AAPM

Tutorial

Characteristics and Control of

Contrast in CT1
I Ed Barnes, PhD

Understanding how contrast is produced and controlled in computed to-

mography (CT) is essential to proper application of this modality. In the typi-
cal CT scanner, a narrow x-ray beam in the section thickness direction and
an air gap in the section plane are used to reduce scatter and improve con-
trast. High- and low-contrast detectability of a CT scanner are important per-
formance parameters contributing to optimal image quality. The limits of
detectability of high-contrast objects (ie, spatial resolution) are affected by
detector aperture size, pixel size of the image, algorithm used to reconstruct
the image, and section thickness. Visibility of low-contrast objects is limited
by image noise and the algorithm. Contrast in CT images can be controlled
by the window level and window width settings used to display the image.
These settings dictate how the actual measurements of tissue attenuation
are translated into a gray-scale image. Wide window widths can be used to
provide an accurate representation of bone, and narrow widths are more
useful for visualizing soft tissues.

INTRODUCTION
When computed tomography (CT) was brought into clinical use in the early 1970s,
it had a profound effect on medical imaging. With CT technology, the image quality
of tomography was significantly improved, and radiologists began to think in terms of
imaging “slices” or sections versus the projection images in use for many years. CT
also provided electronic display of images and complete control over the gray scale.
Dramatic improvement in imaging soft tissues with better contrast was also realized.
The characteristics of that contrast and its control are the topics of this article.

CONTRAST IN CT
In physics, contrast is defined as the difference in signal between an object and its
background divided by the signal from the background (1). In electronic imaging,
gray scale is assigned according to the size of the signal. Visually, contrast is the

Index terms: Computed tomography (CT), image display and recording Computed
#{149} tomography (Cl). physics
Physics

RadloGraphics 1992; 12:825-837

1 From the Medical Technology Management Institute, 9722 Watertown Plank Rd. Box 26337. Milwaukee, WI 53226.
From the 1991 RSNA scientific assembly. Received April 9, 1992; accepted April 15. Address reprint requests to the
author.
C R.SNA, 1992

825
 Without Scatter With Scatter
( Scatter Factor=1 ( Scatter Factor=4 )
-Contrast Level without Scatter

i7 0 80

L / 60
(1)
40
0
0
20

0
1
Contrast= 100%/i = 100% Contrast= iOO%/4 = 25% Amount of Scatter (S)
Figure 1. Diagram illustrates the effect of Figure 2. Graph depicts the variation of
scattered radiation on image contrast (see percentage of contrast with the amount of
text for detailed discussion). (Reprinted, scatter. (Reprinted, with permission, from
with permission, from reference 2.) reference 2.)

brightness of an object compared with its background. The signal measured in CT is

related to the amount of radiation detected. Assuming that the efficiency of detection
has no effect, the signal is proportional to the x rays absorbed and scattered by the
patient’s body.
One of the important parameters in defining image quality is signal-to-noise ratio.
The signal-to-noise ratio is a ratio of the amount of signal received divided by the
amount of noise in the image. The noise is created by statistical variation in x-ray
emission (ie, quantum noise) and any unwanted signal, such as detected scattered
radiation and spurious electronic signals. Another equally important parameter in
describing image quality is contrast-to-noise ratio. The contrast-to-noise ratio is the
contrast of an object divided by the image noise. The contrast-to-noise ratio controls
the detectability ofvarious-sized objects in an image. Maximizing the contrast-to-
noise ratio provides optimal image quality.

