' :v
e
find the mean k = A.
n by r 0 P otons IS a unit of energy, the
SNR = = ,.,/T.
in detail in Chapter 6.
(2.45 P[X S x] = F(x).
P[x) S X S XZ) = F(x
2
) - F(x
1
) = IX' p(x)dx
x,
x= E(X) = I xp(x)dx.
The expected value of the nth moment of X is given by substituting x
n
fJ
equation (2.48). An important statistical parameter is the variance 0';,
represents the second central moment of X as given by
o'} = E[(X - X)Z) = E(X2) - X
z
.
The standard deviation (J x, the square root of the variance, represents t
mean-square (rms) variation of X about the mean Xas given by
(Jx = [E(X
2
) - XZ)112.
If X represents the energy or power, the standard deviation is the
The probability P[) has the property that its value lies between zero
one. F(x), the distribution function, is monotonically increasing from zer
one as x goes from -00 to 00. The probability density function p(x) is al
positive and its integral over all x values is unity. The probability density fune
p(x) takes on a variety of forms, including Gaussian, uniform, and so on. I
cases where x takes on only discrete values, such as the outcomes of dice,
is a sum of delta functions.
The mean of the outcomes, or expected value of X, is given by
The probability that the outcome X is between two values, Xl and x 2, is therefor
given by
where p(x) is the probability density function, which is defined as
d
p(x) = dxF(x).
22 Linear Systems
(b) rect (r ba). where a> b.
(c) gr(ar), where 5{gr(r)} = G(p). .
3 P th
e following properties of b functIOns.
2. rove _ b)
(a) f(x, y) b(X - a, y - b) = (a, b)b(x - a, Y .
. b) - f(x - a y - b).
(b) f(x, y) ** b(X - a, Y - - ,
(c) b(ax, by) = \;b \ b(X, y).
b(r - ro)b() - ()o).
(d) b(x - xo, Y - Yo) = r
b(r)
(e) b(x, y) = -:;:r'
2.4 Prove the following Fourier transform relations.
(a) y)}} = g(-x, -y).
(b) * h} = GH.
1 (U ..!..).
(c) by)} = \ab \ G a' b
x _ a y - b)} = G(u, v) exp [-i21t(ua + vb)].
(d) . robability of being between the
2.5 A random variable x has umform p
a and b.
(a) Find the probability density function p(x).
(b) Find the mean value of the variable
(c) Find the standard deviation of the vanable (J x'
3
Physics of
Projection Radiography
rojection radiography could be referred to as conventional radio-
more familiar, although less descriptive. It refers to the bulk of
where the transmission of x-rays through the body is recorded
arious attenuation effects, we use the simplified geometry of Fig.
sollimated x-ray source is assumed such as would be produced
tee at infinity. In this assumption we avoid the distortions 4lue
rce reasonably close to the object. These distortions are cbn-
I in Chapter 4. This beam is partially absorbed and scattered in
with the remaining transmitted energy traveling in straight
ctor. For purposes of this discussion, it will be assumed that
tie is sufficiently far away so that all scattered radiation fails to
or. The effects of scatter are considered in Chapter 6.
2S
(3.5)
(3.6)
Source Spectrum
A number of sources of energy exist in the x-ray spectrum. However, thus far,
only the x-ray tube, where an energetic electron beam strikes a metal target, has
shown sufficient intensity to provide a usable image in a reasonable exposure
interval. Radioactive isotopes normally have insufficient intensity, although
many do provide monoenergetic radiation.
The source of x-ray energy from x-ray tubes has the Bremsstrahlung
(braking) radiation spectrum [Ter-Pogossian, 1967] derived from collisional
interactions between electrons and matter. This is the energy produced by the
deflection and deceleration of electrons by the nucleus of the atoms in the
material being bombarded. The energy is emitted in the form of x-rays or
high-energy photons whose energy depends on the electron energy, the charge
of the nucleus, and the distance between the electron and the nucleus. The
electrons are accelerated toward the target by an anode voltage E.
In thin targets, a uniform photon energy distribution is produced whose
intensity is proportional to the atomic number Z and whose maximum photbn
energy is the electron energy E. The spectrum is relatively uniform since dif-
ferent events produce different numbers of photons. For example, the photon
can give up all its energy to a single emitted photon of energy E. Similarly, it
can produce n photons, each of energy Eln. Thus the average intensity at each
photon energy is similar. A thick target, such as is used in conventional x-ray
Sources, results in a triangular rather than uniform energy spectrum, as shown
SOURCE SPECTRUM
where Po is the attenuation coefficient at 00'
Id(x, y) = f 10 (0) exp [ - f flex, y, z, o)dzJdo (3.4)
where 1
0
(0) is the incident)x.-raY
h
intensity a funcffitio? of the enher
gy
per
n 0 and flex, y, z, 0 IS t e mear attenuatIOn coe clent at eac regIOn
object under study. The bracketed term represents the intensity transmis-
:t each ....... y coordinate and at each photon energy. Thus the intensity
nsmiss
ion
fthrough the object at photon energy 00 through a thickness I
tra ..... b
at a given x.,ypoSltlOn IS given y
- t(x, y, 00) = exp [ - s: fleX, y, z, oo)dzJ.
If P, has a uniform value of flo throughout the volume, t is given by
t(x, y, 00) = exp ( - flol)
.' 'beam contains a spectrum of different energies, and the linear attenua-
'ncldent, , . f . . d
1. mcient fl is a functIOn 0 posItIon an energy.
non on equation (3.3), the intensity at the detector plane fix, y) is
given by
+
I
FILM
RECORDER
x
/
SCATTERED ENERGY
y
J<:---- z
FIG. 3.1 System for studying x-ray attenuation.
Physics of Projection Radiography
I
NOU'dN fX
N = -fl dx.
Nin 0
Solving equation (3.2) we have the classical attenuation relationship
N
out
= Nine-;<x.
Returning to Fig. 3.1, we use this relationship to formulate the detected intey;
I (x y) in terms of the incident intensity 10 , The intensity attenuation
'same relationship as that of the number of photons. The intensity
energy per unit area, which can be expressed in terms of the number ofphq!
per unit area weighted by the energy per photon. In the more general cast
The transmitted photons either interact with a particle of matter or pass u
fected [Johns and Cunningham, 1974]. The interaction results in the rem
of the photon from the beam by scattering or absorption. This intera
between an x-ray photon and a particle of matter does not affect the 0
photons in the beam. The number of photons interacting and removed
the beam bN in a region of thickness bx is given by
bN = - flNbx (
where N is the total number of impinging photons and fl is a constant of
portionality known as the linear attenuation coefficient. bN is negative sin
beam loses photons. As would be expected, the number of.
is proportional to the number of incident photons, the mteractmg
and the material.:
If we start with N
in
photons and, after a thickness x, have Nout
using equation (3.1) in differential form we have the integral relation!
'ti
ATTENUATION RELATIONSHIP
24
The Atte1Ulation Coefficient 27
foneofthe atoms ofthe target material. The electron is ejected and the vacancy
?filled by an electron from another shell. The loss of potential energy in the
IS nsition between shells is radiated as an x-ray photon having an energy equal
that of the transition. !his characteristic radiation occurs at all levels but
is most pronounced at the mner K shell. It represents a significant fraction of the
total radiated energy from the x-ray tube.
The linear attenuation coefficient J-l of all materials depends on the photon
energy of the beam and the atomic numbers of the elements in the material
[Johns and Cunningham, 1974]. Since it is the mass of the material itself that is
providing the attenuation, attenuation coefficients are often characterized by
fl./p, the. mass attenuation coefficient, and are then multiplied by the density to
get the linear attenuation coefficient in units of inverse distance. In the diag-
nostic range, below 200 kev, three mechanisms dominate the attenuation:
coherent scatter, photoelectric absorption, and Compton scatter, as shown in
Fig. 3.3 for water.
Coherent or Rayleigh scattering is the apparent deflection of x-ray beams
caused by atoms being excited by the incident radiation and then reemitting
waves at the same wavelengths. This phenomenon is useful in x-ray diffraction
studies, where the x-ray energies are of the order of a few kiloelectron volts and
thus the wavelengths are the same order of magnitude as atomic dimensions. It
is relatiyelyunimportant in the energies used in diagnostic radiology, as is seen
bytllep!otor,ur!p in Fig. 3.3.
ei1swer energies of interest, the photoelectric effect dominates the
coefficient, as seen by the plot of ,up!p. The x-ray photon in this
interacting with a tightly bound electron. The kinetic energy
eJectron is dissipated in the matter. The vacancy is filled in a
.od of time by an electron falling into it, usually from the next
mpanied by the emission of characteristic x-ray photons called
ion. Lower-energy excitation is absorbed in the M and L shells,
rgy radiation is absorbed in the inner K shell. This absorption
lady important with the higher-atomic-number materials that
'opaque dyes. The mass attenuation coefficient due to photo-
n varies approximately as the third power ofthe atomic nUnPber
o that the linear attenuation coefficient will vary approximately
wer. Thus photoelectric absorption becomes increasingly impor-
-atomic-number materials.
coefficient undergoes a sharp increase in the energy
ding to the K shell. This is known as the K absorption edge.
omic-number elements, such as are found in water and organic
THE ATTENUATION COEFFICIENT
PHOTON
E EN ERGY (ev)
/
/
/
/ r...- EFFECT OF
/ FILTERING
/
/
/
/
RELATIVE
INTENSITY
FIG. 3.2 Energy spectrum of a thick target, showing effect of filtering and
characteristic radiation.
Many of the x-ray photons, especially those at lower energies, are abs
before they leave the x-ray tube. This filtering of the beam occurs in the
itself, the glass envelope of the x-ray tube, and in other filtering structures
the x-ray beam passes through. The effect of this filtering is shown in the
curve in Fig. 3.2. This filtering is normally desirable since these low-
"soft" x-rays have little penetration in the body and can cause skin
without contributing to the transmission image. Often additional filtering
rial, such as aluminum, is added to remove these lower energies. In spe .
studies involving relatively short path lengths and soft tissue, as in rna
graphy, the lower-energy x-rays are used in image forming.
Were it not for the various filtering actions within the tube, the effi
of x-ray production would be proportional to E since the output power'
portional to E2. Because of the filtering action, however, the output
increases at a greater rate than E2.
In addition to the uniformly decreasing photon energy spectrum,
shows the characteristic radiation lines. These are produced by the ace
electrons colliding with a tightly bound electron, usually bound in the
in Fig. 3.2. The thick target can be modeled as a sequence of thin planes
causing a successive loss of electron energy. Thus the spectrum produced
each succeeding plane is again a uniform spectrum whose maximum val
becomes progressively lower. The sum of the radiation assumes a trian
form due to the reduced electron energy at increasing depths. The total emit'
energy because of this triangular form is approximately proportional to Z
where E is the initial electron energy.
26 Physics of Projection Radiography
28 Physics of Projection Radiography
The Attenuation Coefficient 29
/J./p(cm2;g)
0.1
PHOTON
/lIp TOTAL MASS ATTENUATION COEFFICIENT
\
.
ELECTRON
FIG. 3.5 Collision of a photon and an electron, illustrating Compton
scattering.
20
10
FIG. 3.4 Total and components of mass attenuation of lead (p = 11.5 gjcm
2
).
PHOTON ----.----1
0.5
5.0
"
"
'\
,
'\
,
'\
'\
"
\ .
\ \
\
0.2 "ofp COMPTON RAYLEIGH PHOTOELECTRIC
- J._ _ \\ \
- - - \ " PHOTON
0.1l-__...L- --.L.__ ENERGY
10 20 50 100 200 500 1000 (kevl
1.0
2.0
0.5
0.02
0.05
material, this K edge occurs below the diagnostic energy spectrum being
For higher-atomic-number materials, such as lead shown in Fig. 3.4, this
edge occurs within the spectrum of interest. At energies well beyond the
absorption edge, the attenuation' due to the photoelectric effect diminishes
importance. Over the range of interest, Jlp/ p varies as 1/8
3
-----'0-- --
I
'.. zo---....
SOURCE
fd
w
I
I
I
X-RAYS I /xd
z
FIG. P4.2
(c) Calculate this ratio for W = 20 cm, R = 0.5 cm, using the curves
Fig. 3.7, where the soft tissue is muscle having a density of 1.0 and the bon
density is 1.75. Perform the calculation for x-ray photon energies of 3
and 100 kev.
4.3 A cylindrical object having an attenuation coefficient flo is positioned in
point-source x-ray system as shown in Fig. P4.3.
(a) Find an expression for fa, neglecting the falloff of the source intensi.
over the detector plane due to obliquity.
(b) Find an expression for ld using the object in Problem 4.2 with
cylinder in a layer of soft tissue.
4.4 A cylinder of attenuation coefficient fl, radius R, and length L is plac
on the axis of an x-ray imaging system as shown in Fig. P4.4. Neglecti
all obliquity factors, find an expression for ld versus rd, where the intens
in the absence of the object is a uniform 1
0
,
60 Source Considerations in Radiographic 117Ulging
5
Considerations
Recorder Resolution
SNR = C.jl1N C?l)
number of photons per picture element impinging on the recorber,
or capture efficiency, and C is the contrast of the structure
recorder has a high quantum efficiency but, as will be shown,
resolution. Similarly, very thin recorders exhibit negligible blur-
but capture relatively few of the photons.
We far considered resolution limitations due to the x-ray source. The
resolution-limiting factor in the system is the x-ray recorder,
where the image itself is formed. The principal difficulty is in the attaining of
the' ,high resolution while maintaining a relatively high quantum or
cap lency. The quantum or capture efficiency..epresents the fraction of
ph interact within the recorder material. As will be shown in Chapter
6, captured photons per picture element governs the resultant
ratio. This SNR will be shown to be given by
Vd
\-l!.t------------d ----------
I Y I
'0 t: IT
f--L----i
- - - - -,
I
I Zo I
I
I. d ----------+1
FIG. P4.7
4.8 A rectangular source tilted by an angle eis used to image an opaque obj
tilted at 45 as shown in Fig. P4.8. The projection of the source Yis
nificantly smaller than all other dimensions. Neglecting obliquity fact
plot the relative detected intensity in the Yd direction labeling the Yd axi
the break points. [Hint: Assume space invariance in the vicinity of
break points.]
FIG. P4.8
4.9 A source tilted at an angle of 45 has a projected intensity sex, y)
circ (rjr
o
)' It is used to image a transparency at 2 = 2 0, having a tran
sion t(x, Y) = I: o(x)o(y - i). Find the resultant intensity at z = d,
I
lecting obliquity.
4.7 A tilted source is used to image an opaque planar tilted object infinite
extent and containing three pinholes of equal size as shown in Fig. P4
Neglecting obliquity, plot I d versus Yd, indicating the relative amplitud
and the position of the break points. Space invariance can be assumed'
the vicinity of the pinhole images.
62 Source Considerations in Radiographic Imaging
FIG. 5.1 X-ray recording process.
(5.8)
(5.9)
(5.6)
(5.4)
Screen-Film Systems 65
F(x) = 1 - e-
px
.
p(x) = ixF(x) = p,e-
px
.
Since the system is space invariant, frequency res.ponse the system is
. by the Fourier transform of the pomt-spread functIOn, as gIven by
gtven
r'" Kx
H1(p) = g:{h(r)} = 2n Jo (rZ + xZ)3IZJo(2npr)rdr
For a phosphor screen of thickness d, the distribution function is given! by
1 - e-
PX
F(x) = 1 d' (5.10)
- e P
This distribution function represents captured photons only and ignores those
transmitted beyond x = d since they do not contribute to the resultant image.
Thus the distribution function varies from zero to one as x varies from zero
where p, is the linear attenuation coefficient and e-
PX
represents the fraction of
photons transmitted beyond distance x. The associated probability density is
given by
This normalization can be performed since it only the elimination of
constant terms and no functions of x. In general, howeveI2 the normalization
should be performed after the averaging process, where H(p) is obtained, to
ensure that each individual response HI (p, x) is properly weighted in the
averaging process.
The normalized frequency response, H(p, x), is the transfer function
resulting from a photon giving up its energy at a x. In order to find
the average transfer function ii(p) resulting from a large number of x-ray
photons, we integrate over the probability density function p(x) as given by
ii(p) = f H(p, x)p(x)dx = f e-Zrrxpp(x)dx (5.7)
where p(x) is the density function of the point at which the x-ray photons
interact. The probability density function can be determined from the dis-
tribution function F(x) given in equation (2.45) which is the probability that an
x-ray photon will interact within a distance x. For an infinitely thick scintil-
lating phosphor, the distribution function is given by
h e J (2npr)r is the kernel of the Fourier-Besse1 transform for functions
w e: g cirCUlar symmetry and p is the radial spatial fr.. pency variable as given
havIll I'" .. b
in equation (2.35). The resu tant transJorm IS gIven y
H1(P) = 2nKe-
Zrrxp
. (5.5)
It is convenient' to use a normalized frequency response H(p) as given by
SCINTILLATING
SCREEN
LIGHT PHOTONS
FILM /
X-RAY
PHOTON--_-.......--t--JI----.:----
The invariant impulse response her) along the recording film emulsio
governed by two factors, the obliquity factor of the impinging light phot
which follows a cosine law, and the inverse-square-Iaw falloff with dis
from the scintillation point, which has a cos
2
edependence. Thus her), as
the x-ray source in Fig. 4.3, is given by
x
3
her) = h(O) cos
3
e= h(O)(x2 + r2)3/2
where h(O) is the response at r = 0 given by h(O) = Kjx
2
. Thus her) is give
x
h(r)=K(2 I 2)312
r T x
where K is an intensity proportionality constant. With the assumptions
above, this represents the point-spread function of the x-ray recording pr
X-ray film itself is a relatively inefficient recorder of x-ray photons. To ColI
the x-ray photons efficiently, a scintillating screen is used to convert each x-
photon into a large number of visible photons, which are then recorded
film. The scintillating screen is a dense high-atomic-number material, such
calcium tungstate, which will capture the x-ray photons in a relatively sho
path for resolution considerations. This is illustrated in Fig. 5.1. An incomi
photon gives up its energy to a scintillating phosphor at a distance x as Show
A large number of visible photons are generated in the scintillating crys
The exact nature of this process is quite complex [Bates and Marwood, 1973
We will assume an isotropic radiator of visible light at the scintillation poin
Also, we neglect the granular nature of the phosphor and assume unifo
light propagation. It is reasonably accurate to assume that the x-ray phot
does not interact with the film itself.
