‘U.’
U
AAPM
‘I
Tutorial
,v.
V
CT Image Perry
Sprawls,
Two
important
V
blur age
V
:
characteristics
of the
to visualize
and noise. detail);
jects.
Increased
increased ofblurring
in CT
by the
and
reduces
noise
be regulated
tomographic
structures
blurring visual
can
of the voxels,
computed
anatomic
Sources
(which
Noise1
PhD
the ability
affect
and
Detail
the
include
the
spot
the reconstruction
pathologic
the visibility
reduces focal
and
(CT)
size
size
filter
objects
sampling
the
selected.
are
(im-
of low-contrast
ofthe
and
that
features
of small
visibility
image
aperture
detector
Noise
ob-
size),
is caused
the
size
by the
variation in attenuation coefficients between voxels. Use of small voxels and edge-enhancing filters helps reduce blurring and improve visibility of fine details. However, small voxels absorb fewer photons and therefore result in increased noise. Noise can be reduced by using large voxels, increasing radiation dose, or using a smoothing filter, but this filter increases blurring. An optimized protocol for a specific clinical study must take these physical principles into account and be adjusted to give proper balance among detail, ;V
;
low noise,
and
patient
exposure.
INTRODUCTION Computed tomography (CT) is characterized makes it possible to depict the differences ever, CT has two other characteristics that structures
and
pathologic
of small
bility
objects
low-contrast This
features:
(image
by a high-contrast among
soft
affect
sensitivity
tissues
within
the ability
that
the
to visualize
body.
How-
anatomic
(a) blurring of the image, which affects the visiand (b) visual noise, which reduces visibility of
detail),
objects. article
shows
how
visibility
the sources ofblurring operator to optimize image
is influenced
by
these
two
characteristics,
identi-
and noise, and shows how they can be controlled quality for different clinical requirements.
fies
CHARACTERISTICS two characteristics that In CT, inherent contrast is the extent
by the
LESION Lesions
have
lesion
differs
that
differ
Index
terms:
from
1
From 1991
1992;
the
Department RSNA
6. Address ©RSNA,
from
Computed
RadioGraphics
the
that
greatly
of the
surrounding
of surrounding
(CT),
image
their
to which
those
tomography
visibility:
tissue.
quality
inherent
the attenuation tissue,
#{149} Computed
contrast
and
coefficient
value
If a lesion has attenuation it has high contrast and
tomography,
physics
#{149} Images,
quality
size.
of a
values will be
#{149} Physics
12: 1041-1046 ofRadiology,
scientific
reprint
affect
requests
assembly. to the
Emory Received
University May
28,
School 1992;
ofMedicine, revision
requested
1364 June
Clifton 9 and
Rd NE, Atlanta, received
July
GA 30322. 2; accepted
From July
author.
1992
1041
Figure
1.
matrix herent
Potential lesions are arranged in a according to their incontrast and size.
Figure
2.
(objects)
The
matrix
of po-
tential lesions shown in Figure 1 is depicted here with a curtain
of invisibility,
covers
the small
lesion
ner,
in the
which
low-contrast
upper
lesions.
Because
blurring ponents lesions
cor-
right
as well as many
other
of image
and noise (the cornofthis curtain), some in the body will not be
visible.
visualized. Generally, the easily
detail,
specifically
Physical size is also thought smaller the object, the greater
anatomic
detail.
Figure 1 illustrates how inherent contrast see lesions. The large, high-contrast lesion see. The challenge would be to detect the at the upper right. IMAGE
QUALITY
Each time an imaging procedure tential lesions that are not visible. lesions form
(Fig
in the body that of a contrast detail
of in terms of detail to be discerned. the capabilities must be for detecting and size interact and affect our ability in the lower left corner is very easy to lesion that is both small and low contrast
is conducted, there There is a distinct
to
is actually a large range of podividing line between those
are visible and those that are invisible. This boundary is one curve, which can be thought of as a “curtain of invisibility”
2).