EFFECT OF To generate contrast in an imaging modality such as CT, there must be a significant
SCATFER ON difference between the signal of the object and that of its surroundings. This differ-
CONTRAST ence in signal is generated by two properties of the interaction of radiation with
matter: (a) absorption of radiation by the object (compared with the absorption by
its surrounding structures) and (b) scatter. Absorption of radiation (x-ray photons)
is by the photoelectric process. Scatter can be either coherent or incoherent.
With coherent scatter, the path of a photon changes without a change in enengy.
With incoherent scatter, some energy is transferred to the interacting electron, caus-
ing the direction of the photon to change and a reduction in its energy. Because the
detector in a CT scanner cannot distinguish a scattered photon from an unscattered
one, the scattered photon provides misinformation about the absorption of radia-
tion by the object.
Scatter is an undesirable radiation interaction for producing subject contrast, as is
illustrated in Figure 1. Without any scatter (ie, the imaged object absorbs 100% of
the penetrating radiation), the contrast would be the difference in signal received
by the image receptor under the object compared with the signal received by the
area adjacent to the object or a contrast of i (100%). If a scatter factor S is defined as
the ratio of radiation intensity with scatter compared with the intensity without scat-
ten, then the scatter factor would be 1 (no scatter).

826 U RadioGraphics U Barnes Volume 12 Number 4

 3. 4.
Figures 3, 4. (3) Drawing on the left shows a wide radiation field incident on the medium
surrounding an object (B) to be imaged. A considerable amount ofscatter is received by the
image receptor (A) below the object, and this reduces contrast. In the geometry on the right,
a narrower beam of radiation is used, and the amount of scatter incident on the image recep-
ton (C) beneath the object (D) is reduced. (4) Diagram shows how use of an air gap reduces
the amount of scattered radiation incident on the image receptor. (Reprinted, with penmis-
sion, from reference 2.)

However, if the penetrating radiation is scattered around the object by the sun-
rounding medium, the contrast will be reduced. In Figure 1, if the radiation inten-
sity with scatter increases by a factor of 4 (S = 4), the percentage of contrast de-
creases by a factor of4 and equals 25% (percentage ofcontrast = 100/S [2]). As can
be seen in Figure 2, increasing scatter has a profound effect on the contrast ob-
served in an image.
The amount of scatter in an image is related to the size of the radiation beam and
other factors, such as the thickness of the absorber. As can be seen in Figure 3, a
wide radiation field produces more scatter than a narrow field. Use of a narrow x-ray
beam will reduce scatter and improve image contrast. In CT, the x-ray beam is nan-
row in the section thickness direction, corresponding to the geometry on the right
in Figure 3. Within the plane of a section, however, a wider radiation beam is used
that generates scatter and reduces contrast unless this scatter is eliminated.
In conventional radiography, one method for reducing scatter is to use an air gap
on air grid. The physical separation between the object being imaged and the image
receptor causes some of the scatter to miss the image receptor and thereby not de-
grade contrast (Fig 4). This geometry is typical of that used in a CT scanner in the
section plane. Both of these techniques-the narrow beam used in the section
thickness direction and the air gap used in the section plane-neduce scatter and
improve contrast in CT scanning. This reduction in scatter plus the planar image
generated in CT produce the improved soft-tissue contrast seen in CT images corn-
pared with that observed in conventional radiographs. Manufacturers of CT scan-
ners also use absorbing plates and pins as collimating devices to reduce the amount
of scatter detected by the image receptor.

July 1992 Barnes U RadioGraphics U 827

 5. 6a. 6b.
Figures 5, 6. (5) Phantom for measuring low-contrast image performance. (6) CT images of
the phantom in Figure 5 were reconstructed with the typical soft-tissue algorithm (a) and
with an algorithm that provides a substantial amount of smoothing (b).

DEMONSTRAT- The ability to detect contrast in an image is generally separated into high-contrast
ING CONTRAST detectability and low-contrast detectability. The detectability of contrast can be
DEThCTABILITY demonstrated with an appropriately constructed object commonly referred to as a
phantom. A common phantom used to demonstrate low-contrast detectability con-
sists of a cylindrical plastic disk that is immersed in a water bath for imaging (Fig 5).
Each sector ofthe disk contains a pattern ofholes ofvarying diameter, and the thick-
ness of each sector is varied to create the desired object contrast between the water-
filled holes and the surrounding plastic. CT images of a phantom can be used to
provide a visual measurement of the detectability ofvarious-sized objects at a given
contrast (Fig 6).
Another common phantom is made of solid plastic with attenuation values equal
to those ofwater (Fig 7) The cylindrical phantom
. approximates the size of a typical
head. A ring of plastic with attenuation values equal to those of bone can be placed
around the phantom to simulate the absorption and scatter caused by the skull. A