SCREEN-FILM SYSTEMS
(5.14)
(5.12)
PHOSPHOR
SCREEN
Screen-Film Systems 67
DOUBLE
EMULSION
FILM
FIG. 5.3 Dual-screen recording system.
PHOSPHOR
SCREEN
this can be clearly approximated as
H(h) = k (21lPk + :>(1 - e-Pd>'
b tituting 11 = 1- e-
pd
, the capture efficiency of the screen, and using the
Su snable approximation 11k ], we get an expression for the cutoff frequency:
reasO .
Pk (5.13)
the currently used of is the -10% response.
e for this case the hmItmg resolutIOn. In cycles/m.m would be f.lIO.21l'f/. ThIS
shows the trade-off bet,:een and the
as regards the thicknessd. A coeffiCIent. f.l is always
since it provides for the stoppmg of photons a short dIstance. The thIckness
d however, must be a trade-off between efficIency and the cutoff frequency.
practice a variety of screens are n:ade available 0\ differing so
that the desired trade-off between efficIency and resolutIOn can be achIeved for
each study. .
Modern x-ray recording systems utilize a double screen-film cassette
structure which helps this trade-off among screen thickness, resolution, and
efficiency, as shown in Fig. 5.3. The recording film has a photographic emul-
sian on each side and is sandwiched between two phosphor layers. The efficiency
is now based on the entire phosphor thickness d =. d
1
+ d
2
Scintillations in
either layer will be recorded on one of the two emulsions. The frequency
ponse due to a scintillation at x is given by I
H(p, x) = e-
2np
(d,-x) for 0 < x < d
l
= for d
1
< x < d.
The film itself is essentially transparent to x-rays.
10.0 1.0 0.1
.-.---
."".
'"
.""
0'----------'----------'----------'-.....
0.01
0.6
0.2
0.4
1.0
0.8
FIG. 5.2 Frequency response of a screen-film system.
It is important to establish some figure of merit for the frequency respo
such as a cutoff frequency or effective bandwidth, to evaluate various c
figurations. Referring to equation (5.11), we note that, using typical values,
bracketed expression can be approximated as unity above relatively low spa
frequencies. For example, at p = ].0 cycle/mm, H(p) = 0.53 and the bracke
expression is 0.85. Clearly, in establishing a cutoff frequency, it is conveni
to assume that the bracketed expression is unity. We define the cutofffreque
Pk as that spatial frequency where H(Pk) = k. For values of k less than
Equation (5.11) represents a monotonically decreasing response wi
increasing spatial frequency. A representative value for the thickness d of
typical screen is 0.25 mm. A calcium tungstate screen at the center of t
diagnostic photon energy spectrum will have an attenuation coefficient J1,
about 15 em-I. A plot using these values is shown in Fig. 5.2.
to d. The resultant probability density function is f.le-IIX/(l - e-Jicl) with
normalized spectrum given by .
H(p) = 1 }:e-i-<d fad e-Z7tXPe-i-<Xdx
66 Recorder Resolution Considerations
NORMALIZED
RESPONSE
68 Recorder Resolution Considerations
We will establish an effective cutoff frequency Pk for this configuration as was
done for the previous case. Since dl and d2 are comparable in width, each about
half of d, we can again neglect the exponents of the form exp (- 2npd), since
they will become negligible at relatively low spatial frequencies. We also make
the approximation
The frequency response averaged over many events using equation (5.7) be.
comes
Ii(p) = 1 I'd {fad
l
exp [-2np(dj - x) + J.lx]dx
+ exp[-2np(x - dj ) + J.lX]dX}
(5.20)
(5.21a)
p(x) = J.l e-I'X
1 - e-
pd
.
The mean interaction distance xis given by
x = J
O
d
xp(x)dx
x= 1 - e-pd(1Ui + 1)
p(l - e I'd) . (5.21b)
the values used, d = 0.25 mm and p = 15 cm-
I
, the
IS 0.1 a?proxImately equal to d/2. ,Por thicker screens used for
maxImum senSItIVIty, WIth reduced resolution, the front screen de th d
be half of the back screen depth d
2
p j can
Critical-Angle Considerations 69
probability density function of the photon interaction po' t
derived, is given by In, as previously
(5.16)
(5.15)
1 -L 1
2np - J.l I 2np + J.l 2np
(5.25)
I
(5.24)
which is valid as long as (2np)2 J.l2. As before, this is appropriate at all but
the lower spatial frequencies. Using these approximations we again define the
cutoff frequency Pk as that frequency at which Ii = k as given by
- II 2
H(h) = k:::::: .l:::-.e-I'dl __
1J 2npk
resulting in a cutoff frequency
P
.--.f!:-2e-l'dl
k - 2nk1J .
This indicates that this configuration, for a given efficiency, has an improvem
given by 2e-l'd
l
. It is erroneous to assume that a reduction of d
l
to zero
maximize the resolution since this negates the approximation of Ii(p) at big
frequencies and simply returns us to the original single-layer configurati
The optimum condition, as would be expected, is approximately at dj =
Under these conditions we have
J.l --
Pk - 2nk1J2,-/1 - 1J
with a resultant improvement over the single screen of 2,-/1 - 1J. The dou
screen improvement factor at high spatial frequencies for commercial ser
where 1J :::::: 0.3 is about 1.7. This substantial factor can be used to pro
improved frequency response at a given efficiency, improved efficiency
given resolution, or any intermediate combination.
The optimum division of the two screens between d
l
and d
2
for a giv
is a somewhat complex subject. A simplified approach is to place the div'
at the mean stopping distance of the photons, thus ensuring that the s
lations in both screens will be as close as possible to the film emulsions
CRITICAL-ANGLE CONSIDERATIONS
Afurther refinement of the frequency response of the h h
f th bl f h .. P osp or screen makes
useo/ e pro em 0 t e angle. Assuming perfect contact between the
ph()sphor and t?e film, the maXImum or critical angle ec at which Ii ht from
tJJ.epb.psphor WIll enter the film is given by g
LJ nf
sm U c = n (5.22)
p
refractive index of the film emulsion and . th. f .
. where np > n
f
. If we R th;
scmtI atlOn pomt at which light enters the film emulsion, we have
nf R
np = ,-/R2 + x2
R= x
'-/(n
p
/n
f
)2 - 1 (5.23)
hthe of the light as shown in Pig. 5.1, the
hm:ted to. a . CIrcle of radius R. Thus the original point
system IS multIplIed by eirc (r/ R) as given by
her) = K x circ (r'-/ii2=!)
(r
2
+ X
2
)3/2 x
the cire function is defined in Table 2.2. The Fourier
CIrc function is given by
g: {eirc r'-/f22=l} = x J (-.J.llXP )
x p,-/n
2
- 1 1 Vn2 - I .
Recorder Resolution Considerations
70
. the convolution of the originally derived
The total frequency response IS. . b
d
that due to the circ function as given y
ponse an __-)
xl (2nxp/Jn
2
- 1 .
()
K - 2r.xp * :::.::....!..C
1
HI P = e Jn
2
- Ip
'd tion increases the resolution but decrea
1 h itical angle conSI era (:t"
In genera, t e cr - . h h t s per x-ray photon. The ellect on noise
the number of captured llg t P 0 on
considered in Chapter 6.
ENERGY SPECTRUM CONSIDERATIONS
h b done under the assumption of a mo
The analysis of the recorder Wa.sh ethen usual broad energy spectrum gener
. spectrum It e
chromatic energy . d (8) of the scintillating phosphor shoul
by x-ray tubes, energy depen ence,ul t analysis. Thus a more accurate ex
taken into account for a more compoe ese in the svstem of Fig. 5.1 is give
sion for the averaged frequency respon -'
- fd Ii;' _
2rrxp
e-
p
(i;lx ,u(8) S'(8)d8dx (
H
1
(p) = K e 1 _ e-p(i;)d
o i;, .
. d 5'(8) is the enercry spectrum leaVI
. alizmcr constant an b f
where K IS a norm b F a parallel x-ray geometry, or or r
body and entering the recorder' .or f the rays can be ignored, this spe
close to the axis where the ob IqUlty 0
is given by [ r l
5'(8) = 5(8) exp -. ,ul(8, z)dz J
d is the attenuation coefficient
where 5(8) is the source spectrum an ,ul
body as a function of energy and depth.
ALTERNATIVE APPROACHES
TO RECORDER SYSTEMS
. . . h hors the light photons were
In recorders using scmtillatmg. P 'lol
sP
. 'This gave rise to the fu
. h oint of scmtI atlon. .
isotroplcally at t e p. d ffi' C I Three configuratIOnS ca
trade-off between resolution an e Clen).
to avoid this . can be structured in the form
First the scmtlllatmg phosphor h't If and as with co
, . t .ai is the phosp or 1 se , . d
fibers where the mner rna en d . h I yer of lower-refractive-lil e
fiber optics, the phosphor is cla v-: It a a in one of the fibers, t
hoton gives up Its energy Th
When an x-ray P lete internal reflection. us
light is trapped within the fiber by t the film The resolu
bounces back and forth untIl It arnves a .
Alternative Approaches to Recorder Systems 71
determined by the of. the fibers. the fiber structure
h isotropic spreadmg of the lIght photons, WIth Its resultant loss of resolutIOn.
eincrease the collection efficiency, the fiber bundles are merely made longer,
a resultant loss in resolution. The photographic film can be placed on
WI e side of the fiber bundle with an optically reflecting surface on the other
to ensure that most of the light photons eventually reach the film. Initial
51 perim
ents
have been made with structures of this type, although thus far the
construction difficulties have prevented its commercial availability.
P The second alternative approach uses a high-pressure gas chamber, usually
enon, as the detector. The x-ray photon gives up its energy in an ionization
xrocess that creates electron-hole pairs. A strong electric field is placed across
ihe gas chamber so that the charged particles, once generated, will follow field
linesa.l1d not disperse. The field is thus similar to the fiber structure, which
ionized particles to follow straight lines. Thus a relatively thick gas
be used to ensure that most of the x-ray photons interact and
. ovide high collection efficiency. The resultant charged particles follow
it field and are deposited on a dielectric sheet. The charge pattern on
represents the latent image. If developed with a toner, as is done in
; the desired image results. One of the practical difficulties is the
;the dielectric sheet from the high-pressure gas chamber.
alternative approach is a scanning system where the information
in time rather than in space. A narrow pencil x-ray beam scans
he detector is made very thick so as to have a very high capture
e detected signal is used to intensity modulate a synchronously
of a cathode ray tube to create the desired image. The important
.of this system is that the geometry of the detector does not affect
solution. Thus a thick, high-efficiency scintillating crystal, such
de, can be used followed by a photomultiplier to develop the
. The noise considerations of configurations of this type are
hapter 6. In any case the resolution is governed solely by the
e scanning beam and is independent of the detection process.
s of this type have been impractical since they use a very small
x-rays emitted from the source. This causes a long .exposure
e associated distortions due to respiratory and cardiovascular
ears line detector arrays have been used which have been
mputerized tomography systems. Here the x-rays are initi'il.lly
planar or fan-shaped beam. This planar beam is
y onto the line detector array. The resolution normal to the
.ed solely by the beam thickness. The resolution along the
bythe array of detectors. Each detector is usually shielded to
products, be they light or charged particles, from entering
ements. Thus, as with optical fibers, the resolution along the
ed by the size of the detectors. Detector arrays with O.5-mm
where
OVERALL SYSTEM RESPONSE
VIRTUAL
...... IMAGE
/
/'
t
MIRROR
FIG. P5.3
I, .... --d--+-II
t
EMULSION
Overall System Response 73
and Hbac'k are the individual normalized responses. Find the
cutoff Pk of the overall transparency using the appropriate
By what factor does this cutoff frequency differ from the
gle emulsion on the front side where the x-rays impinge?
to capture the escaping half of the light photons which are
st, a mirror of reflectivity R is placed in back of the screen as
P5.3. For this case calculate the following:
PROBLEMS
5.1 Using the appropriate approximations find the ratio of the high-frequency
cutoff.h for the case shown in Fig. 5.1, with the film in front where the
x-ray photons impinge, to that of a screen with the same geometry having
the film in the back.
5.2 Afilm emulsion is placed on both sides of a phosphor screen of thickness
d. 'The resultant transparencies are combined such as to provide an overall
transparency having a small-signal normalized frequency response
jj _ Hrront + Hbac'k
overall - 2
The maximization of this function determines the depth z which provides
the best With a source, and a broad
S(P), the optimum Will occur at small values of z SInce !JCp) will be
the dominant. factor. the OppOSIte for a larg.e source haVIng a narrow
S(P), the Will oc.cur where the plane IS the re.corder: with
zapproaching d. The opt.lmum depth plane, In general, IS determIned by dif-
ferentiating HoCpfM, z) WIth respect to the depth z.
Recorder Resolution Considerations
Ho(ft) =
This transfer function, in terms of object spatial frequencies pfM, ena
study the conditions for maximizing the response to of t
For example, placing the response in terms of depth z, we obtam
H0 (ft) = s[(1 - PJH( p) .
The overall response of a projection imaging system, including the source"
recorder considerations, is given bye
I (x y) = Kt(Xd Yd) ** ** herd)
d d, d M'M m
2
m
where for convenience, the source response, like that of the recorder, is ass
to be 'radially symmetric. In this case the intensity f d represents the reco
light photons. In the frequency domain this becomes
flu, v) = KM2T(Mu, Mv)Ho(u, v)
where Ho(u, v), the overall transfer function between the
object and the image, is given by
Ho(u, v) = Ho(p) = S(mp)H(p).
This transfer function is based on spatial frequencies at the r
representing spatial frequencies of the magnified object t(Xd/M, Yd!M
often desirable to evaluate the ability of the system to resolve spec!fic
frequencies of the object itself. The output spectrum in terms of. object
frequencies is given by
= K
M
2
T
(u,v)H
o
(ft)
detector elements have been constructed. Although the resolution is poc
than film screen systems, these alfays exhibit high efficiency and have electri
outputs that can be coupled to digital systems. The be
and line array are scanned, relative to the subject, to create a two-dlmensio
image. . .
In systems using discrete detector arrays wIth the reso
tion is governed by the detector size itself. Here the l.n resolution
based on fabrication considerations. A more fundamental hmltatlOn on detec
size is the signal-to-noise ratio, which is considered in Chapter 6.
72
6
Noise Considerations in
Radiography and Fluoroscopy
-----n--of-----d2 -.....
FILM
74 Recorder Resolution Considerations
(a) H(p, x), the normalized frequency response for a photon stop'
distance x. PIng
(b) H(p), the average frequency response.
(c) Pk> the high-frequency cutoff using the appropriate approximation
[Hint: For the reflected light use the virtual image position of the s .s..
.
latIOn, as shown.]
5.4 A more complete model of a film-screen system includes the effect
light from each screen reaching the opposite emulsion as shown in p.o
Assuming that the film is transparent to light and x-rays,
H1(p) and H
2
(p), the normaliz!d frequency responses of each emulsion
the overall response Ho(p) that it is the average ofth
mdlVldual responses. By what factor does Ho(p) differ from the previo
case, where the film thickness is ignored?
FIG. P5.4
5.5 An x-ray imaging system has a uniform circular source of radius r
l
the impulse response of the recorder is a uniform circle of radius "2. An
opaque planar object at plane z has two pinholes separated by a distance s.
(a) What is the overall impulse response of the system to a planar object
at depth z?
(b) What is the minimum distance s at which the images of the holes are
separable, that is, the resultant responses do not overlap?
(c) At what ratio of rdrz is this minimum distance independent of the
depth z?
5.6 An extended source x-ray system is used with the Source parallel to the
recorder plane and having a distribution s(r) = e- ar'. The recording plane,
a distance d from the source, has an impulse response her) = e- br'. Neg-
lecting all obliquity factors, at what distance Zo from the source should a
transparency be placed so as to maximize the relative response at a spatial
frequency Po in the transparency? Discuss the optimum Zo where a: b
and b a.
The ability to visualize a structure in a no.ise-free environment depends, among
other factors, on the local contrast C, whIch we define as
C = /).! (6.1)
1
where i is the average background intensity and /).[ is the intensity
in the region of interest. Contrast, however, is not a fundamental hmIt .on
visualization since it can be artificially enhanced by, for example,
part of the background or raising the intensity pattern some NOIse,
however, represents a fundamental limitation on the. :0
tures. The signal-to-noise ratio, a basic measure of VIsualIzatIOn., IS .
by the ratio of the desired intensity variations to .the random mtensIty vana-
tions, which are governed by the statistical propertIes of the system. In general,
our signal-to-noise ratio will be defined as
SNR = /).1 = ci (6.2)
(11 (11
77
(6.7)
(6.9)
(6.8)
Noise Considerations in Radiography and Fluoroscopy
i1N , 1M
SNR = ;JN = Cv N
a Poisson process of rate pN
o
. We have shown that the
Vntlot()nS from the object continue to be Poisson distributed with the
by the attenuation of the object p = exp (- f fJ.dz). The photons
m the object have a mean value N given by
al is i1N, the variation in the number of photons per element
ructure of interest, and the noise is ,.)N, the standard deviation
of photons per element. As is seen for a given subject contrast
-noise ratio is proportional to the square root of the number/ of
the signal-to-noise ratio can be made arbitrarily high except for
imitation. We will briefly consider the nature of the radiation
process of mean N the variance is N and the standard deviation
zen, 1960). We can thus calculate the signal-to-noise ratio of a
ing a contrast of C as given by
t of radiation [Johns and Cunningham, 1974; Sprawls, 1977] is
lex and will be treated here in a relatively simplified fashion.
summation is equal to e
qN
Substituting this into the formula for
obtain
of the probabilities of the various combinations transmitting k photons as
given by
Q(k) =P(k)( Z)pk + P(k + l)(k t l)pkq + ... + P(k + n)(k t n)pkq"
(6.4)
where each term represents a combination of a photon source producing k +
xphotons,P(k +x), and a binomial transmittance (k t X)pkqX, which combine
to transmit k photons. Since P is a Poisson process, we have
P(k + n)(k + n)pkqn = (k + n)! pk
q
"
k (k + n)! k! n!