We are generally concerned with image quality and improving image quality, which entails pushing the curtain of invisibility back so that more potential lesions or other objects are revealed. By changing certain imaging factors, we can raise the curtain and increase visibility. In doing this, however, we must be aware of other compromises in the imaging process. Two other specific characteristics of an image besides
contrast
oflow-contrast
anatomic
1042
U
RadioGrapbics
U
have objects,
detail
Sprawls
(Fig
an impact on visibility: (a) visual and (b) blurring, which affects 2).
noise, which affects visibility visibility of small objects and
Volume
12
Number
5
Focal Spot Size
Motion
I
III II
+
+
Sampling Aperture
Filter Smoothing
tI/-#{149}.. LII + LII I Ill
+
L1J.i Matrix
Detail
LI
-
V
I
Voxel Size
I Ill
Image
Figure
3.
(Blurring)
principles
imaging.
Aperture
pen,
fine
print,
noise
ing
procedures
from and
and
some
level
other
the
protocol
suffer
blurring.
In human
limits
our ability
which
Operators
of blurring
factors.
from
of blurring,
details.
amount
and
users
by selecting
In general,
as we
of CT can
appropriate
set up
the level of noise and the level lesions that will be visible.
As blurring
the
increases,
there
visibility
ofdetail
visibility
are specific
of the
sources
is limited
in
smaller
of image
by blurring
from
three
we
occasion-
to see small
vision,
objects,
control
the for
for a specific
objects
of imag-
clinical
proce-
We are affecting
is reduced. the
amount
various
In each
In conventional
sources:
As-
Md:
1987.)
of blurring.
blurring.
of medical
Rockville,
values
a protocol
dure, we are adjusting the range of potential
modality,
of detail
image.
physical
All image-forming
illusof blurring
FOV = field of view. (Reprinted, with permission, from Sprawls P. The a CT
FOV
Detector
ally suffer
Diagram
trates the sources that limit visibility
focal
imaging
Blurring
radiography, spot
size,
the
of receptor, and motion. Unlike radiography, which is affected during image acquisition, CT can be affected in the scanning stage reconstruction stage (Fig 3). During the scanning step of the image-forming process, blurring
by blurring and in the
the finite
size of the radia-
type
size of the focal
tion
detector,
tors.
For example, spot on one
focal
sectional
similar
(ie,
the
ray)
the object blurred.
in principle
of a ray,
from
both
is wider
but
also
than
the
In some CT scanners, would select the small user
can
select
(as in radiography) to the
and
blurring
is produced
by the finite
associated
with
radiographic
by recep-
a ray is the segment of an x-ray beam defined by the size of the end and by the size ofthe detector on the other end. The cross-
dimension
on the distance
spot
only image
and
often
referred
the focal the
object
surrounding
the user focal spot
change
the
to as the
spot
and
sampling
the detector.
being
imaged
tissue),
the
(ie,
image
If the the
ray
effective
size
of the
detector;
depends aperature
is “seeing”
ofthis
can choose the focal spot to give the greatest detail.
aperature, sampling
small
object
not
just
will
be
size and, obviously, In a few scanners, again,
for
the
maximum
detail, a relatively small detector aperture should be used. In the image reconstruction process, the section of tissue that was imaged is subdivided into individual tissue voxels, each ofwhich has a dimension. All structures within
ber. within achieve
September
an individual
In principle,
voxel
a voxel
are
mixed
together
is a three-dimensional
a voxel; when we look at images, we are an image with high detail, it is necessary
1992
and
represented
blur.
We cannot
seeing voxels to use small
by a single
CT num-
see any detail side by side. voxels.
To
Sprawls
U
RadioGrapbk.s
U
1043
FOV
(mm)
Voxel
+
Figure trates vidual
three
4. how voxel
factors:
Diagram illusthe size of an mdiis determined
by
field
of view size, and sec-
(FOV), matrix tion thickness. (Reprinted, with permission, from Sprawls P. The physical primciples of medical imaging. Rockville, Md: Aspen, 1987.)
5.
4
Matrix Size (voxels) (64, 128, 256, 512)
d=
L+
Matrix
Size
6.