828 U RadioGraphics U Barnes Volume 12 Number 4

 Figures 7-9. (7) Commer-
cial phantom for measuring
CT scanner performance.
(8) CT image of the phantom
in Figure 7, with a 6% con-
trast insert. (9) CT image of a
phantom with a high-contrast
insert. (Figs 7 and 8, courtesy
r ofRadiation Measurements,
Middleton,Wis.)

;. .. .

-. . - . . . I

‘ . , ::, .. #{149}#{149}:.#{149}
#{149}‘#{149}.#{149}

. .. .

‘: : . .

larger plastic ring with water attenuation values can also be placed around the phan-
tom to simulate the absorption and scatter caused by the body. Inserts that contain
structures of known low contrast can be placed in the plastic material. Figure 8
shows the phantom with an insert that simulates a structure with 6% contrast. Deter-
mination of the average CT numbers in the regions of interest located within the
structure and in the adjacent background allows calculation of image contrast pro-
duced by the CT scanner.
Figure 9 shows the phantom with a high-contrast insert. The insert consists of a
pattern ofholes ofdecreasing diameter, with the spacing between the holes equal
to the hole diameter. With such a pattern, the detectability of high-contrast objects
near the upper limit ofcontrast can be determined. This measurement is typically
referred to as the spatial resolution of the system. The smallest-diameter objects that
can be distinguished from each other at the limit of resolution of the system are
demonstrated in this image.

July 1992 Barnes U RadioGrapbics U 829

 2 3 4 6 810

a. b.
Figure 10. Contrast detail curves. The small rectangular boxes highlight the high-contrast
(a) and low-contrast (b) portions of the curves.

These demonstrations of detectability over the full range of object contrast can be
used to construct a contrast detail curve (Fig 10) (3), which is a plot of the percent-
age contrast of an object compared with the smallest diameter object visible at that
contrast. The high-contrast portion ofthese curves is an indication ofthe high-con-
trast object detectability, also referred to as the spatial resolution of the system.
Several factors affect the limit of detectability of high-contrast objects, including ap-
erture of the detector used to collect the CT data, pixel size in the image, the convo-
lution filter used to reconstruct the image, and section thickness. The low-contrast
portion of the curve demonstrates the contrast detectability at the low-contrast lim-
its of the system. The visibility oflow-contrast objects is limited by image noise and
the convolution filter used to reconstruct the image.
Both the high-contrast spatial resolution and the low-contrast detectability of a CT
scanner are important performance parameters contributing to optimal image qual-
ity, as demonstrated with clinical examples in Figures 1 1 and 12. Images of an ab-
dominal plane obtained at two different doses illustrate the effect of noise on low-
contrast detectability (Fig 1 1). The radiation dose used to obtain the images differed
by a factor of approximately 10. The difference in quantum noise is readily apparent
in these two images, and the visibility oflow-contrast structures in the liver is notice-
ably degraded in the noisier image. A similar factor-of-lO difference in radiation
dose was used to image the chest (Fig 12), but imaging the high-contrast air pas-
sages in the lungs was not influenced by the resulting variation in noise. The addi-
tional noise present in the lower-dose scan did not have a substantial effect on ob-
ject detectability compared with that in the higher-dose, lower-noise scan.

830 U RadioGraphics U Barnes Volume 12 Number 4

 12a.
Figures 11, 12. (11) CT scans ofthe abdomen obtained at 20 mM for a radiation dose of 2
mGy (0.2 rad) (a) and 240 mM for a dose of 22 mGy (2.2 rad) (b) illustrate how noise at the
lower dose degrades low-contrast detectability. (12) CT scans ofthe chest obtained at 40 mM
for a radiation dose of 3.6 mGy (0.36 rad) (a) and 400 mM for a dose of 36 mGy (3.6 rad)
(b) illustrate how high-contrast detectability is not affected by noise.