_ e-N'(pNo)k (qN
o
)"
- k! -n-!-' (6.5)
Substituting this result in the general equation for Q(k), we obtain
Q(k) = t (6.6)
k. n=O n.
Noise Considerations in Radiography and Fluoroscopy
SNR = ;JNhv + 4KT
where N is the number of photons per time element, is the signal ener
per element, and JNhv is the standard deviation or nOIse energy per
In most communications systems 4KT hv, so that, for. any reason.able Si
strength, the additive thermal noise term in the denomInator F
example at v = 106 Hz, 4KT/hv = 2.5 x 10
7
Even at v = 10
2
GHz, thIS ra
is 2.5 x'102 At the nominal x-ray frequencies of v = 10
19
Hz, correspon
to A = 0.2 A, this ratio is 2.5 x 10-
6
Clearly, equation (6.3) reduces to
SNR given by fl. This means that in the x-ray region the to.tal re
sents a countable number of photons whose statistical uncertaInty is the
contribution to the noise.
The emission of photons from the x-ray source is a Poisson process [Pa
1960) whose probability density was given in equation (2.53) as
Pk = -k-'-.-
where P is the probability, in a given time interval, of emitting k photon
No is th; average number of photons emitted that interval.
In Chapter 3 we characterized the transmiSSion of photons throu
body as a binary process where photons either inte.racted and were r
from
the beam or did not interact and were transmitted to the recorde
, bT fa
represents a binomial process [Parzen, 1960) where the proba 1 P0
being transmitted is exp (- S j1dz) and the probability q it
1 - exp (- S j1dz). The cascading of a Poisson and bmomlal dis
results in a Poisson distribution as shown below. .
The probability of getting k photons through an object, Q(k), is
where (JJ is the standard deviation of the background intensity representing the
rms value of the intensity fluctuations. .
The noise properties of most communications systems Involve
noise only. The energy per photon, hv, in these spectra is rela.
tively small, so that copius amounts of a:e avaIlable for the
sicrnals being considered. Thus the uncertaInties lIe almost completely ill the
added by the measurement system rather those of the itself.
Quantum noise, which is the primary n?ise source In x-ray sys.tems, IS the noise
due to the quantization of the energy Into photons each haVIng an energy hv.
This quantum noise is Poisson distributed. .. .
We can appreciate the dominance of the quantum nOIse by considenng a
communications system having an additive noise power of 4KTB, where Kis
Boltzmann's constant, T is the absolute temperature, and B
The noise energy per time element 1/B is therefore 4KT. The ratiO of Signal-to-
noise energy per element can be structured as
Nhv
76
78 Noise Considerations in Radiography and Fluoroscopy
Our principal intent is to relate dose considerations to the incident photon
density and hence to the signal-to-noise ratio. Two quantities are of interest.
the ex'posure, or amount of radiation delivered to a point, and the absorbed
dose or the radiation energy absorbed in a region.
The unit of exposure is the roentgen (R), which is defined as producing
ionization of 2.58 X 10-
4
coulomb/kilogram in air. The absorbed dose unit is
the rad, which is defined as an expenditure of energy of 100 ergs/gram. In air
an exposure of 1 R corresponds to an absorbed dose of 0.87 rad. The number
of rads per roentgen varies with different materials and with the energy used as
shown in Fig. 6.1. As is seen, for soft tissue, there is approximately 1 rad per
Noise Considerations in Radiography and Fluoroscopy 79
SNR = C,Jf1>ARt. (6.11)
;:tl-to-noise ratio given above is that of the photons emerging from
thus would represent the performance of a detector which cap-
ese photons. For a recorder having a quantum efficiency 1] the
becomes C,J1]C1>ARt. Here we see the fundamental t:ade-
A dose R. For a typical chest x-ray the exposure is
the tran.smIsslon t through regions devoid of bone is about 0 05
effiCle?cy of the screen is about 0.25. For a linear
the. ratio is about 400C or about 16 db for
n a 10 %change in transmission. i
to-noIse :-atio been defined in terms of the incident radiatitn
us effectIvely In terms of the surface dose. The maximum dose
t the surface. Although this is an effective measure a
r dose are used. It is important to notehowe;er
e relates dIrectly the incident photon density and {hus would
constant for a gIven body material. Integral dose, or total
the number of roentgens incident on the subiect A I'S th . f
t . 2 (J' 'J, e area 0 a
en III cm - pdz) IS the transmission t through the body
densIty per roentgen. The signal-to-noise ratio in terms of
Iven as
__-L_ PHOTON
, 40 ENERGY
120 140 160 (key)
G. 6.2 The relationship of photon f1uence to exposure over the diagnos-
energy range.
3.0
2.5
2.0
X10'0 PHOTONS/cm
2
/ROENTGEN
160 140 120 100 80 60 40
MUSCLE
20
FIG. 6.1 The relationship of an absorbed dose to exposure in body tissues
over the diagnostic energy range.
_...........
/. .
. '"
I .
. "
.-------.
.-.-*-e-e-e-e---.---. .
o
2.0
1.0
4.0
3.0
RADS!
RENTGEN
roentgen throughout the diagnostic energy range. Thus with the excepti
bony regions, where the absorbed dose increases at lower energies, the r
roentgen become essentially equivalent for our purposes.
Figure 6.2 shows the relationship of photon density to the expo
roentgens. At low photon energies most of the photons interact, but the
imparted per interaction is small. At very high photon energies, very few
interact. These conflicting factors result in a peak at about 60 kev,
average over the diagnostic range being about = 2.5 X 10
10
photons!,
roentgen. We use this relationship to characterize the number of pho
pixel N in terms of the incoming radiation as given by
N = <PAR exp ( - Sf1,dz)
Recorder Statistics 81
the valence band to the conduction band. In its return to the valence band
energy is radiated in the form of visible light photons. The number of light
photons produced due to each captured x-ray photon is also a Poisson-distrib-
uted random variable. The total number of light photons produced is given by
'd . 'n Radiography and FluoroscOpY
80 Noise COnsl eratzons I
d
. -rads where a O"ram-rad is equivalent to
. expresse In gram, 0 .
absorbed energy, IS f d e used is the surface Integral exposure
A
th Jobal measure 0 os )
100 ergs. no er g . roentgens and the total surface area
which is the product of the exposure In
exposed.
M
Y = L: Xm
m=l
(6.12)
(6.14) a; = E(M)a; + airE2(X)
epresents the expected value and a
2
the variance, These equations
ely reasonable since the resultant mean would be expected to be the
the individual means. The variance a
2
is due to the uncertainty of
r of captured x-ray photons and the uncertainty in the number of
ons produced per x-ray photon. Since each x-ray photon produces
ns having a variance a;, the total uncertainty due to light photons
.iSimilarly, the variance in the number of captured x-ray photons air
an average gain of E(X), so that this resultant component of the
eighted by E2(X),
use these results to analyze the signal-to-noise ration of a radio-
n to determine whether the uncertainty in the emission of light
ences the overall performance. We let E(X), the mean value of
flight photons produced from each captured x-ray photon, be
the average gain. It should be emphasized that gl includes both
flight photons and their attenuation or transmission loss to the
yare used. The number of light photons captured is often limited
gle or transmission through a material. As with the previous
Poisson process is attenuated by one having a probability lof
he effective rate of the Poisson process is simply multiplied J:>y
product of the average number of light photons produced
multiplied by the probability of transmission to the point of
poisson distributed, g1 is also the variance of X. E(M), the
,of captured photons per picture element is, as before, l1N.
oisson distributed, its variance is l1N. Substituting these into
where Yis a random variable representing the total number of light photons
produced, M is a random variable representing the number of captured x-ray
pMt()ns, and X
m
is a random variable representing the gain or the number of
light photons produced per x-ray photon.
Our resultant SNR will be determined by the statistics of Y, the total
We will study to what extent this SNR has been reduced com-
that of equation (6.11). ":Ie do not to know the density
of Y [Feller, 1957] smce the SNR IS completely determmed by the
variance. Using probability theory, these are given by
E(Y) = E(M)E(X) (6.13)
RECORDER STATISTICS
s er icture element transmitted thro
Ideally, the number of photon P lt
P
t SNR performance. This, ho
1 d
t mines the resu an h
body complete y e er . (6 11) thus represents t e
only true with an ideal only a portion 1J of
, NR f iven radIatIOn. n , .' the
SIble S or a g dditional noise source anses m
P
hotons are captured. Also, an a .
h
t record the Image. d fi
of the captured P otons 0 f ently used in radiography an .
The recording process most requ elatively high atom
I
.' These screens use r
is the scintIllatIOn screen. h the x-ray photonS ar
1 . tungstate were .
materials, such as ca clUm. i this process an electron IS r
. '1 by the photoelectnc e ect. n
pnman Y
RESOLUTION CONSIDERATIONS OF THE SNR
. d (6 11) is clearly proportional to the contrast
The SNR in equatIOnS (6.9) an. .' aused by the structure of interest. It
. 1 h O"e in transmISSIOn c
or fractIOna c ano . t C is the recorded contrast. For larg
h
.zed that thIS contras
should be emp aSI bl . due to the finite source and the record
lesions C is unaffected by. the urlnn
g
ds the edges of the lesion and does n
. . thOs blurnng mere y roun d d
resolutIOn SInce 1 II t uctures however the recor e contr
. 1 alue For sma er s r " .
change ItS centra v h a specific definition must be aSSl
will definitely be affected. In t ese d image. Hopefully, this definition
t
o C relating to the shape of the recor e
, . r . b observers.
strongly relate to VIsua IzatIOn. Y ( ) can be blurred due to a finite sou
For example, a planar t hapter 4 of the form s(x/m, ylm)
'd' image as gIven In . bl 1
size, pro
vl
Ing an h't recorded contrast consIdera y at
t(x/ M, ylM). A can
It
to note that this system ha,
by this convolutIOn operatIO? h structure being visualized. A relatt
. dependIna on tel. 1
optimum source SIze . 0 d d contrast but with re atIVe y
.d th hIghest recor e ,
sm'lll source pro
vI
es .e . f a given photon density, the nu
h ce SIze Increases, or h
photons. As t e sour , eral the contrast can decrease due to t
of photons increases m gen " tronaly dependent on the depth
volution operation. ThIs, of course, IS s 0
tion of the lesion of interest.
and
Fluoroscopy 83
VI EWER
X-RAY
SOURCE --r-----_L
tag
e
does not significantly change the resultant signal-to-noise ratio. The final
s ocess, the film emulsion, is highly nonlinear and not subject to simple analysis.
a linearizing approximation, however, about 200 photons are required on
the average for each developed silver grain in x-ray film; thus gz :::::::: 1/200. The
g P
roduct of 2.5 does result in about a 20 %degradation of the performance
gj 2 --=-=
since,j1 + 2/5 :::::::: 1.2.
FLUORESCENT SCREEN
FIG. 6.3 An early fluoroscopy system.
In fluoroscopy [Ter-Pogossian, 1967; et al,. 1978] the image is
displayed in real time rather than being recorded on film. This allows moving
structures such as the heart to be visualized. It is also widely used for the inser-
tion of catheters to monitor their position. In the early forms of fluoroscopy
the radiologist directly observed the fluorescent screen as shown in Fig. 6.3.
FLUOROSCOPY
has a significant noise problem due to a deteriorated light collection
. e . Approximately 10-
5
of the light quanta produced at the screen
aci:].\y appear at the retina of the eye because of the small solid angle inter-
cete<ibythe pupil n
e
and the losses within the eye. The capture efficiency of
tlit'fe is given by
'"iIi . = Tene = TeA (6 23)
x.i . 71e 4n 4nrz . ;
is the light transmission to the retina, which is about 0.1, A is Jhe
pupil/area and r is the distance of the eye from the screen. For the dark-adapted
eye,Ais about 0.5 cm
z
, corresponding to an 8-mm pupil. The closest reasonable
vieWirfgdistance is about 20 em, resulting in maximum value of 1fe of about
of 10-
7
to 10-
8
are much more typical. For a typical screen the
product of the screen gain of 10
3
and the retinal transfer of 10-
5
is about lO-
z
(6.16)
where Y, as before, is the number of captured light photons and 2 m is a rando
variable representing the number of events, such as developed film grain
produced for each captured light photon. Using the previous relationship
we obtain
a; = E(Y)a; + a;EZ(2)
= gjgz71N + (71Ngj +
a
w
= gjgzJiiN /1 +1- + g1
'\j gj z
SNR = CE(W) = c,v'11R
aw ,v'1 + 11gj + 11gjgz
This representation can be generalized for q successive stages as
SNR = cJiiN .
J1 + gj)
If every product term, gjgZg3' ... , is appreciably greater than unity,
SNR will be dominated by the captured number of x-ray photons per P
In general, this is a desirable goal since it ensures that the minimum pa
dosage is used for a given image quality.
In the particular case of radiographic film-screen cassettes, g1, th.e g
the scintillating screen, is limited by energy conservation to the ratiO 0
wavelength of the emitted visible light, 5000 A, to that of the x-ray pho
0.25 A, which is 20,000. These screens have an energy efficiency of
about half of which escapes, resulting in a gj of about 0.5 x 10
3
ThiS
a; = 71 Ngj + 71 Ngi
The resultant signal-to-noise ratio is given by
SNR = CE(Y) = . (6.17)
ay ,v'1 + 1/gj
If an appreciable number of light photons g j are collected for each captured
x-ray photon, the resultant statistics are essentially determined by the numbe
of captured photons 71 N.
In most processes, such as the radiographic screen-film combination,
least one additional gain stage is involved having a similar statistical model. In.
the second stage each light photon generates a Poisson-distributed array 0
events whose mean is gz. We can characterize the sum of the recorded eventS
in this process by W, where
equations (6.13) and (6.14), we obtain
E( Y) = 71 Ngj
82 Noise Considerations in Radiography and Fluoroscopy
OBSERVED
FLUOROSCOPIC
SCREEN
........... OBSERVED
INTENSIFIER
OUTPUT
FLUOROSCOPIC SCREEN
LIGHT PHOTONS
EMITTED BY
FLUOROSCOPIC SCREEN
\
\
\
\
\
\
\
\
\
\
STAGE
\
\
\
LIGHT PHOTONS
EMITTED BY
OUTPUT SCREEN
LOSS FROM
OUTPUT SCREEN
TO RETINA
FIG. 6.5 Quantum values at different levels.
'"
ONE X-RAY PHOTON
ABSORBED BY INPUT
SCREEN
ELECTRONS EMITTED
BY PHOTO-EMITTER
IGHT PHOTONS
EMITTED BY
INPUT SCREEN
/e
/
/
/
/
/
/
/
/
/
Image Intensifier 85
,gIg, product of the phosphor screen and h .. .
W
hich will leave the incident sI'gnal to . p ?toemltter gams of about 100
- -nOIse ratIo e . II '
emitted electrons are focused by the v I ssentla y undisturbed. The
d
anous e ectrostatic Ie d
to repro uce the incident image with good fidelit nses an magnets
sec;en. These electrons are accelerated to a yon an output phosphor
pro?uce about 10' visible photons per elect n of about 25 kev so as to
can either be observed directly or through . e resultant mtensIfied image
the light loss to the retina of film If observed
of about 1, so that the signal to' . .results In an overall gain
h - -nOIse ratIO IS slightly red d P'
o. t e quantum values at different 1 I' uce. 19ure
eve s In the system. In a dose-
INPUT
FLUORESCENT
SCREEN
PHOTOCATHODE
\
\
\
ELECTROSTATIC
LENS
ELECTRONS
FIG. 6.4 Diagram of an x-ray image intensifier tube and a photOgra
a typical intensifier. (Courtesy of the Siemens AG-Bereich Medizi
Technik.)
\
\
\
\
/
!
I
!
\
--
Modern fluoroscopic systems [Ter-Pogossian, 1967; McLean and Schag
1979) solve these problems through the use of the image intensifier, shown
Fig. 6.4. In these devices the light from the phosphor scintillator is ampli
before being utilized by the eye. A photoemissive material is placed against
scintillating phosphor. These materials have quantum efficiencies of about 1
so that one electron is emitted for about every 10 light photons. This pro
OUTPUT
__--- FLUORESCENT-
SCREEN
1-4----- ANODE
IMAGE INTENSIFIER
at best. This reduces the signal-to-noise ratio by 10. Stated in other words, the
photon flux, or radiation, required to provide an image having a signal-to_
noise ratio comparable to that recorded on film would require a lOO-fold
increase. As fluoroscopy was normally practiced, however, the radiologist con-
tented himself with a poorer signal-to-noise ratio. He usually employed dark-
adapted vision, which increased pupil size and thus maximized the collection
efficiencY ofJigbt photons.. In the dark-adapted state, however, the visual accuity
IS conSIderably reduced smce the denSIty of rods m the retma IS considerably
less than that of the cones, which dominate vision at normal light levels.
Thus early fluoroscopy was characterized by poor statistics and poor
visual performance of the observer. It is the function of the image intensifier
our next topic, to both improve the statistics, in terms of available independen
events per pixel, and to increase the brightness sufficiently so as to provi
normal visual acuity.
84 Noise Considerations in RadiographY and Fluoroscopy
(6.25)
/
ELECTRON
OUTPUT IN TV CAMERA
SNR = Cl1
N
,y'N; + l1N
OUTPUT LIGHT
PHOTONS
/
1.0 ABSORBED
X-RAY PHOTON
Additive Noise 87
EMITIED LIGHT
\
\
EMITIED
ELECTRONS
FIG. 6.6 Quantum values in a system employing a TV camera.
component in the output signal of the television
. e ynamic range and can represent the limiting noise com on nt
high.photon transmission. Since this additive noise is rnt
nOIse due to the x-ray h t h' .
p 0 ons, t e signal-to-nOIse ratio can be
STAGE
10
4
IMAGE
QUANTA/STAGE
(10
3
)(10-
1
)(0.04 x 10
3
)(0.1) = 4 x 10
2
where g1 is the gain of the scintillating phosphor in the image intensifi
light photons per x-ray photon, g2 the efficiency of the photocathode in el
per light photon, g3 the gain of the output phosphor of the image inte
with the loss in the optics in light photons per electron, and g4 the q ..
efficiency of the television camera photocathode in electrons per light p
The result of 4 x 10
2
ensures that the signal-to-noise ratio of the s
essentially determined by the number of quanta emerging from the b
the quantum efficiency in capturing these photons. Fluoroscopy bee
efficient process. The resolution through the many cascaded imaging st
is reduced over film radiography, although adequate for most studies.
tion to resolution, the dynamic range is reduced considerably compared
tographic film because of the television camera and the glare of th
intensifier. Figure 6.6 illustrates the quantum values at different level
system employing a television camera.
where M is the magnification, f the ratio of image distance to lens diamet
and T the light transmission of the optics. This expression is essentially d
mined by the solid angle intercepted by the optical system. Systems that eo
the image intensifier output to a television camera have f numbers of a
1.0, a magnification of about 1.0, and a transmission T of about 0.6, resul
in an 110 of about 0.04. The quantum efficiency of the photoemitter of
television camera is about 10%. This represents the lowest point in the quan
amplification chain, since beyond this point there are a variety of amplifica
mechanisms. The glg2g3g4 at this lowest point is given by
86 Noise Considerations in Radiography and Fluoroscopy
efficient system the minimum quantum levels, representing the product of th
average gains, remain well above that of the absorbed x-ray photons. e
The demagnification of the input image to a smaller output image in Pi
6.4 does not affect the SNR, since it is based on events per picture
!his does, the brightness of the
Image, whIch does affect the vIsual accUlty of the observer. Increased brightnes
can ensure that the visual accuity will be dominated by the
cones rather than the rods.
With many image intensifiers the small image size precludes direct vieWing
so that additional devices are required. If optics are used to magnify the image
of the intensifier will be lost, thus pro.viding either a nois;
Image or addItIOnal exposure. ThIs problem can be allevIated by coupling the
output of the image intensifier to a television camera tube. The optical syste
coupling the image intensifier output to the television camera is usually a len
The photon collection efficiency of a lens 110 is given by
88 Noise Considerations in Radiography and Fluoroscopy
SNR of the Line Integral 89
(6.34)
(6.35)
(6.37)
SNR = c In
= c (J ,udz),fiiN.
E[g(X)] = g(N) + g/J(N)(J2 +
2 . . . (6.32)
as before, is the variance of the number of h .
valuating the second derivative of the log p btO?S per element gIven
,we 0 tam
E[g(X)] = In (NN) .+.- J... -L- (6 33
. 2N I )
he number of counts per element exceeds 106 .
ected value of the line intearal can simpl b most CIrcumstances,
,our of by In (No/N), the
of the line integral, (J;, we use eq (6.31) to evaluate
= g2(N) + [g'(N)2 + g(N)g"(N)](J2 + ....
the relatIonship from equation (2.49) yields
(J; = E[g2(X)] - E2[g(X)]
[g'(N)J2(J2 - + ...
+ terms of degree and higher
.......,1
-N'
al to' ' .
. - -nOIse ratIo IS given, from equation (6.29), by
SNR = cln
detection efficiency,
in a Taylor series about N as given by
g(X) = In No - In X
g(N) + g'(N)(X - N) + g/J(N)(X -; N)2 + ... + g(n\N)(X _ N)n
. n!
h X
d . (6.30)
were. IS a ran om varIable representing the b f
per element and N is its mean value. num er 0 transmitted photons
Using the general relationship
E[g(X)] = Jg(X)p(X)dx (6.31)
use the series to evaluate the various statist' I
(6 3
1) d' t' . lca averages of g. Inserting
IV""VJ"nv an m egratmg, we obtam
g = In No - In N = f fldz
We have thus far dealt solely with the signal and noise considerations of
transmitted photon intensity. This transmitted intensity is a nonlinear fune
of the line integral of the attenuation coefficient N = No exp (- f fldz). In
recent x-ray imaging systems employing digital processing and electroni
plays, the line integral g itself is calculated and displayed through a log oper
as given by
where the detector efficiency has been assumed unity.
In this case our SNR is defined as
SNR OF THE LINE INTEGRAL
where c is the fractional change in the average value of SJ.!dz cause
region of interest and (Jg is the standard deviation.
To evaluate the mean and standard deviation of the line integr
make use of the fact that g is a function of the random variable X [
1965], which is known to be Poisson distributed. We expand the f
ell
SNR = T'i/ 1 + 1/k211
N
'
In cases where k211N 1, corresponding to a high photon count and/or
relatively high additive noise fraction, the SNR reduces to C/k, the ratio of the
fractional signal component to the additive noise. In this case further increases
in radiation No will not improve the performance, since it is being dominat
by the additive noise of the system. In the other extreme, where k211N
the additive noise is negligible and the SNR returns to the dose-depende
case of Cjl1N.
N a = kl1N.
The SNR in equation (6.25) can be rewritten as
where N is the .average or of tran.srr:itted photons per
element, N = No exp (- f fldz), C IS the fractIonal vanatIOn of the region
under study, and N; is the variance of the additive noise component. Thus
N
a
, the standard deviation of the additive noise, is being expressed, for COn.
venience, as a number of photons per element.
In general, the additive noise Na will occupy a fraction k of the dynamic
range or the average value of the signal. This is conveniently expressed as
90
Noise Considerations in Radiography and Fluoroscopy Scatter Analysis 91
Two appropriate assumptions have been made in this derivati?n, which
should be pointed out for completeness. First, to use .the Taylor senes. expan-
sion we must assume that the probability of collectmg zero counts IS zero;
oth:rwise the log function blows up. This is clearly a reasonable assumption,
given an 'average of> 10
6
Second, the of number of i.ncident
photons per element, No, has been ignored. ThIs IS agam SInce
. enerally a multiplicative factor of > 10
2
greater than N, so that ItS relatIve
IS g . . d f
statistical variation can be neglected. Also, monItorIng .etectors are 0 ten used
to measure No, thus minimizing the effect of its .
The expression for c, the fractional change In the lIne Integral, can be
structured in terms of the contrast C, as given by
N N
- SI'dt N - (I + <l SI'dt J
C _ - oe - oe :::::: c /ldz.
- N - Noe- SI'dt
Therefore, logarithmic processing of the image,. to provide a display of theyne
integral of the attenuation coefficient, results In the same SNR as the dIre
intensity presentation as given by
cg ci -
SNR = - = - = ejYfN.
(lg (l1
SCATTER
The most significant additive noise component !s that of [Te:-Pogoss'
1967]. The attenuation mechanism that domInates :adIOgraphy IS Com
scattering. Depending on the energy used and the atomIC of the mate
the attenuation mechanism is divided between absorptIOn scatter.
soft-tissue regions, which occupy most of the body, scattenng IS the doml
mechanism. Unfortunately, many of the scattered photons reach the sc
representing additive noise. .
Scatter, as an additive n?ise c?mponent, ha.s two effect;
of contrast and an increase In nOIse. The first IS a determInIstIc phen
where the scatter produces an added intensity Is which adds to the
intensity It' The resultant contrast C, due to a transmission contrast CIS gI
C CIt C
, = It + Is = 1 + Is/It
where (1 -L I /I)-I is the contrast reduction factor. This reduction is
, st.. h I b comparable t
evident in regions of low transmISSIon were t ecomes d
addition to its deterministic effect, the scattered photons produce an a
noise component in the counting statistics. If N represents tran
photons per picture element and N
s
the scattered photons per pIcture
. . d b f ndependence. The
their vanances add In the etector ecause 0 1
signal.to-noise ratio becomes
SNR = CYfN c,.,/ijN
+ l1Ns + NsjN (6.41)
Scatter causes a serious deterioration of performance if nothing is done to
reduce it. This scatter reduction process must somehow distinguish between the
scattered and transmitted photons. One mechanism, as described in Chapter 3,
is to make use of the photon energy loss that takes place in the process of
Compton scattering. Unfortunately this can only be used for the case of mono-
epergetic sources, since, with broadband sources, the transmitted and scattered
photons have overlapping energy spectra. This procedure of scatter elimination
tlirough spectral analysis is thus only used in nuclear medicine, where the
iS9tope sources are monoenergetic and of sufficiently high initial energy to
ca.usea significant energy loss. The other mechanism that distinguishes scatter
of direction. Transmitted photons all appear to come from the source.
SAAttered photons can thus be minimized by collimating structures that are
. -at the source. These structures, known as grids, absorb many of the
d photons because they arrive at angles other than that determined by
ition of the source. They will be analyzed in some detail following the
ment of a model to analyze the amount of scatter.
in Fig. 6.7, the area under study is a cylinder of length Land
which is assumed to be homogeneous with an attenuation coefficient
'ncident x-rays are assumed to be parallel having a photon intensity
nsjcm
2
Thus the intensity reaching the incremental section at plane
oton density given by n(z). The incremental density of scattered
Ilerated in this section is given by
dnsCz) = n(z)J1.
s
dz = noe-
JlZ
J1.
s
dz (6.42)
the Compton scattering portion of the attenuation coefficient.
consider what fraction of the scattered photons will reach the
he scatter were perfectly isotropic, half of the scattered photons
the direction of the detector and the other half toward the source.
ostic range is a slight from isotropy, wl1-ere
k In the forward dIrectIOn [Klein and Nishilma,
oXImated;by
k = 0.52 + 0.07ffci
kev
). (6.43)
ical average photon energy of 40 kev, k = 0.55.
ume that the collection angle at each point in the detector plane
FIG. 6.7 Model for scatter analysis.
(6.47)
(6.49)
(6.48)
(6.52)
(6.54)
Scatter Analysis 93
F(r) = I - e-
P
.'. (6.51)
to the probability that the photon will reach a distance
by the conditional distribution function
detector plane in Fig. 6.7, we have
dO. = dA cos (t _ rdrd(J(L - z)
r
2
+(L - Z)2 - [r
2
+(L - Z)2]3/2
here dA cos (t is the incremental area element in the plane of the section with
; being the angle to the axis of the cylinder. Integrating, we obtain
Q(z) = (L - z) lR1
2
" [r 2 + (frc:! Z)2]3/2
= 2n[1 - L - z J.
,.jR2 +(L - Z)2
The integral in equation (6.46) is then given by
1
LQ(z)dz = L + R - ,.jL2 + R2
o 2n
with the scattered photon density at the detector given by
n
s
= Bnoke-pLf.l.(L + R - ,.jL2 + R2). (6.50)
The multiplicative buildup factor B accounts for those scattered photons
not collected after the first scattering. If they scatter again, the fraction col-
lected can be structured exactly as the previous analysis-hence the multipli-
cative nature of the factor. A simplified estimate of B would therefore be the
estimated number of scattering events as a photon travels to the screen. We
first calculate the mean distance in the z direction to a scattering interaction of
an entering photon. The estimate of B is then the total depth L divided by this
mean distance.
Assume that a photon is moving toward the detector at an angle (t to the
cylindrical axis. The distribution function of a scattering interaction along its
or probability that the photon will reach a distance r before interacting
F(z I(t) = 1 - exp (-f.ls
Z
) .
cos (t
,smiffelreI11tiating provides the probability density function,
p(z I(t) = -!:!:.L exp (-f.ls
Z
') .
cos (t cos (t
the cOlhditiolnal mean in the z direction, given the angle (t, becomes
E(z I(t) = Szp(z I(t)dz
= ~ 1"" z exp (- f.lsZ)dz = cos (t.
cos (t 0 cos (t f.ls
X-RAY DETECTOR
------
--'
'"
"
T
I
z
\ /'
I //-
-l-" dz
no PHOTONS/cm
2
t t t t
F(z) = ke-p(L-Z) Q(z)
2n
where k is the fraction scattered forward, e-p(L-z) is the fraction transmi
through the material, and Q(z)j2n is the fraction subtended by the solid an
of the detector. If these scattered photons were the only ones produced, negl
ing secondary scattering processes, the total scattered photon density at
detector, n ~ , would be given by
n: = Sn(z)flsF(z)dz.
However, this analysis has neglected a variety of secondary process
These include the scattered photons within the solid angle which have ex
rienced an additional scattering event but remain within the solid angle.
analysis has also neglected photons which were scattered outside the solid an
which, as a result of a subsequent scattering event, are scattered back into
solid angle. We can summarize this process by a multiplicative buildup fa
B providing a total scattering density at the detector ns given by
n
s
= S: Bn(z)f.l.F(z)dz
B k
-pL (L Q(z) d
= no e fls Jo 21t z.
We first evaluate Q(z) and its integral for the cylindrical geometry of
6.7 and then evaluate the buildup factor B. Using r, () polar coordinates
is constant at Q(z), the fraction of scattered photons F(z) that reach the detect
is given by
92 Noise Considerations in Radiography and Fluoroscopy
SCATTER REDUCTION THROUGH SEPARATION
Scatter Retblction Tltrough Separation 95
."
.\.
\.
\
"\ 00'
0.0 I-.. -O-_..L-'--'-----'__...... \ __ .......l__
0.Q1 0.1 0.20.30.5 1.0 2 3 5 10 100
0.2
0.4
0.8
0.6
1.0
0.99
parable to the object size, the scatter collection fraction is about 0.3, leaving
the resultant scatter intensity larger than the transmitted intensity for the object
previously considered. For relatively large separations, however, this reduction
factor becomes appreciable. However, at these larger separations, we experieI}ce
resolution the finite source size as studied in Chapter 4. Wfth
separatIon the SNR IS gIven by :
SNR = C"'/1/noe Jd-A . (6.62)
.../1 + 'IIQ(s)/2n
- Equation (6.62) was derived with the assumption of a parallel beam.
Although this assumption gives reasonably accurate results for the previous
FIG. 6.8 Scatter collection fraction vs. separation/radius.
by some distance s, the number of received scattered photons will be
object d while those transmitted will be unaffected. The solid collecting angle
the detector and the emitted photons from the object has the
:eform as derived for Q(z) in equation (6.48) and is given by
Q(s) = 2n(1 - .../RZ
s
+ sz)' (6.61)
Th efore the fraction of scattered photons collected by the screen is given by
Q(:{;21C. Aplot of th.is collecti.on versus s/R, the ratio of the
to the object radius, IS shown In FIg. 6.8. As can be seen, for separatIons com-
S'2(s}
2i
Noise Considerations in Radiography and Fluoroscopy
SNR = C.../'flno
e
J1LA
.../1 + ns/noe J1L
where A is again the area of a picture element. It is convenient to struct
the SNR as
SNR = C.../'flno
e
J1LA
.../1 + If
where If is the ratio of scattered-to-transmitted photon density at the detecto
given by
III = = Il;L2k(L + R - .../L2 + R2).
'I' noe J1L
Using typical values of a body section where L = 20 cm, R = 10
with Ils = 0.2 cm-
1
, and k = 0.5, we have a If of approximately 6.1..
This degree of scatter is clearly unacceptable. It would reqUIre
responding increase of greater than 6 to 1 in radiation dose to resto:e
original SNR. In addition, the contrast, especially in areas of 10"':'
is seriously deteriorated. It is therefore obvious that scatter reductIOn IS requ
Because of the many multiple scatter processes, the scattered e
from the volume in Fig. 6.7 can be assumed to be approximately ISotr
covering a solid angle of 2n steradians. If the detector is separated fro
= ("/2 cos a sin rJ. da
Jo Ils
1
,,/2 1 1
= -sin2adrJ.=2
o 21ls Ils
The average number of interactions along the entire length L is given by
L
B = E(z) = 21ls
L
with the scattered photon density at the detector given by
n
s
= noe-J1LIl;L2k(L + R - .../L2 + R2).
We can use this result to find the signal-to-noise ratio
given by
The mean or average distance to an interaction in the z direction is obtain
by integrating the conditional expectation multiplied .the probability dens
of a. Since the scatter is isotropic, it has equal probabIlIty of appearing in an
two-dimensional angular interval. This corresponds to pea) = sin a, where
varies from 0 to n/2. The resultant mean distance is
E(z) = SE(z Ia)p(rJ.)da
94
97 Scatter Reduction Using Grid Structures
where niO) is assumed uniformly distributed over 1C radians and T(O) is even.
The function of the grid structure is thus to provide as low as possible an
integrated T(O) without unduly attenuating the desired transmitted photons. If
1/ is the grid efficiency for transmitted photons, the improvement in the ratio
of transmitted to scattered photons at the detector is given by 11t/RS' T(O) and
R
s
are evaluated using Fig. 6. I O.
Noise Considerations in Radiography and FluoroscoPY
I
" 't does give an optimistic result in the case of separation. In the case
ana )SIS. 1 . h '.
f d
' t b am the detected photon density due to t e transmItted sIgnal
o a Ivergen e , . .
photons will decrease due to the The scatter,
is unaffected by the beam divergence SInce It rep:esents an IsotropIC
source. Thus the SNR is reduced from that shown In equatIOn (6.62).
96
J I
FIG. 6.10 Dimensions of grid strips.
Scatter reduction is achieved by providing a relatively high transmission to
the collimated desired photons, and a relatively low transmission to the iso-
tropic scattered photons.
We first subdivide T(O) into specific angular regions which have uniform
properties. Within each angular region, at each 0, we use the fact that each ray
has a uniform probability distribution of occupying each lateral position. We
thus further subdivide each angular region into regions having different attenua-
tion mechanisms and apportion each region uniformly. In those regions having
attenuations which vary with ray translation, we integrate to find the mean
attenuation. For example, for rays having angles to the normal 0 < 0 <
tan'-l(t/h), the rays are either totally in the metal strip, partially in the metal
strip, or unattenuated. T(O) is given by
T(e) = t - h tan 0e-lihlcos8 (totally in metal)
o<8<tlln-1 (tlh) 5
2h tan 0 1 J
ht
an8.