Contrast detail curves can be used to illustrate the impact of selected parameters
on the overall object detectability ofvarious CT scanner configurations (4). For ex-
ample, the GE 8800 scanner (GE Medical Systems, Milwaukee) has a larger number
of detectors of substantially smaller aperture than the older model GE 7800 unit.
This change in scanner configuration improved the high-contrast spatial resolution,
as is readily apparent from the high-contrast portions of the curves shown in Figure
13. Consequently, smaller high-contrast objects can be detected in images obtained
with a GE 8800 unit compared with those detectable in images from a GE 7800 unit.

July 1992 Barnes U RadioGraphic’s U 831

 100
Contrast (%) Contrast Dst.iI Curv#{149}

100-

50-
10

C
0 20
C-)
C) 10

0
5

0.1 2
0.5 1 2 4 10 20
Object Size (mm)
Figure 13. Contrast detail curves for two
CT scanners (models 7800 and 8800; GE 0.5
Medical Systems) illustrate the effect of de-
tector aperture (the GE 8800 scanner had
detectors with smaller apertures compared 0.2

with those of the GE 7800 unit). Obiect

I t I I I diameter
0.2 0.5 1 2 5 10 (mm)

Figure 14. Contrast detail curves illustrate

effect ofconvolution filter algorithms. Two
levels ofracliation dose are also shown. FC2 =

smoothing convolution filter. FC3 = edge-

enhancing, high-detail convolution filter.

The effect ofradiation dose is also illustrated in Figure 13. Improvement in low-
contrast detectability with use of the higher dose (50 mGy [5 rad]) is evident, with
both scanners yielding a very similar performance. The lower dose of 12.5 mGy
(1.25 rad) resulted in degraded low-contrast detectability produced by a higher
noise level, with the effect being similar both scanners.
for This variation in noise,
however, had a negligible effect on the high-contrast spatial resolution
(detectabil-
ity) of each scanner, as expected.
Contrast detail curves are also affected by the convolution filter algorithm used to
reconstruct the image. In Figure 14, two different convolution filters were used to
produce the contrast detail curves: an edge-enhancing, high-detail filter and a
smoothing filter. The high-detail convolution filter improved the high-contrast spa-
tial resolution measurement. As expected, the variation in radiation dose had a neg-
ligible effect on the high-contrast detectability for the smoothing algorithm. At the
low-contrast end of these curves, the smoothing algorithm and, as expected, the
higher radiation dose both resulted in an improved low-contrast detectability.

CONTRAST The relative amount ofradiation attenuated by each volume element (voxel) in an
CONTROL object is represented by the CT number (5). The CT number is defined as the differ-
ence in attenuation of the contents of a voxel relative to water:

CT number = [(ji5 - p)] F,

where p.5 = attenuation of the structure and p, attenuation =

ofwater (water is
used as the reference material to construct a CT image). A scaling factor F is used to
define the scale of CT numbers over the range of attenuation values encountered in
the body. This is commonly referred to as the Hounsfield scale.

832 U Ra4ioGrapbics U Barnes Volume 12 Number 4

 CT Numbers Hounsfield Scale
alr=-1000 water=O
I )IL I I I
-1000 0 2000 3095

(“;“ I
-100 -80 -60 -40 -20 0 20 40 60 80 100

Fatty Tissues BlOOd Clot Figure 15. Illustration of

CSF D Tissue the Hounsfield scale and im-
White Gray age display gray scale. The
Image Display typical range ofCT numbers
0 8 bIts of gray scale 28=256 levels of gray- 255 for variousCSF soft cerebrospinal
shown. =tissues is
- I fluid.

.#{149}- 500-.- Figure 16. Graph depicts

the mapping ofthe gray scale
onto the Hounsfield scale by
setting the window level and
window width.