+ ------ e-lix/sm8 dx (partIally In metal)
5 h tan 0 0
+ s - t -.;. h tan B X 1 (unattenuated) (6.r)
whereJl is the attenuation coefficient of the metal strip. The first term .lias
thefactor (t - h tan 0)/5, indicating the fraction of the period 5 that the rays
are'!otally in the length of the metal strip and experience the attenuation
exp{-Jlh/cos 0). In the second term the duty cycle is (2h tan 0)/5, and the
attenuation term is the integrated average over the different parts of the metal
SOURCE
o
o
0 0'%
DETECTOR
LEAD STRIPS
DETECTOR
FIG. 6.9 Scatter-reducing grids.
. . I' focused arid we will stu
Since the analysIs IS unduly comp ex USIng a 1;;'
the collimated grid and make the reasonabl.y that t
scatter-reduction performance will be essentIally IdentIcal .In both
evaluating the performance of a grid structure, we calculate Its
a function of scatter angle in one dimension, T(e). The reductIOn factor
of the scattered photons is then given by
J
7f/
2
nsCe)T(e)de 2 i
7f
/2
R = 7fi2 = - T(e)de (6.
S J7fi2 n
-7f/
2
n.(e)de 0
The scatter can be reduced, with a small loss o.f by using
a collimated structure which is parallel to the dlrectIO.n of transmItted beam
[T P
. 1967] As shown in Fig. 6.9, the gnd consIstS of an array of
er- ogossJan, . .
thin absorptive strips which are parallel for the case of a collImated beam
and are pointed toward the source for a d!vergin
g
beam. The IS called a
d
'd' 't 's designed for a specI1k source to detector dIstance.
focuse gn SInce 1 1
COLLIMATED BEAM
I I I
t ,
SCATTER REDUCTION USING GRID STRUCTURES
99
(6.71)
(6.70)
(6.68)
Scatter Reduction Using Grid Structures
llsing a scatter reducing grid is given by
SNR =
+ Rsl/l/tlt
.the quantum. et?ciency of the no is the incom'ng
e IS the transmISSIOn of the body, A IS the area of a picture
e ratio of scattered to transmitted photons at the detector, and
n of the scattered photons passed by the grid.
ation (6.69), we can calculate the performance of some typical
curves of T(e) are shown in Fig. 6.11. In general, the values
ratIO of scattered to transmitted photons following the grid, is
attenuation coefficient of the filler material. Normally, relatively
>tt11mlber materials such as plastics and aluminum are used as filler
attenuation. Except for the use of filler materials, in theory,
could be made arbitrarily high for increased scatter reduction at
smission efficiency. When filler materials are used, T(e) is modi-
tiplicative factor e-Plh/cos8, ignofing the small path through the
3s - h tan e e-2Jtt/sinO + h tan e - 2se-3pt/sinO
s s'
tan-
1
e:) < e< tan-
l
e:)
Or, in general,
T(O) = +n(n + l)s - h tan O]e-npt/sino + (h tan 0 - ns)e-(n+l)pt/sin8},
tan-
1
(n:) < 0 < tan-{(n 1
1
)S] (6.69)
where n is an integer that takes on values from zero to infinity.
The transmission efficiency tit for the dimensions shown in Fig. 6.10 is
given by T(O) = (s - t)/s, assuming that the rays are completely stopped by
passing through the length of a strip. Often, however, filler material is used in
the space between the metal strips for structural purposes. Under these con-
ditionsl1t is given by
re t. In that case we get either zero attenuation or the complete attenuation
we .'
of one or more stripS as gIven by
h e e-Jtt/sinO + s - : tan e, 0 < e < tan-1
2s - h tan ee-Jtt/sino + h tan e - s e-2pt/sinO
s s'
T(e) = < e< tan-Ie:)
(partially in strip)
(
passing through entire)
strip width
...L h tan e- t -Jtf/sinO
I e
s
Noise Considerations in Radiography and Fluoroscopy
=+[s - t - h tan e + (h tan 0 - t)e-Ilf/sino
+ 2 e (1 _ e-Jtf/sinO)}
In the next angular region rays either pass through the entire strip wid
pass partially through a strip as given by
T(e) = h tan e - t e-Jtf/sine (through entire strip width)
tan-I[(S-f) /hj<O<tan-
1
(s/h) S
I S + t - h tan e 1 (' e-IlX/sin8
T S S - h tan e JhtanO-(s-f)
(partially in strip)
_ 1 [(h tan e _ t)e-Jtf/sino + (s + t - h tan () sin
- S s - h tane
This procedure is continued until T(e) becomes vanishingly small as
angle rays go through many metal strips and experience large
resulting T(e) is then integrated to find the effectiveness of.the gnd.
The calculations can be significantly simplified by igonnng the :fi
of the metal strips but considering their exact attenuation. a. ra
throuah a lead strip of thickness t at e would have a
I:;> I
However, because of the zero thickness approximatIOn, we can
ravs that pass partially through the strip. We thus use a model of
st;ips of height h and separation s whose attenuation behaves as If I
strip traversed by the ray. The third term is the fraction of the duty
this angular range, over which the rays do not strike the metal strip, and t
are unattenuated. Collecting terms and performing the integration, we obtai
T(e) = -l[(t _ h tan e)e-Jth/cOSO + 2 sin e (1 - e-Jth/cosO)
o<O<tan- 1 (f/h) S f.l
+ s - t - h tan ()J.
Similarly, we find the transmission in the next angular region tan-let/h) < 9
tan-l[(s - t)/h], where the rays are either unattenuated, pass through the en
strip width, or pass through a part of the strip as given by
T(e) = s - t - h tan e X I (unattenuated)
tan-J (f/h) <O<tan-J[(S-f) /hj S
98
Noise Considerations in Radiography and Fluoroscopy 101
MOVING SLITS
Linear Detector Arrays
FIG. 6.12 Scatter reducing, using translated slits.
of Fig. 6.12 is the use of a one-line detector array in place of
This linear array follows the translating sheet beam, achieving
DETECTOR
X-RAY
SOURCE
only a sheet beam to pass through the body at anyone time. The
ators on either side of the body must be translated at different speeds
that the transmitted beam is not intercepted. Most of the scattered
iII fail to reach the detector, as shown. Although these systems have
)VeIl, they do have the practical difficulty of requiring increased
wer output from the x-ray tube. Multiple-slit systems mediate this
,>ther configurations using rotating slits have also been studied because
';convenient geometry. I
[the order of unity. This can, of course, be considerably more in a region of
attenuation, such as in the path of a large bone, where the transmitted
are reduced and the scattered photons from the volume are relatively
unaffected.
Although the grids provide reasonable performance, there are clearly many
linical situations where they are inadequate and the remaining scatter con-
to limit visualization. Significant research is under way on a variety of
'mproved scatter-reducing mechanisms.
lOne approach is the sequential irradiation of the body with translated
slits [Sorenson and Nelson, 1976] as illustrated in Fig. 6.12. Here two translated
0.4 0.2
BRAD
0.6 0.8 1.0
o
0.1
FIG. 6.11 Scatter transmission vs. angle for two typical grid structures
using lead, where Jl = 60 cm-
1
Grid (a): s = 0.3 rom, h = 2.0 rom, and
t = 0.05 rom, resulting in R
s
= 0.106. Grid (b) has the same parameters
except that h = 4.0 rom, resulting in Rs = 0.034.
0.3
0.2
0.4
T{B)
0.5
0.6
0.7
0.8
0.9
1.0
100
FIG. P6.1
PROBLEMS
103
RECORDER
RECORDING
PLANE
PHOSPHOR SCREEN
[)
--1 r-
t
I
I
--jw
FIG. P6.4
Il,
r------
Noise Considerations in RadiographyQnd Fluoroscopy
FIG. P6.3
obliquity. factors calculate the signal-to-noise ratio using a
havmg an efficIency 11 and a resolution element of area A. The
signal is defined as the background minus the value at the center
image.
what ratio of r1 to r2 is the signal-to-noise ratio a maximum?
.ess section of an object is shown in Fig. P6.4, where the desired
tion is represented by the small structure of width w.
the SNR the x-ray photons emerging from the object
e SIze of a resolutIon element is A cm
2
late the SNR of the recording where the phosphor screen has a
fficiency of 11 and produces L light photons per x-ray photon, and
OJ
6.3 Acircular x-ray .source of radius r1 emits n photons/unit area during
the exposure tIme (FIg. P6.3). A cylindrical lesion of thickness t radius r
' 2,
and attenuatIon coefficient Jl2 is within a slab of thickness L where L t.
Noise Considerations in Radiography and Fluoroscopy
_ I
-
-_ J.l
__: t
i
-
-
6.1 (a) Neglecting scatter and assuming that 11 = 1, find the SNR for ima
the thicker region where Jll I (Fig. P6.1).
SNR:::= Cj$ART
which essentially approaches the ultimate that is achievable. These linear arra
are used in conjunction with the computerized scanners where the
provide both cross-sectional and In thiS context the outp
signals are digitized to facilitate processmg by digital computers and subsequ
storage and display.
(b) Assuming that the energy dependence of Jl is given by f.l
find the energy that maximizes the SNR.
6.2 An x-ray transparency t = a + b cos 2nfox at a depth z is imaged
L x L rectangular source having an intensity of n photons per
a distance d f;om the recorder. Find the signal-ta-noise ratio of
tant image, which is defined as the ratio of the peak amplitude ofth
to the standard deviation of the average value. The area of eac
has a negligible effect on the ability to resolve the sinusoidal im
the desired scatter reduction since the only sensitive area is the line itself.
one-line array has a number of added desirable it can be made
using high-quantum-efficiency electronic detectors usmg scintillators
and photodetectors. Thus not only does 11 approach umty, but. the resultant
signal is sufficiently high to override any subsequent electromc noise. The
effects of neglible scatter, high efficiency, and the lack of subsequent noise
processes provide a signal-to-noise ratio determined almost solely by the trans-
mitted photons as given by
102
105
SCREEN
/
I
J.L /
J.Lo dJ(
\
w--j
\
\
'-
L----.--!\
Noise Considerations in Radiograrph" and 'rl " ctuoroscopy
FIG. P6.8
f the scatter-reducing grid the ..
pproximated as a parallel'b from a point
on efficiency 11t = T(O) ,...., ( earn entermg a parallel grid
W X W source a - t)js. Calculate 11t for
, period s, thickness t and d from t.he detector. Tht
The angles involved 'are coefficient p.
nd cos e:::::: 1. suffiCIently small such that
no [1 + m cos 211'fOXl
FIG. P6.7
modulated x-ray image is recorded .
as shown in Fig. P6 8 FInd th by a one-sIded screen-
. . e record d SNR
as the peak of the sinusoid and th . e where the signal
erage background On th e nOIse as the standard deviation
er x-ray photon . t of ahverage the produces I light
th
' IC are transmItted t h
e events are recorded Th . 1 . ate emulsion
, 1..b . e pIxe area IS A h' h . '
eg 19I Ie effect on the resolut' ' w IC IS assumed
Ion.
no PHOTONS/cm
2
rs \
t \
T8
t \
I
I
\
__--- -d-----------
(a) Find the value of r
s
that provides the maximum SNR wit
at some specific plane z o .
(b) Using this value of r
n
find the SNR versus depth z.
(c) Assuming that the planar object produces P isotropi
photons per incident photon, find the SNR versus depth z.
6.7 (a) Calculate the SNR of the lesion image (Fig. P6.7) assu
Compton scatter component of flo is flS" The difference in
the lesion can be neglected. The detector has a capture effie
a resolvable element area of A.
(b) Calculate the modified SNR using a grid structure havin
T(e) placed against the detector plane. The separation s .
this case with the detector against the object.
FIG. P6.6
R events per light photon are recorded. Due to critical-angle consid
. d. d 1 h h I era
tlOnS, the recor er receIves emitte Ig t p otons on Y over a solid angle n
6.5 Equation provides If!, the :atio of scattered to transmitted photo
for the cylmdncal geometry of FIg. 6.7. Calculate If! for an object havin
square cross section W X W with the same length L. You can leave y:
anSwer m mtegral form.
6.6 A circular source of radius rs emits no photons per unit area (Fig. P6.
A planar object of radius R has a transmission t with an opaque cente
radius roo The image signal is defined as the difference in photons
pixel, with a pixel area A, between the background and the center of
image of the opaque disk at the center. Ignore all obliquity considera
and assume that 1] = 1.
104 Noise Considerations in Radiography and FluoroscOPY
7
Tomography
In this chapter we consider systems that provide important "tomographi
three-dimensional capability. The tomogram is effectively an image of a
taken through a three-dimensional volume. Ideally, it is free of the effi
intervening structures, thus providing a distinct improvement in the abi
visualize structures of interest.
In single-projection radiography the resultant image is the superp
of all the planes normal to the direction of propagation. In essence the
has infinite depth of focus, although, as was shown in Chapter 4, th
source size causes planes closer to the recorder to have better res
Ignoring this effect and assuming parallel rays, the recorded image is gi
Id(x, y) = 1
0
exp [- Sp(x, y, z)dz].
This integration over z often prevents a suitable diagnosis of the
of a section at a given depth plane. Since all other planes are supenmpo
subtle contrast variations of the desired plane are often obscured.
particularly true in studies of lung lesions where the superimposed rib st
obscure the visualjzation.
MOTiON TOMOGRAPHY
V t'l very recently the only method of isolating a view of a single plane was
tomography [Meredith and Massey, 1977], as shown in Fig. 7.1. The
J!lO lee and the recorder are moved in opposite directions. As shown, one plane
object remains in focus while all others have their images blurred. The
In ture and degree of the deblurring is determined by the distance of each plane
f
na mthe focused plane and by the extent and type of motion of the source
ro f' h . k
dfilm. These systems are 0 ten classIfied by t e type of motIon underta en,
an h'd .
eh as linear, circular, and ypercyclOi al tomography. The mechamsms that
these motions are quite elaborate since they must be both accurate
rapid, so that the motions can be completed in a few-second breath-holding
interval.
MOVING
FILM-SCREEN
CASSETTE
MOVING
SOURCE
FIG. 7.1 Motion tomography.
the source undergoes a specific motion in a plane parallel to
The path as a function of time can be characterized as
in general, is a two-dimensional delta line function which
otion of the source. From this motion and the corresponding
'efilm, we can calculate the resultant impulse response.
source motion of g(x, y, t), we immediately see that the resultant
nary film using a pinhole transparency is g(xlm, ylm, t) by straight-
etry. Thus the path due to source motion experiences the same
; m = -Cd - z)/z, as did the source image in Chapter 4. As indi-
vide a tomographic plane, the film is moved in a scaled
motion. Whenever the source is displaced from the axis at a
center is positioned at - kx, - ky, where k is a positive num-
'ng the scaling of the film motion. Thus the center of the film
th g(x/-k, y/-k, t).
lin moves, the resultant path of a beflm on the film moves in the
etion with respect to film coordinates. Thus the resultant path
Tomography
I
(7.14)
(7.12)
k = -m = d - Zo
Zo
__l_S(X
d
,Yd) ** 1 f( Xd Yd)
4nd
2
m
2
mm L(k + m)2 k + m' k + m .
graphic plane z = Zo where m = -k and f becomes a delta
tion continues to be limited by the source size. The expres-
ges to that of the point source as given by (7.5) and (7.7) if
) = KN5(x,y).
IaCu, v) = I
o
M2T(Mu, Mv) sine [X(k + m)u]. (7.13)
irection, a plane at z is multiplied by a low-pass filter having
width of approximately rX(k + m)r
1
At the desired plane
mes an infinite-bandwidth filter and does not affect the fre-
d(U, v). All other planes experience various degrees of filtering.
a finite source rather than a point source, the total impulse
onvolutionof the motion path function and source size as
function becomes a narrow delta function and reproduces the
in its original form. The plane of interest is at the depth z0 =
"hest tomogram"using linear motion is shown in Fig. 7.2. Note
of the ribs and spine.
uency domain the Fourier transform of t is multiplied by the
rm of the point-spread function h. For the case shown above,
Motion Tomography 109
( ) - t(X
d
Yd) * 1
0
f( Xd Yd) (78)
Ia Xd,Yd - M' M * L(k + m)2 k + m' k + m . .
The most widely used form of motion tomography is linear tomography,
here the source and film are both moved uniformly in straight lines in opposite
The source motion is by
g(x, Y, t) = 5(x - vt)5(y) reet (7.9)
where v is the velocity of the source in the x direction and X is the extent of the
traVerse. The resultant source path is given by
f(x, y) = Sg(x, y, t)d(vt) = rect (7.10)
ting a line of length X in the x direction. The resultant detected image
Jbecomes
Iixa, Yd) = t(::, r;) ** X(k1+ m) rect m)]5(Ya). (7.11)
llthe expression above could be eliminated by defining the convolu-
gone-dimensional in the x direction only. Thus each point at plane
into a horizontal line of length X(k + m). At the desired plane
IiXd' Yd) = t ** h
(
X Y) 'f'( Xd 3-),
= t M' M** B
j
k+'Yz' k + m
, alizin constant B, we use similar reas
In evaluatmg the norm g that the total number of pho
, Ch t 4 We assume .
that employed In er If the source is translated the
during the exposure IS l'!., d ring the exposure IS NIL, whe
photons emitted per UTIlt dIstance . u b
line integral of the pathf(x, y) as gIVen y
L = JJf(x, y)dxdy.
, 'it represents a
. ' dimensional mtegratlO
n
,
Although thIS IS a two-. f ' Thus the constant B, rep
. .)' delta lme unctIOn. , . b
smce f(x, y Is.a . t the detector plane, IS gIVen Y
intensity per Unlt dIstance a I
KN cos
3
() =
B = L4nd2 (k + m)2 L(k + m
" . ' defined in (4.2), (4.4), and
where Ii is the InCIdent mtenslty Ignoring obliquity fact
proportional to the energy pe: p that cos
3
() ::::: 1
Ii = 1
0
, the intensity at the aXIS, w IC
detected intensity becomes
108
h two motions is the impulse response,
incident on the film due to L e
is given by (X Yd t)
h(Xd,Yd,t) = Bg k -: m'k+'Yz'
due to both motions and B is a no
. h t tal maanlllCa IOn ,.
where (k + m) IS teo. ::> bse uently evaluated. The mtegrated Impul
malizing constant that will be su q
response is therefore )
Xd t d(vt)
h(Xd' YJ) = B Jg(k + m)' (k + m)'
(
Xd Yd)
= Bf k+'Yz' k+'Yz
. d th traversed by the source as given by
where f (x, y) is the Integrate pa
f(x, y) = Jg(x, Y, t)d(vt)
. ' h d' ection of motion. The resultant reeo
where v is the source velOCIty III t e lr
P
ath, as given above, is )
(
Xd Yd .
f k+'Yz' k+'Yz
f the geometry, is independent
This impulse response, ,becausde 0 thus be placed in convolutiona.
. f the object an can
lateral coordInates 0 1 t 'maae due to a transparency t a
h d pth P
lane '7, The resu tan 1 ::>
for eac e
z is given by
7.3 Sequence of radiographs used to create a tomogram.
<t,"4yltiple-Radiograph Tomography; Tomosynthesis
111
SUBJECT
Motion Tomography
--1 x r- ./SOURCE POSITIONS
. . .
SCREEN-FILM CASSETTE
'--"-'(xz,Yz)
ro
z
---- i
TRANSDUCER i
s(xo, Yo)
r
oz
c
FIG. 9.6 Basic propagation model.
Diffraction Formulation
Diffraction spreading [Goodman 1968' N t 1
ultrasonic wavelength repre;ents th or on
i
976] due to. the relatively
limits in ultrasonic imaging. In stu; de.ternunmg the resolu-
sider the propagation ignoring attenuat' y gb 1 ractIOn problem we con-
o . . ' lon, etween a point th d
,n.t?, x" y, plane and a point at depth z in the x on e .trans ncer
The basic propagation delay is model d b plane as FIg. 9.6.
of . e y t - roz/c), IndIcatIng a delay
DIFFRACTION FORMULATION
FIG. 9.4 Manually scanned B-mode system. (Courtesy of Siemens Gam-
masonics, Inc., Vetrasound Division.)
An abdominal B scan is shown in Fig. 9.5. This is an image of the
where the upper border is the skin line along which the transducer was sea
The hepatic vein is clearly shown. The curved lower boundary is the diaph
This analysis was made with certain assumptions about the ultr
model of the region of the body being studied and of the nature of the pr
tion phenomenon. The region was assumed to have a constant velocity of
gation c, constant attenuation rx, and composed of an array of weakly re
isotropic scatterers. The propagation phenomenon of the transmitted wa
assumed to be governed by geometric optics with diffraction neglected. W
proceed to examine these more closely and to refine them where appropri
We first consider the important problem of diffraction spreading an
consider our ultrasonic model of tissue.
The B scan represents, by far, the most widely used modality in ult .
. . I 'd d" . f h rasoIllc
Imagmg. t prO'll es a lrect representatIOn 0 t e cross-sectional anatomy hi
can be readily evaluated. A typical B-scan system with a manually tra WI ch
. h "F' 94 ns ate<!
arm IS sown mIg. . .
180 Basic Ultrasonic Imaging
182 Basic Ultrasonic Imaging
. affect the resultant waveform, including the
A number of lin.ear 'c to the propagation phenomenon and the
" eratIOn WhIC IS aSI . d'
denvatlve op "F venience these are combme mto a single
d h ractenstlcs or con , ,
trans ucer C a ' h erall impulse response representmg the sIgnal
impulse response a(t), Thus t e ov
received is given by JZ
, )_[J (t - roz) * aCt) 2"
h(xo,Yo;xz,yp
t
- c roz
th P
roduct of an obliquity factor zl '0., the.
h I
2 term represents e , h d'
where t e z ro
z
" d II' the normal falloff WIt Istance of an
cosine of the angle of mCldence, an Oz,
isotropic radiator. . h to find the field amplitude at any plane
We will use the Impulse responsle where '7 = 0 This field amplitude can
, I t the transducer pane -' '"
z due to a SIgna a .' tal parameters of the medIUm, mcludm
f the followmg mcremen Th
represent any 0 'I 1 'ty density and pressure. ese are a
, d' I ment partIc e ve OCI " .,. I
partIcle ISp ace, 'd artures from larger eqUIlIbnum va u
. 1 tities representmg ep f h I'
small-SIgna quan. '1 ferred to as a component 0 t e m
, .' eratIOn prevIOUS y re ,
The denvatIve op 'arameter is required for propagatIon
filtering aCt) indicates that a change map d at some point is proportional to
occur. For example, the pressure
f at a source pomt. , .
rate of change 0 pressure h ., driven by a sinusoidal burst providi
Consider a transducer at t e ongm
a field amplitude given by , )
u(xo, Yo, t) = s(xo, yo)p(t) exp (-IOJot
. 1 l't de distribution at the transducer, pet)
(
) 's the spatIa amp 1 u f
where s x o, Yo 1 ')' the sinusoidal carrier at a requency coo'
pulse envelope, and exp (-IC?ot IS d to a radiating point on the transd
field amplitude at any point m z lue response of equation (9.14), The t
l
ution with t e Impu se . ' h
is found by convo ) . found by convolvmg WIt equa
field amplitude at, plane z, plane as given by
(9.14) and integratmg over teen
u(x
z
, Yz, t) = JJu(xo, Yo, t) * h(t)dxodyo
. ' 'th respect to time, Substituting for u(x
o
, Yo, t
where the convolutIon WI ,
performing the convolutIOn gIVeS
JJ
)
(
ro:) eXP(ikro.)(+)dXodyo
u(xzoyz,t) = s(xo,yop t-
c
. roz
X exp (-icoot) * aCt)
_ c = 2nlA, Equation (9.17) repres
e
where the wavenumber k - OJo/_ Th 'ntegral operation prOVIdes the
. . ttern at plane ., e 1 t' s de
insonIficatIOn pa l' The temporal func IOn
extent of the pattern or its lateral reso utIOn,
depth resolution, as will be _ ' ( Y t), We now st
. f ch pomt at plane . IS U x:, z' ,
The insonificatIOn 0 ea h t ansducer, Assummg
response from the reflected wave back to t e r
Steady-State Approximations to the Diffraction Formulation 183
reflecting point at Xz' Yz, the received signal eh(t) is given by
eh(Xz,Yz, t) = SSu(xz,Yz, t) * [J(t - * (9.18)
where 6(t - 'ozlc) is the impulse response from the reflecting point to each point
in the transducer, bet) represents the linear operations between the reflecting
point and the received electrical signal, and zl is again the falloff in amplitude
with distance, including the obliquity factor zl rOz' The signal due to the reflecting
point is thus derived by integrating over the transducer area s(x
o
, yo)' Performing
the temporal convolution eh(t) becomes
eix.,yz,t) = SS[55 s(xo, Yo) exp (ikroz)p(t - -
X exp ( 2) * aCt) *bet) (9.19)
'Oz
where the primed coordinates are used to distinguish the integration of the
reflected components. The uncompensated envelope response e(t) for a general
object with reflectivity R(x, y, z) having a uniform attenuation is then given by
e(t) = Ifff e-
2CXZ
R(x, Y, z)eix, y, t)dxdydz I (9.20)
STEADY-STATE APPROXIMATIONS
TO THE DIFFRACTION FORMULATION
In equations (9.17) and (9.19) the pulse envelope pet) appears in the spatial
defining the system spatial response. Physically, the transmitted pulse
from different portions of the transducer arrives at each reflecting point at
times, and the reflected pulse arrives at different portions of the trans-
different times. If the pulse is relatively short, the resultant transducer
will depend on which portion is being excited. For very long pulse envel-
the entire transducer is excited by essentially the same sinusoidal
so that a steady-state analysis represents a good approximation.
In making the steady-state approximation, we assume that
p(t - - (9.21)
1Jlakes p( .) independent of x and y, and allows it to be moved outsidejthe
I in equations (9.17) and (9.19), where it affects only temporal or depth
ion, not lateral resolution. Physically, we are assuming that the envelope
transmitted waveform from all parts of the transducer arrives at each
lane z at approximately the same time. Similarly, the waveforms reflected
reflecting point arrive at all portions of the transducer at approximately
Thus the lateral resolution considerations are governed solely by
Fresnel Approximation 185
f
Oz
Z + (xo - xz)2 t (Yo - yz)2. (9.30)
This approximation is valid in regions where Z3 (nJ4l)[(x
o
- x
z
)2 +
This inequality ensures that the exponent due to the third term
in expansion is significantly less than unity. As in optics, this
PRgximation is quite accurate for systems with reasonable angular fields. The
at plane z with the Fresnel approximation becomes
h(xz>yz) = h(x,y,z) = eikZS(X;y) ** exp [i;/X
2
+ y2)J (9.31)
rent depths. The convenient convolution form can be used since, using
approximation, the impulse response is space invariant, depending
difference between the spatial coordinates.
impulse response can be used to find the field amplitude h(x
z
, yz) at
z due to any transducer distribution sex, y). The effect of this convolu-
In the Fresnel region [Goodman, 1968J we approximate 'Oz in the exponent as
the first two terms of the binomial expansion of
'Oz = z,J1 + (xo - x z)2 -; (Yo - yz)2 (9.29)
FRESNEL APPROXIMATION
h(xz, Yz) = ffs(xo, Yo) exp dxodyo (9.28)
As a first approximation we apply the paraxial approximation where the
term !/z z --: 'Oz the axis. It should that
this approximatIOn IS relatIvely msensItlve because of the multIplIcatIve effect of
this term. This is in sharp distinction to the 'Oz term in the exponent because of
its greater sensitivity. The approximations of 'Oz in the exponent are divided into
the Fresnel and Fraunhofer regions, often referred to as the near-field and far-
field regions.
Note in equations (9.25) and (9.27) that the integration the
olution is squared because of the symmetry of the transmlttmg and receIvmg
reseration. Thus, in the steady-state approximation, the overall lateral resolu-
nis the product of the identical transmitter and receiver patterns. We will now
tJoceed to study approximations to the lateral response function in the brackets
pr as to make the resultant expressions more tractable and subject to analysis.
;e separate out the lateral response of the transmit and receive operations as
given by
, - Z
1:
S
elatively narrow angles are usually in
,.fi(Df2)2 I Z2 mce r 'd' .
where 'OZmox = T' .' :::: Z + (Df2)2f2z, provl mg a
lved we can use the 'oz"";
'for the steady-state approXImatIOn gIven by
D2
1: 8zc'
d d pth
of z = 10 cm, the pulse durati.
. f D - 2 cm an a e
For a transducer sIZe 0 - 03 c This is achieved by most sys
must be significantly greater than . l' However for larger transducers
f b
tl 0 "sec are typlca .
since durations 0 a ou . t'" 'tion will be poor. A more ace
shorter depths, this steady-state
analysis will be considered f n of (9.21), the transmitted
Applying the steady-state approxlma 10
amplitude at depth z is given by z
u(x" y" t) = ell s(x., y.) exp (ikr,,) (r;,) dx.dy.J
x p (t - exp (-iWot) * aCt)
I r depth resolution functions are clearly
where the lateral and tempo.ra 0 t a point reflector eh(xZ ' Yz,
rated. The overall round-tnp response 0
equation (9.19) then becomes 2 . ( 2z)
e,(t) = U1s(x., y,) exp ikr., dx.dy.Je-'WO'P C
1 considering all the 1mear
where pet) is the effective pulse enve ope
where (2Z)' (2Z) e-iwot * aCt) * bet).
-,wol - p t --
pt-- e - c
c 'eneralized ob
ut of the system usmg a g
The envelope-detected outp . by
t d for
tissue attenuation is then gIVen 2 )
compensa e 2 Z
\
rrf )[f rS(xo,yo)eikro,(+)dXodYo] p(t - c
ec(t) = K \ J J JR(x, y, Z J J 'Oz
184 Basic Ultrasonic Imaging
. d veforms at different portions of the trans-
the relative phases of the receIVe wa
ducer, not by the P(t)f this steady-state approximation, we can calcu-
To estimate the 0 b teen the center and edge of a transducer
. d 1 dIfference e w
late the maxImum e ay " th center of the beam. For the steady-state
fl
fng pomt m e
of extent D for a re ec 1 . f the pulse envelope pet) should be long.
. . h ld the duratIOn 1: 0 . d'
approXImatIOn to 0,. Th' ensures that the entIre trans ucer IS
h
d lay dIfference, IS . .
compared to t IS e ble duration Thus the apprOXImation...
. 'fi d for a reasona' .
simultaneously mSODl e
criterion may be stated as
186 Basic Ultrasonic Imaging
(9.36) evaluated at x = vnT, y = Yo
as before, is the time of each scan line.
instructive to c?nsi.der the affect of the phase-shift term exp (i2kz)
the round-tnp tIme to each plane. If we have reflections at one
only: each will experience the same phase shift. This phase
exp WIll then the magnitude is taken, representing
p.' e However, gIven reflectIOns at different depths, each reflectik.n
WIth a phase shift. The resultant signals, as represeAt-
(9.35), add. or destructively, depending on
elatIve phases. ThIS addItIOn and cancellation of signals results in a
d pattern known as "speckle" whose properties are studied later in this
If this phase factor is ignored, the resultant image will be the inco-
of the reflectivity, devoid of speckle.
Fresnel Approximation 187
and the diffraction spreading in both directions, llz
z
. As before, we can assume
that this factor varies very slowly compared to the effective pulse
envelope pet - 2z/c). Thus the envelope function acts as a delta function on the
attenuation factor, allowing it to be taken outside the integral as a time variation
As in equation (9.3), a compensating gain variation is used where
get) = (ct)ZelX.cr, resulting in a compensated output signal given by
etCt) = KI fffR(x,y, z)e
i2k
: {s(x,y) ** exp [i;/X
Z
+YZ)J} Z
X p(t - I (9.33)
For A using a stationary transducer positioned at x = 0, y = 0,
the resultmg estImate of the reflection coefficient along the z direction is given by
R(O, 0, z) = K!R(X,y, z)e
i2kz
*** {s(x,y) ** exp [i;/X
Z
+ yZ)J} Zpe:)!
evaluated at x = 0, y = O. (9.34)
For a B. scan: again the transducer moving in the x direction along the
r== Yo lme WIth a velOCIty v the gain-compensated signal is approximated as
ec(t) = K IfffR(x, y, z)e
iZkZ
{sex - vnT,y - Yo) ** exp [i;/X
Z
+ yZ)J} Zp(t - 2:)dXdYdz!. (9.35)
resu!t:ant displayed estimate of the reflectivity in the y = Yo plane is then
by
z) = K !R(X,y, z)e
lZkz
*** {S(x,y) ** exp [ifz(x
Z
+ yZ)J} Z
x pe:)!
-6 dB
Lateral distance
On.axis pressure amplitude
Having described the steady-state diffraction behavior in the Fresnel regi
we can to defining the received signal from a reflecting object as in eq
tion (9.21). In the Fresnel region the envelope detected signal is given by
e(t) = K \ JJJexp (;;2iXZ) R(x, y, z)e
i2k
:
X !s(X,y) ** exp [i fz(x' +Y'lJl 'p(t -
The overall attenuation factor is exp (- 2iXZ)j z2 due to the tissue attenuatio
FIG. 9.7 Steady-state field pattern of a circular transducer. (Courtesy of
Siemens Gammasonics, Inc., Vetrasound Division.)
o 0.25 0.5 0.75 D
2
1f... 1.0
I -----+-- Axial distance
'l-Nearfield ------Farfield ------+-
tion is relatively complex. The amplitude distribution of a circular disk trans-
ducer is shown in Fig. 9.7. As is seen, the field amplitude in the immediate
vicinity of the transducer has an oscillatory pattern whose extent is approxi-
mately a geometric extension of the transducer. At a depth of D2/4A, the
oscillatory pattern diminishes and the pattern begins to diverge uniformly. This
distance is referred to as the near field by some authors. The oscillatory pattern
extends to a depth of about D2/2A for a square transducer, twice that of the
circular transducer. At a distance of D2 jA the 3-db width of the beam equals the
transducer diameter. Beyond this depth we are clearly in the Fraunhofer or
far-field pattern, which is the next subject of discussion. In this region the shape
of the pattern remains fixed and its size linearly increases with depth, represent-
ing a fixed angular pattern having an angle of approximately A/ D.
FRAUNHOFER APPROXIMATION
Fraunlzo/er Approximation
Consider a square transd 189
It tIl' ucer where sex y) -
res
u
an atera spatIal patter . th f. .'. - rect (xl D) rect (yl D) The
nIne ar field IS glVen by .
h(xz,Yz) = e;V;y {rect(Xo)rect(:to.)} = eivDz. (DX).
. D D Z SInC T smc(DYz) (941
where SInC x = sin (nx)/nx Th fti.. Z AZ .)
d
fi d h . e e ectIve WIdth f th .
e ne as t e region where its ar um . a e smc function may be
beam WIdth at any depth z is ).zl D. :nt IS bet,,:een -j,. In this region the
In FIg. 9.8, showing the near collim t dPproxlI:natIOn to the total field is show
diverging behavior in the far behaVIOr to a depth D'il and then
reglon shown In Fig. 9.7 hav b . varIOUS perturbatIOns within the F 1
h
. . e een Ignored A . h resne
avmg a maXImum diameter of D .' S IS S own, a large transducer
depth Dill. If a smaller .approximately collimated to
Improve lateral resolution its perfor D
z
IS used in an attempt t
beca
use f th d ' mance rapIdly det . 0
. . a ere uced value of D'I). enorates at greater d th
FIgure 9.8 illustrates the diffi
z
ul't' .. ep s
One ap h . c YIn desIgmng I
proac IS to provide approxim t 1 an u trasonic transducer
a e y near-field performance at all depths:
.//
... ././
"....