In the Hounsfield scale, water is given a value ofzeno and air a value of 1,000. - A
12-bit number is used to define the scale. Since 212 equals 4,096, this corresponds
to a CT number scale from 1,000 for air - (lung) to +3,095 for the densest object
that can be measured by the CT scanner (compact bone). As illustrated in Figure 15,
the soft tissues of most clinical interest are contained within a CT number range of
-lOOto +100.
Although the CT scanner is capable of dividing its measurement of tissue attenua-
tion into a range of 4,096 CT numbers, the eye is not capable of distinguishing this
much detail in an image. The image display ofa CT scanner represents only 256 1ev-
els of gray, which must therefore be mapped onto the portion of the Hounsfield
scale that is to be displayed. Adjustments called window level and window width are
used to define this mapping. Selection of the window level specifies the CT number
for centering the gray scale, and choice of the window width defines the range of CT
numbers over which the gray scale is to extend. These adjustments can be thought
ofas defining the “slope” ofthe gray scale. The correlation between the gray scale
and Hounsfield scale is shown in Figure 16. When the center of the gray scale is
placed at a window level of 100 and the window width is set at 500, the gray scale
permits display ofCT numbers from - 150 to +350. All CT numbers below the
lower limit of the window width are displayed as black, and all those above the up-
per limit are displayed as white in the image.

July 1992 Barnes U RadioGraphics U 833

 Figure 17 Gray-scale mapping with the 0 1000 1500
same window level but with wide and nar- 1
row window widths. Each gray scale is cen- I
tered at a CT number of 0 corresponding to
water.

Figure 18. CT scans of the

head displayed with the same
window level of 40 and dif-
ferent window widths. (a) CT
scan displayed with a narrow
window width of 50 demon-
strates fine details in the soft
tissues of the brain. (b) CT
scan displayed with a wide
window width of 2,500 de-
picts bony structure of the
skull more accurately than a
but provides little information
about soft tissues.

Figures 17 and 18 illustrate the use of a wide window width versus a narrow win-
dow width when the window level is kept the same. As seen in the clinical images, a
narrow-window image provides the most useful representation of soft-tissue struc-
tune, while the wide-window image provides a more accurate representation of
bone (Fig 18).
Figures 19 and 20 illustrate the use ofdifferent window levels with the same win-
dow width. As seen in the clinical images, the selection of the window level will de-
termine how well the soft tissues are displayed. In Figure 20a, a window level of 40
provides a useful clinical image. With a window level positioned at - 10 (Fig 20b),
only the fluid-filled ventricles of the brain are displayed. This window level therefore
has limited clinical usefulness in this situation.
The gray scale can be applied more than once to the image display, representing
different window widths and levels. This is known as dual-window display (Fig 21).
Dual-window scales can also be overlapping and combine both wide and narrow
window widths in one display. With this capability, the technologist can select a
wide window width for one assignment ofgray scale to define the bony structure of
the skull while using a standard window width and level to display the soft-tissue
structures of the brain (Fig 22). This provides an accurate representation of both the
bone and soft tissue in one image. The technologist can also combine a standard
window width and level for viewing the brain with a narrow window width to high-
light the structures filled with cerebrospinal fluid. These dual-window displays,
however, have not found wide acceptance for routine clinical use.

834 U Ra4ioGrapbics U Barnes Volume 12 Number 4

 Figure 19. Gray-scale mapping with dii-
fenent window levels but the same window
width. Here, the same slope of the gray scale
is used, but it is centered at different points
on the Hounsfield scale.

Figure 20. CT scans of the

head displayed with the same
window width of8O but dii-
ferent window levels. (a) CT
scan displayed with a window
level of 40 is useful for dem-
onstrating soft tissues of the
brain. 0’) CT scan displayed
with a window level of -10
shows only fluid-filled ventri-
des.

2000

Figure 21. Diagram shows two examples

ofgray-scale mapping with dual windows
80 and levels. Dual windows are applied to
J different portions ofthe Hounsfield scale.
. Wide window widths (top) and narrow
. window widths (bottom) are illustrated.

a. b.
Figure 22.CT scans ofthe head displayed with dual win-
dows. (a) CT scan displayed with wide window width
(2,000; window level, 1,000) and standard window width
and level (80 and 40, respectively) shows both bony and
soft-tissue structures of the head. (b) CT scan displayed with
standard window width and level (80 and 40, respectively)
and with narrow window width (10; window level, 10)
shows the fluid-filled structures.

July 1992 Barnes U RadioGraphics U 835

 Figure 23. CT scans of the chest with win-
dow widths and levels set to display detail in
the lungs (a), soft tissues (b), and bones (c).
a.

Adjustments ofwindow width and level can be useful in imaging various struc-
tures of clinical interest in the chest (Fig 23). In Figure 23a, a window level centered
around air is used to bring out the bronchioles of the lung. In Figure 23b, a window
level centered on soft tissue is used to bring out detail in the mediastinum and vas-
cular structures near the heart. In Figure 23c, a window level appropriate for dis-
playing the bony structures of the ribs and spine has been selected. A wide window
width has been used in this image, and the vascular structures within the mediasti-
num can also be appreciated.
The contrast in a CT image can also be controlled through the use of contrast
agents. These agents typically are injected and use an iodinated material to increase
x-ray absorption in the blood, which improves visualization of the vascular struc-
tures in an image (Fig 24).

836 U Ra4ioGrapbics U Barnes Volume 12 Number 4

 a. b.
Figure 24. CT scans obtained in a plane through the liver without contrast material (a) and
after injection ofcontrast material (b). The enhanced contrast ofblood assists in visualizing
the vascular structures in the liver.

The CT scanner has had a dramatic impact on our ability to image the human body. SUMMARY
It provides improved capabilities to control the contrast in an image that can be
used to significant diagnostic advantage. Window width and level controls must be
understood and used prudently to ensure that all useful diagnostic information is
displayed and used in image interpretation. Understanding the mechanisms of con-
trast production in CT is important to obtain better images, control the contrast for
each application, and improve the accuracy of diagnosis.

Acknowledgments: Appreciation is expressed to RobertJ. Wilson, PhD, Dennis W. Foley,

MD, Stephen Balter, PhD, Bryan R. Westerman, PhD, and Joel Schenk, RTR, for assistance in
preparing and acquiring the illustrations for this manuscript. Many thanks to the members of
the 1991 RSNA/AAPM Physics Mini Symposium Committee and to the symposium director,
Perry Sprawls, PhD, for constructive criticism and suggestions for improving this work. Appre-
ciation is also expressed to Tara A. SaIakS and Sheila D. Kirschbaum, BA, for assistance in pre-
paring the manuscript and to Dian Barnes for lifelong inspiration.

1. Zatz LM. Basic principles of computed 4. Cohen G, DiBianca FA. Information REFERENCES
tomography scanning. In: Newton TH, content and dose efficiency of computed
Potts DG, eds. Radiology of the skull and tomography scanners. In: Haus AG, ed.
brain: technical aspects ofcomputed to. The physics of medical imaging: record-
mography. St Louis: Mosby, 1981; 3853- ing system measurements and tech-
3876. niques. New York: American Institute of
2. Sprawls P. Physical principles of medi- Physics, 1979; 356-365.
cal imaging. Gaithersburg, Md: Aspen, 5. McCullough EC. CT imaging parame-
1987; 179-193. ters. In: Fullerton GD, ZagzebskiJA, eds.
3. Cohen G, DiBianca FA. The use of con- Medical physics ofCT and ultrasound:
trast-detail-dose evaluation of image tissue imaging and characteristics. New
quality in a computed tomography scan- York: American Institute of Physics,
ner. J Comput Assist Tomogr 1979; 1980; 70-88.
3:189-195.

July 1992 Barnes U RadioGrapbic.s U 837