"...../
./
./
"....// -----
D, D2
.-l :'--
-l i r- "" --------
................... ----
"
"
"
"
"
""
FIG. 9.8 Approxim t fi .........
a e eld patterns for different transduce . r SIZes,
the transducer dimension D' h
DZ/A., is equal to the maxim IS c d
sen
such that the limit of t
hi
..
optimu t um epth z Th t"
m ransducer size becomes max' us lor collimatetl
. depth of 20 cm and a mlax. (9.42)
r SIze of 10 wave ength of 05 m th'
. . cm. This is a representativ fi . m IS represents a
ents In resolution can be obtained b e for modern instruments.
y redUCIng the wavelength. How-
where v = k(r;12z + z) and is the Fourier transform operator using spa
frequency coordinates u and vas indicated. Thus in the far-field or Fraunh
region, where z D' I)., the compensated received signal is given by
dt) = K\ 111 R(x, y, y)JJ'p(, - ) dxdydz I
In this far-field region the received signals and reflectivity estimates c
obtained by substituting for the lateral response termS (s(x, y) ** exp [i
(x' + y')]}' in equations (9.32) through (9.36) the simpler exp
y)J],. This formulation serves to illustrate the performance probl
collimated ultrasonic imaging systems in the far field. This far-field respOD
Fourier transform of the source function, is known as the "diffraction-Ii
response since it represents the best resolution for a given source confi
where r; = x; + y; and rl =xl + yl As before, the phase factor outsidethe
integral can be ignored when investigating a specific plane. Bowever, for reflec-
tions at various depths this z_dependent phase shift results in coherent speckle.
As for the quadratic phase factor within the integral, for a transducer having a
maximum lateral dimension of D, the exponent will have a maximum value of
"D'14).2, or approximately D'/).z. Thus for depths greater than D'I)., the
exponent will be less than 1 radian, and can be neglected. This leaves only
two-dimensional Fourier transform kernel in the integral, so that (9.3&) can
approximated as
r-J e
iv
rr r .(2n\ J
h(x" y,) - z JJ s(xo, Yo) exp L-' ).z}(xoX, + yoy,) dxodyo
eiv""'r ( )l u- _ X
z
Y
:::::: z;)" 1.. S X 0, Yo J ;'z V = ;:z
The general results in the Fresnel region, under the approximations outlined
are valid at a11 depths. Bowever, in the far-field or Fraunhofer region, the expres:
sions can he simplified. This not only provides simpler mathematical operations
but also represents considerable insight into the nature of the imaging patterns:
We return to the convolution relationship of equation (9.31) in integral
form representing the lateral impulse response as given by
h(x" y,) = e;' 11 s(xo, Yo) exp H', [(x, - x 0)' + (y, - Yo)'J\ dxodyo. (9.37)
This equation can be restructured as
h(
_ ) _ exp [ik(r;/2z + z)1 rr ( )
Xz,Yz - z J J S xo,Yo exp 2Z
X exp + Yoyz)Jdxod
Yo
(9.38)
191
(9.47)
Widehand Diffraction
The previous formulations all d h u1 1 . use t e steady-stat .
p se enve ope m equation (9.17), pet _ r c' e where the
eau:atlcm. (9.24). As previously indicat:i as pet - zlc), as
WIth short pulses and/or 1 ,IS apprOXImatIOn can be poor l'
arge apertures T d n
1976] for the short pulse or wideba 0 stu ya m?re exact solu-
field pattern at a depth z W . r nd case we WIll consider the
. e SImp Ify the formulation of (9.17) as
u(xzo y., t) = SS sex ) ( ( 0, Yo exp ikroz)p t - r;z)
X exp(-icoot)
WlDEBAND DIFFRACTION
bution in the Fresnel equation (9 38) . . as gIVen by
h( ez'V SS x
z
, y z) = - s x k
z ( o,Yo) exp exp
x exp [-i(21r)(x J )'z oxz +YoYz) dxodyo. (9.46)
As can be seen, where z = f, t d
amplitude is the Fourier :qual to the focal length, the field
results m an effective lateral heam width ource dlStnbution. As before, this
large aperture system can result' 1" at the focal plane, of ).+1D Thus
h A d
mare at!vely s 11 :J I' . a
dept. t epths outside this regio th ,rna ,well-defined beam at thI'S
. ' d' , n e quadratIC ph ....
In vanous Ivergmg patterns Fo" k ase lactor returns result!'
b fil
' . r wea ly focused " , ng
er D IS appreciably greater than I the systems. where the system F
In FIg. 9.9 where the resolution is b ' behaVIOr IS slmilar to that sho
on eiilier side. est at the focal plane and degrades gradua7I;
p(t - r;z) exp (-icoot) = p (t - r
oz
) exp ( , )
. c -lcoot * aCt)
IS the effective pulse envelo .
again use the Fresnel app P: m :he transmitted waveform.
ct. roxnnatIOn of equaf (9
ons ramts apply to the wideb d Ion .30) since the same
II d an case The r I z an the r
oz
in the ex onent ' . amp ltude factor I
by the first two terms of the b' P . I and the argument of p are I
. momla expanSIOn, giving
t) = _e'_kz SS sex ) {k
z 0, Yo exp i-rex - x )2 +( } 2z z 0 Yz - Y0)2]
X p[t _ _ (xz - X O)2 _ (yz - yo)2J
2zc 2zc dxodyo (9.48)
T
U
' o \ (/,JOlf
\1\
J- \
r---
f
single depth plane and its vicinity. The acoustic lens is generally made 0
plastic material which has a velocity of propagation greater than that of wa
Thus, unlike the case of light optics, the refractive index n is less than nni
where unity represents the index of the surrounding water. The relative pli
delay through the material at each lateral position x, Y is given by
{)(x, y) = ken - l)d(x, y)
where d(x, y) is the thickness of the lens at each position. We can approxi
the spherical surface as a quadratic surface in the paraxial region. In ilia
d(x, y) can be approximated by
d(x, y) :::::::: K + x
2
t.
y2
where R is the radius of curvature of the lens and K is a constant. I
constant phase factors the phase shift of the lens is thus given by
x2 -L y2 x
2
+ y2
{)(x, y) = ken - 0--TIr- = -k-V-
where f, the focal 1ength is - (n - 1)/ R. Since the refractive index is
one, this system is a positive converging lens.
To study the field patterns this phase factor is added to the sour
FIG. 9.9 Acoustic focusing system.
One method of modifying the performance shown in Figs. 9.7 and 9.8 is through
the use of acoustic focusing, as illustrated hy the lens system of Fig. 9.9 [Good-
man, 1968]. Bere an acoustic lens is used to ohtain optimum resolution at
ACOUSTIC FOCUSING
ever, since the attenuation rt is frequency dependent, the increased frequenc
results in excessive attenuation at the greater depths. The situation is qui!e
different, however, in systems designed for superficial imaging of the eyes,
tbyrold, vessels m the neck, and so on. These systems mvolve a maximum depth
of about 4 em. Because of the reduced attenuatIOn problem frequencies as high
as 10 MHz (1 = 0.15 mm) can be used. In these cases the optimum transducer
size D is 2.4 mm, representing a considerable improvement in resolution.
190 Basic Ultrasonic Imaging
192
Basic Ultrasonic Imaging
Wideband Diffraction 193
(9.56)
(9.54)
lu(ep,O)! =
is plotted in Fig. 9.11. Here the response is seen to be a very short pulse,
function, for ep = and increasing in width as J increases. The shading
for r' > D Iep I and
t')1 = Isine Irect (Dep r/) -t Isin [1 (t' + ,,+ ] I
x rect (t' +/!>/2) + Isin [1 (t
l
- t + ] Irect (t' -/:/2)} (9.53)
for r' < D 14> I. In this case the absolute value of the sum becomes the sum of
the absolute values because the terms occur at different times. Notice that in
the second and third terms of each expression it becomes more convenient to
express the transform as a sine rather than a sine function.
These relatively complex equations show the interaction between the tem-
poral or depth response, using the coordinate t' , and the lateral or angular
response, using the coordinate . We can study some special cases in the interest
of clarification. Setting t ' equal to zero, we limit ourselves to the depth plane
z = Tc and obtain the lateral response as
Similarly, we obtain the response along the z axis by setting ep = and obtaining
D (t')
Iu(O, t') I = z rect 7' . (9.55)
On the t' and ep axes we observe the relatively straightforward responses
that were used in the steady-state analysis. In the steady-state analysis these
were also the off-axis responses. If we plot the effective extent of the response in
for example a plot of ep versus t ' indicating the lateral and depth resolu-
it would be a rectangle under the steady-state approximation since the
ternpc)raI and spatial responses are assumed to be noninteracting.
typical response using the more accurate formulation of equation (9.52)
in Fig. 9.10 for a relatively wide pulse. This figure can only show the
c<o,l'\",r,",v,rn",t", extent of the response since the amplitude at each value of ep and
r..,'l11r.." the third dimension. It does illustrate the important deviations from
ste,amf-state approximation. We see the diagonal arms resulting primarily
the last two terms in equation (9.52). These fall off as I/ep, which is not
in the figure. These represent undesired responses well removed from the
of interest. As the pulse length r increases, the relative amount of energy
diagonal arms decreases to the point where the effective extent of the
approaches a rectangle, as in the steady-state approximation. I
In the other extreme, for very short pulses, the response is dominated by'the
term in (9.53), as given by
1 (r') (t' )
!u(>, t') I::::::: z J rect D>
we obtain the two-dimensional transform
Expanding the quadratic terms,
relationship
Iv "[ !- _ (xz - X
O
)2 _ (Yz - Yo)2J}
u(x
z
, YZ> t) = e
z
;:F {s(Xo, Yo) exp 2Z p t - c 2zc 2zc
x exp (-iOJot), u = l;, v = f;' (9.49)
, , I' the steady-state case, (9.49) is difficult to evaluate,
As wIth the Fresnel mtegra m , h d'
a' , lif the formulation in the far-field regIOn were qua ratle
We can abam h
Y
h ent and the argument of p can be neglected. In
,n x y m bot t e expon ,"
terms 1 0, o. f lation to the x dimensIOn to aVOId exceSSIve
addition we restnct our ormu " , b
: ' , I that case the transmItted field IS gIven y
complexIty m the notatIOn, n
(
t) ::::::: e
lv
{S(XO)pll (t - T + xzxo)} exp (-iOJot) (9.50)
u x:, Z zc
T
_ I i 2/2zc representing the approximate time delay from the
where - Z c T X z ,
" to the reflecting point, I
ongm , , I ssume a rectangular source and a rectangu ar
As an IllustratIve examp e, a fi ld l' d
( )
_ t( ID) andj5(t) = rect(tlr). The e amp Itu e
pulse envelope where s Xo - rec xo
is then given by
u(x" t) exp - v)]5' lrect rect (x
o
- (t.:;;./;;(Zc/x,))\. (9.51)
, I tions' the cases where D > \uclXz I and
This product two
h
g::r:
he
o;maller' rectangle dominates the product of
where D < 1rzclx
z
)' ,n c , have three regions, corresponding to
the two functions. WIthm each optIOn we function either fits into the
differen,t time,S, where the is reduced. Thus each option
one or IS at eIther end where th r f ach being a Fourier transform of a
represented by the sum of three
l
IOn
r
s, et'ons whl'ch define the time intervals.
1 ' l' d b ectangu ar tIme unc 1
rectangle mu tIp Ie y r , S' the time response represents 0
in which the different occur. _ T _ t)c representing time as
depth resolution, it is convementffto tt -d (d tl'me S'I'milarly we let r ' = 't'
. f n e ectlve re ar e ,
equivalent depth m terms 0 a WId fi A" = x Iz as a close approxim
the equivalent length of the pulse. e a so ,e ne 'f' "z
1
h b' ct point makes WIth the ongm. . .
tion to the ang e eac 0 Je, , h factors we limit our express
lO
For simplification, to aVOId carrymg P " d of this chapte
A" ') I Th' . 'n keeping WIth the remam er
to Iu(xz , t) \ = Iu('f', t. IS IS
d
I d here appropriate since envelope dete
where phase factors have been roppe w ,
, , ssumed The overall response is therefore gIVen by
tIon IS a '
DA" \ ( t
'
) I A \ ' [l.(t
l
+ + -
\u(<P, t')\ = +{D\ sinc (-f) rect r' _ D4> T n4> sm A 2. 2
X rect (I' t;12) + :4> Isin U (I' - i - J\ rect (" :ot)\ (9.?
195
10
0.18
3-10
40
1.65-1.75
1.35-1.68
0.002
Attenuation
Coefficient (db/em)
Ultrasonic Characteristics of Tissue
Air
Blood
Bone
Lung
Muscle
Other soft tissues
Water
Material
ATTENUATION AT 1.0 MHz
TABLE 9.1
Attenuation
We have thus far assumed a uniform attenuation coefficient throughout the
object being studied. However, in practice, the attenuation coefficient is a func-
tion of both the particular tissue and of the frequency of the propagating wave
[Woodcock, 1979]. The attenuation mechanisms in biological materials are not
well understood. In many common fluids, such as water, the attenuation is
primarily due to viscous absorption. In these cases the attenuation is propor-
tional to the square of the frequency. In most biological materials, however, in
the frequency range 1.0 to 10 MHz, the attenuation varies directly with the
frequency. This mechanism is usually attributed to a relaxation process in which
energy is removed from the propagating wave by an oscillating particle and then
returned at a later time. Some typical values for attenuation at 1.0 MHz are
given in Table 9.1.
indicates the falloff in amplitude with . As can be seen from Figs. 9.10 and 9.11,
the pulse waveform should be chosen to minimize the volume in the t'- space,
thuS providing the best compromise between depth and lateral resolution.
Using the same techniques, the complete transmitter pattern u(x
z
, Yn t) can
be studied. The round-trip response to a unity reflecting point e,,(xz, Yz, t) is
again derived using equation (9.17). The received signal from a generalized
reflecting object R(x, y, z) is derived using equation (9.20). In each of these
equations the far-field and paraxial approximations can again be applied to
simplify the computations and achieve insightful results.
ULTRASONIC CHARACTERISTICS OF TISSUE
2t'
= --
D
= ~
D
FIG. 9.11 Transmitted field pattern for very short pulses.
_______- ~ : - - - - - - - - - t '
_------+--4E-:---k--------t'
FIG. 9.10 Typical transmitted pattern, using wideband diffraction con-
siderations.
Basic Ultrasonic Imaging
194
196
Basic Ultrasonic Imaging
Ultrasonic Characteristics of Tissue 197
TABLE 9.2
PROPAGATION VELOCITY
(9.59)
(9.60) v{ cos B{ - V
r
cos B
r
= V
r
cos B:
i, r, and t indicate incident, reflected, and transmitted
Reflectivity
Our development assumes that the reflectivity of tissues can be modeled as
an array of weakly reflecting isotropic scatterers having a reflectivity R(x, y, z).
Because of the isotropic nature of the reflections, a falloff of liz was assumed as
in equation (9.1). In general, the reflectivity depends on both the shape and the
material in a relatively complex manner [Nicholas, 1977]. The simplest behavior
occurs at a planar interface between two materials. The resultant reflection is
called a specular reflection, as differentiated from the diffuse reflections we have
assumed. The planar surface acts as a mirror and reflects the wave at an angle
equal and opposite to the angle of incidence. In this case the amplitude of the
reflected wave received by the transducer becomes a strong function of the posi-
tion and angle of the transducer. The mathematical development for specular
reflectors of general shapes is more complex than that given for isotropic scat-
terers. However, the general concepts of resolution and diffraction considera-
tions remain the same.
In the earlier ultrasonic instruments, where binary images were produced
which essentially outlined organs and lesions, these specular reflections were the
significant information. In modern instrumentation with large dynamic
and gray-scale displays, the diffusely reflecting, isotropically scattering
have become the most significant, hence the use of the model in this
contains clinically useful information. For example, certain malignant tumors
exhibit increased propagation velocity. A method for reconstructing both
c(x, y) and lX(X, y), using the techniques of computerized tomography, is shown
in Chapter 11.
The reflectivity is dependent on changes in the acoustic impedance. This
\'IlllPedallce Z relates the pressure P to the particle velocity v as given by
P = Zv (9.57)
the acoustic impedance is given by
Z = pc (9.58)
P is the density and c the velocity.
behavior at a planar interface between two materials can be studied
Fig. 9.12. For equilibrium, the total pressure on each side of the interf1ce
equal, and the particle velocity on each side of the interface must !be
These conditions are satisfied when
330
1450
1500
1520
1540
1541
1549
1561
1566
1570
1585
1620
4080
1480
Mean Velocity
(m/sec)
Air
Fat
Aqueous humor of eye
Vitreous humor of eye
Human tissue, mean value
Brain
Liver
Kidney
Spleen
Blood
Muscle
Lens of eye
Skull bone
Water
Tissue
Velocity
., . . the reflection mode is the assumption
The basis for ultrasomc m h t the body The round-trip time of
f velOCIty t roug ou . . .
of a propaga Ion. . s de tho Fortunately, although vanous matenals
each echo IS used to detern:me It. P t'c velocity the soft tissues of the body
d
h es m theIr acous 1, . .
exhibit profoun c ang 5 0 / Some representative values are gIven In
are limited to a range of about /0'
Table 9.2.
. 11 ometric distortions in the reproduc
Variations in velOCIty cause sma ge tant propagation velocity.
. d' 1 ystem assumes a cons . b m
images smce the ISp ay s . d deflect the propagatmg ea ....
. . t' can dIStort an h t
addition velOCIty vana Ions 't has been found t a
, . s In some cases 1
causing additional geometnc erro!'. h hout the object of interest, c(x, Y
distribution of propagation velOCIty t roug
ffi' t f the biological tissues varies approximately
As indicated, the coe . 0 f ater and air vary as the square of the
directly with frequ.enc
y
whIle t a :tion path is primarily soft-tissue struc-
frequency. In studIes where p Phg the abdomen a fixed-gain compensa-
tures of comparable attenuatIOn, I
SUC
th
as
cases involving blood pools or fluid
. 11y adequate n 0 er .
tion system IS usua . h onent of the gain compensatIOn over
., f d' able to vary t e exp
regions, It IS 0 ten eSlr . 1 rs make this option available.
the path. Many commerCIa scanne
198 Basic Ultrasonic Imaging
(9.63)
(9.62)
COMPOUND SCAN FOR SPECULAR INTERFACES
At specular interfaces, as shown in Fi 9 12 . .
fashion where 8 = -8 As' d' t d
g
: '.' the wave IS reflected In mirror-like
r i' In Ica e In FIg 9 13th' .
reflection from an organ interface co I' t'l ' . IS.can result In the specular
resultant image will therefore be mis . mp e e y the The
the interface. SIng some 0 the InfOrmatIOn relating to
TRANSMITTED
MATERIAL 2
----
0,-----
P" c,
INCIDENT
MATERIAL 1
components, respectively. Using Snell's law, we have
sin Of C1
sin Or = G (9.61)
and, as in electromagnetic theory, we set the angle of incidence Of equal to the
angle of reflectance Or'
The reflectivity, defined as the ratio of the reflected pressure to the incident
pressure, is found using equations (9.57) through (9.61) and is given by
R _ Pr _ Z z cos 0f - Z 1 cos Or
- Pi - Z2 cos 8f + Z1 cos 8r
where Z 1 and Z2 are the acoustic impedances in the two interfacing media. At
normal incidence where Oi = 8r = 0, we have
R=ZZ-
Z
1.
Z2 +21
Table 9.3 gives the reflectivity R at normal incidence for a variety of tissue
interfaces.
TABLE 9.3
REFLECTIVITY OF NORMALLY INCIDENT WAYES
REFLECTED
Materials at Interface
Reflectivity
ANGULATED TRANSDUCER RECEIVES SPECULAR REFLECTION
MISSED
REFLECTION
I I
./ DISPLAY
I I ",.
;z: G
ORGAN BODY SURFACE
INTERFACE
FIG. 9.12 Behavior at a plane surface.
FIG. 9.13 Effect of a compound B scan. '
;n,is problem can be alleviated with ma 1 I
,r
und
translation-rocking by utilizing a
:l number ofline scans are obtained 'th ' . t each lateral pOSI-
tions, all in the plane of the desiredWI the in different
.on and angle of the transd crhoss sectIOn. record the
ucer so t at the receIved e h t
appropriately recorded in the display Th th c oes a each angle
. us e transducer eventually
0.66
0.69
0.08
0.08
0.10
0.09
0.10
0.09
0.03
0.03
0.01
0.05
0.9995
0.89
Brain-skull bone
Fat-bone
Fat-blood
Fat-kidney
Fat-muscle
Fat-liver
Lens-aqueous humor
Lens-vitreous humor
Musde-blood
Muscle-kidney
Muscle-liver
Soft tissue (mean value)-water
Soft tissue-air
Soft tissue-PZT5 crystal
Here we see that the interfaces between soft tissues have a
under 0.10, representing less than 1%of the energy being reflected. This
cides with our "weakly reflecting" assumption, where multiple reflections we
ignored. However, a number of other interfaces, such as between tissue a
bone, tissue and air, and tissue and the transducer, have strong reflections. Th
certain clinical situations can result in multiple reverberations, giving rise
false echoes.
Speckle Noise 201
The calculation of SNR using (9.64) and (9.66) is complicated somewhat
by the constant K. This depends on a variety of factors, including the piezo-
electric constants of the transducer. Equation (9.66) can be restructured in terms
of measured values using a water tank. Assume that a specular
reflector wIth umty or known reflectivity is placed close to the transducer face
so as to. be within the near field. This is an easily performed experiment using
a matenal, such as a metal, whose acoustic impedance is very different than
water. Since the reflector is in the near field, no diffraction effects are involved.
In addition, the atter:uation in water tank, over the short path, is negligible.
The peak reference SIgnal E
r
wIth R = 1.0 and no diffraction or attenuation is
given by
200 Basic Ultrasonic Imaging
reaches a position and angle, as shown in Fig. 9.13, where the beam is perpen-
dicular to the interface so that the specular reflection is received.
The use of higher frequencies has increased the ratio of diffuse to specular
echoes since the amplitude of the diffuse echoes increases as the square of the
frequency. However, these are still considerably lower in amplitude than the
specular echoes so that a large dynamic range is req.uired to them.
Following detection, most ultrasonic systems use nonlmear compreSSIOn, prior
to display, to compress the large specular echoes and enhance the weaker diffuse
echoes.
The array systems, discussed in the next chapter, use stationary transducer
arrays and thus do not involve compound scanning. However, if a large array
subtends a relatively large angle with the region of interest, it reduces its angular
sensitivity. One mechanical commercial scanner uses eight rotating transducers
immersed in a water bath. These separate views are added to provide the effect
of a compound scan and minimize the angular sensitivity of specular interfaces.
NOISE CONSIDERATIONS
Er = KP ff [sex, y)]2dxdy.
Using this measured value, the value of Eo is given by
Eo = If e-''''E, ISS R(x,y, zo)[s(x,y) ** ei(kr2/2zol]2dXdyl
[sex, y)]2dxdy
(9.67)
(9.68)
Unlike x-ray, with its signal-dependent Poisson noise, the noise in ultrasonic
systems is governed by additive Gaussian noise resulting from the transducer
and the first amplifier. The resultant signal-to-noise ratio is therefore the ratio
of the received signal power at the transducer terminals to the average noise
power e;. .'
In estimating the signal, it must be emphasIzed that the attenuation com-
pensation, defined in equation (9.3), is performed beyond the transducer and
thus does not generally affect the signal-to-noise ratio. Thus, as would be. ex-
pected, reflections from greater depths,which experience increased attenuatlOn,
result in a reduced signal-to-noise ratio.
The sicrnal-to-noise ratio at each depth Zo is defined as the peak signal
power at that depth divided by the noise power as given by
SNR = (9.64)
e;
where Eo is the peak value of eo(t), the signal envelope e(t) derived from depth
plane z = zoo This signal eo(t), in a single transducer system, using steady-state
diffraction theory, from equation (9.32) is given by
eo(l) = K \ e-;;"ft(1 - 0) SSR(x, y, zo)[s(x, y) *>e;{,,'n"'j'dxdy \. (9.65)
The peak value Eo is given by
Eo K
e
-;;" p \ SSR(x, y, z 0) [s(x, y) ** e;{"'!'''']'dxdy \ (9.66)
where P is the peak value of p(t),
This expression can be used for Eo in (9.61) to find the signal-to-noise ratio
based on experimentally measured values.
. Th.e integral expressions in (9.66) and (9.68) represent the product of the
patterns of the source and the reflectivity at plane zoo If the reflecti-
VIty functIOn R, representing the object being studied at plane z is small
t.he beam. size, the integration is essentially over R itself.
such as a lesion, is large compared to the beampattern,
the IntegratIOn IS effectIvely over the beam pattern and is independent of the
size of the object.
SPECKLE NOISE
The noise studied thus !s of electrical noise at the input of the system.
ultrasor:Ic Imagmg has substantial coherence properties. This
results In the mtroductIOn of a spatial noise component known as "speckle"
[Burckhardt, 1978]. The origin of this component is seen if we model our
function .an array of Because of the finite resolution, r-t
any tIme we are receIvmg from a dIstribution of scatterers within the resolutiqn
element. scattered signals add coherently; that is, they add constructively
and depending on the relative phases of each scattered waveform.
, !he. nOIse properties of speckle are based on the statistical nature of the
dIstrIbutIOn of the sum of sinusoids reflected from the randomly distributed
scatterers. The resultant phasors add in a random walk distribution. If there are
a large number of scatterers within each resolution element, and the received
PROBLEMS
Basic Ultrasonic Imaging 203
I
I I
I C2' P2. a2 I C3. P3. a3
I I
I I
I Z2
9.4 In an ultrasonic imaging system, two isotropic point scatterers are in the
y = each a distance Z from an L X L square transducer. We define
two Images as being separable if the points are separated by at least
the betwee,n .the first zeros of the impulse response.
(a) IS the ml.mmum separation of the points in the x direction to
prOVIde. separable Images where Z = Zl in the geometric near field and
z = Z2 m the far field?
(b) What is the transducer size L that will achieve the same minimum sepa-
ration of points in both the near and far fields at Z = Zl and Z2 ?
9.5 An ima.ging .has an additive noise power component *.
A reflectIng havIng a reflectivity Acirc (rl ro) o(z - zo) is ad-
WIth an CIrcular transducer of radius rt where r
t
> ro. The
me.dmm a umform attenuation ct. Calculate the ratio of the signal-to-
nOIse ratIo In the near field at Z = Z 1 to that of the far field where Z = Z
Assume geometric imaging for depth Zl and Fraunhofer behavior for depth
Z2'
FIG. P9.3
compensated for and that the effective pulse envelope is rectangular with a
period -r where c-r/2 < A.
(b) Repeat part (a) where 3A < c-r/2 < 4A.
In the reflectivity of the body we have assumed a weakly reflecting
where b?th the energy lost due to the reflection and multiple reflec-
tIons . be Assume a simple one-dimensional planar model
contammg two reflectmg surfaces which is modeled as
R(x, y, z) = R1o(z - Zl) +R2o(z - Z2)'
(a) Find the estimated reflectivity R(O, 0, z) using a perfect delta function
as the effective pulse envelope taking both the energy loss in reflection and
multiple into account. Ignore attenuation and calculate only the
first receIved multIple reflection.
(b) Find the value of Rusing R
1
= R
2
= 0.1. Compare this to R. using the
weakly reflecting assumption.
9.3 An ultrasonic pulse ,:ith aCt) = rect (tiT) is transmitted through
the ,shown m FI,g. P9.3. Plot the envelope of the received signal,
labelmg amplItudes and tImes. Use the weakly reflecting assumption and
neglect multiple reverberations.
(9.70)
(9.69)
E
SNR = /_
-"\/ 2 _ 2
9.1 A region of the body is modeled as a uniform array of scatterers. A fluid-
filled nonreflecting cyst is contained within the volume so that the reflectiv-
ity is given by
R(x,Y, z) = comb comb comb )(1 - rect x
2
+ y
2
(z- zoY),
(a) Plot the estimated reflectivity versus depth in the A mode using an
L x L transducer at the origin where L < A B. Assume geometric
imaging where diffraction is ignored. Assume that the attenuation has been
We use a fundamental property of Rayleigh probability distributions which
relates the mean of the envelope square to the square of the mean, as given by
2 = ..'2. (9.71)
n
Substituting this in (9.70) yields a signal-to-noise ratio of
SNR = ;E = (4 rcY/2 = 1.91. (9.72)
This relatively low ratio emphasizes the importance of this noise source. It often
overrides the system electrical noise and is especially visible in larger organs
consisting of uniformly distributed scatterers such as the liver. It is not yet well
understood how much this noise component contributes to reducing the diag-
nostic accuracy of the image.
One mechanism of reducing this speckle noise is the summation of a num-
ber of images of the same object, each with independent speckle patterns. This
will reduce the noise by the square root of the number of images. These inde-
pendent images can be obtained by acquiring the data from different angular
views. For example, compound scanning, illustrated in Fig. 9.13, reduces the
speckle noise by acquiring and combining views of the same region from
different angles.
where f2 is the average of the envelope squared. We define the signal-to-noise
ratio as the ratio of the mean of the envelope to its standard deviation as given
by
phases are uniformly distributed from to 2n radians, the envelope amplitude
E obeys a Rayleigh probability density function given by
2 (_2)
peE) = fi exp E2
202 Basic Ultrasonic Imaging
fNG ARRAYS
.. z
BODY
R(x,Y,z)
FIG. 10.1 Collimated imaging array.
Y
I
TRANSDUCER ARRAY
? a.re defined as transducer arrays which are in an image plane.
is III to the arrays we will subsequently consider where the array
a nommagmg plane and the various transducer signals are delayed and
ed to provide image information.
Abasic imaging array is shown in Fig. 10.1. The system shown uses near-
or collimated imaging. The individual transducers shown are fired in
sequence. Intially, we assume that we are operating within the near field of the
individual transducers so that the propagated wave is a geometric extension of
each transducer sn(x, y). Using rectangular transducers each individual source
is described by
six, y) = rect rect (Y hnd). (10./>
As shown in Fig. 10.1, each transducer is driven by the pulse pet) in a
sequence separated by time T. This time, as indicated in Chapter 9, must be
greater than the maximum round-trip time 2z
max
/ c, where Zmax is the maximum
depth: Each rece:ived subjected to time-varying gain and envelope
detected to proVIde a sIgnal IdentIcal to that of equation (9.4). These signals are
10
Ultrasonic Imaging
Using Arrays
The various systems described in Chapter 9 all used a single transducer that was
manually scanned to provide a two-dimensional image. These systems lack two
desirable characteristics: real-time imaging and dynamic focus. Although real-
time operation can be achieved by rapidly oscillating mechanical systems, it is
presumed that an electronic scanning approach, with a stationary transducer
array, is more desirable from the point of view of size, weight, and reliability.
Dynamic focus, to overcome some of the basic diffraction problems illustrated
in Figures 9.7,9.8, and 9.9, can be achieved only through electronically con-
trolled transducer arrays.
In our analysis we lean heavily on the results of Chapter 9 in developing
the impulse response due to diffraction and the various considerations of
attenuation, velocity, and reflectivity. We will consider all array configurations
that are presently used or whose use is being considered [Macovski, 1979].
R(O, y, z) = K IR(x, y, z) *** p (2:) [Sn(X, y) ** exp JI
evaluated at x = O. (10.5)
This expression is valid for essentially all depth ranges, within the fairly
broad limits of the Fresnel approximation.
A commercial imaging array is shown in Fig. 10.2 together with a typical
image of a fetal head.
ec(t) = K fff R(x, y, Z)Sn(X, y)p (t - nT - 2:) dxdydz I (10.2)
where, as in Chapter 9, ec(t) is the gain-compensated, envelope-detected signal
R(x, y, z) is the three-dimensional reflectivity of the object, and pet) is th'
received pulse as modified by the various linear parameters of the system
e
including propagation and filtering. '
This array is an attempt to provide a cross-sectional image of the reflectivity
in the x = 0 plane. When ec(t) is synchronously displayed, the resultant estimate
of the reflectivity is given by
(10.3)
As before, the triple asterisk refers to a three-dimensional convolution. Each
line in the reflectivity image is blurred by sn(x, y) in the lateral dimensions and
p(2z/c) in the depth dimension. Clearly a high-resolution image would require
a relatively short pulse and a large number of relatively small, closely spaced
transducers. However, as the transducers become smaller, our assumption of
collimated imaging, which ignores diffraction, becomes less and less accurate.
In considering diffraction we will use the steady-state approximation of
equation (9.21), which assumes that the pulse envelope arrives at each lateral
position at the same time. We make use of the entire diffraction analysis of
Chapter 9 and use the result of equation (9.34), the resultant signal using the
Fresnel approximation. Applying this to the collimated array, the compensated
detected signal is given by
ec(t) = K SSSR(x, y, Z{Sn(X, y) ** exp (i k;;)J
X p(t - nT - 2:) dxdydz I (10.4)
where r
2
= x
2
+ y2. The lateral response has been degraded by diffraction as
noted by convolution with the quadratic phase factor. The estimate of the
reflectivity, including diffraction, is given by
Imaging Arrays 207
The resultant signal produced is called a spin echo. A SImple analogy IS a num?er
of racing cars leaving a starting point at different speeds and slowly
"out of phase" after a time T. If their directions are reversed, after another tIme
T, they will again be "in phase" at the starting line. . .
The spin echo signal has an amplitude somewhat less than orIgl?al
FID. This loss is due to the spin-spin relaxation process whereby the
of each precessing nuclei with its neighbor causes some dephasing. ThIS IS a
random process, unlike the dephasing due to the nonuniform field, and cannot
equilibrium with the relaxation time T!. Therefore, if a 90 is applied
at a time t
b
after the 180 inverting excitation, the resultant FID sIgnal amplitude
is given by
DIGITAL SUBTRACTION RADIOGRAPHY
The principal motivation of this technique is the noninvasive study of vessels
[Ovitt et aI., 1978]. With invasive procedures where catheters are inserted into
vessels, large amounts of iodinated contrast material are present, resulting in
good vessel visualization, despite the intervening anatomical structures. In
noninvasive studies, however, the very low iodine concentration is insufficient
to be visualized in a normal radiograph. As a result, subtraction techniques are
employed to eliminate the intervening tissue so that the iodine visualization is
limited solely by noise. The subtraction technique involves obtaining data both
before and after the administration of the contrast agent, and subtracting the
result.
Let f.1r(x, y, z) be the attenuation coefficient of the tissue in the anatomical
region under study. Let f.1ix, y, z) be the attenuation coefficient distribution of
the administered contrast agent. The subtraction operation involves first taking
the logs of the measured intensities to derive the desired line integrals and then
subtracting. The measured intensities, assuming monoenergetic parallel x-rays
are
II = 10 exp [ - f f.1l x, y, z)dzJ (11.9)
12 = 10 'exp [ - f LuC<x,Y, z) + f.1cCx,y, z)]dz]. (11.10)
Subtracting the logs, we have
I n ( ~ : ) - I n ( ~ ~ ) = ff.1cCx,y,z)dz (11.11)
the desired projection image of the contrast agent alone.
The quality of this image is limited by two factors, motion and noise.
Various physiological motions, often involuntary, occur between II' the pre-
pontrast image, and 1
2
, the postcontrast image. This provides a motion "noise"
gomponent in the output image given by
(11.12)
Where t 1 and t 2 are the acquisition times of the images. This effect can be
minimized by storing a number of precontrast and/or postcontrast images and
finding a pair that provides acceptable tissue subtraction performance with a
small n
m