Figures 5, 6. (5) CT scam with a moderate level of image noise demonstrates mires within the liver. (6) CT scan with a higher level of noise demonstrates duces the visibility of the vessels.
vascular
how
There are three protocol factors that can be adjusted that control voxel image detail: field ofview, matrix size, and section thickness (Fig 4). The mension of a voxel (sometimes erroneously referred to as the dimension
the ratio of the field ofview are imaging) to the matrix size (the pixel)
image). instances,
is
The
other dimension will be the largest
re-
size
and
face diof the
(the actual size of the anatomic region number of voxels across the dimension
of the voxel is the dimension. Section
struc-
noise
that we of the
section thickness, which, in most thickness is typically adjusted from
about 1 to 10 mm. The final factor that determines the amount of blurring produced in a CT image is the reconstruction filter. Reconstruction filters are mathematical operations that are used to alter characteristics of the image. In certain cases, it is desirable to use a socalled smoothing filter, which, in principle, is a blurring filter. In summary, the total blurring in an image that limits our ability to see the small features, potential lesions, or other anatomic structures has three basic sources. During scanning, the amount of blurring is determined by the sampling aperture (regulated primarily by focal spot size and detector size). At the time of reconstruction, blurring is determined by the size of the voxel and the type of filter. We need to recognize the characteristics ofblurring. Ifwe introduce a large amount of blurring at one point, there is no way to counteract it by decreasing the blurring at other points in the process. A chain is no stronger than its weakest link.
1044
U
RadioGraphics
U
Sprawls
Volume
12
Number
5
Figure
7. Diagram illushow noise results bethe same tissue can pro-
trates cause
duce
different
Note
that
CT numbers.
the
pixel
in the
right corner
lower than
the
Figure
other
8.
tamer
is brighter
one.
CT scan
ofwater
of a con-
in which
a cir-
cular area of interest is mdicated. Values for the mean .
.
VVVV
VVVVV.V:VV
(ME)
and
(SD)
ofthe
standard
deviation
CT numbers
are
shown.
L
‘
:
V;
h.
t
/
;
r
‘
‘
; ;;
‘
‘
Visual noise affects visibility of low-contrast objects. In an image with low noise, we can see more low-contrast lesions. As the image noise increases, fewer and fewer low-contrast objects are visible (Figs 5, 6). Noise is caused by the variation in attenuation coefficients between individual
voxels.
In CT, we are measuring
the attenuation
coefficients
of the individual
Noise
tissue
voxels. Ifwe have two voxels ofidentical tissue, we would expect to measure identical attenuation coefficient values, and, when these are translated into CT numbers, we would expect to get similar numbers. In reality we do not. In Figure 7, two voxels of the same tissue produce different CT numbers. This is a statistical variation that is visualized as image noise. One way of evaluating noise is to produce a CT im-
age of a container of water. Water zero. However, a CT scan ofwater the
noise Noise
pixels
(Fig can
within
and
systems
and
quantified.
the
calculating
of interest
way
the
ofx-ray
photons
reduce
the
September
of of
and
do
a statistical
analysis
ofthe
number
of
the
of defining
amount
of noise a region
in CT
of interest
images.
All CT
and
measuring
deviation.
we must measure the attenuation coefficient A more precise measurement can be made
photons
CT number
and
capability
standard
in a voxel,
1992
count
of quantifying
software
To decrease noise, voxel more precisely. more
CT number because
the various CT numbers, the resultant statistiwould most likely be a bell-shaped curve. The amount of spread or be evaluated by determining the standard deviation, which is the easi-
excellent
have
number
Ifwe
a region
an
and has a defined in pixel brightness
8). be
cal distribution deviation can est
is homogeneous reveals variations
absorbed
in each
voxel
the more
precise
the measurement
error
that
appears
in the
during image
the
imaging
values for each by increasing the process.
will be, and
The
this will
as noise.
Sprawls
U
RadioGraphics
U
1045
Figure
9.
trates
how
reduced quantity sorbed
Diagram CT image
illusnoise
by increasing
the
photons abtissue voxel.
ofx-ray by each
This is determined posure size.
(MAS)
by the exand the voxel
The quantity ing
the
size
is
ofradiation
of the
photons
voxel
and
the
absorbed
exposure
can be controlled
(Fig
9).
Obviously,
in two ways: a large
voxel
will
cept and absorb more photons than a small voxel and will result in decreased in the image. The other way of controlling the quantity of absorbed radiation
adjustinter-
noise pho-
tons per voxel is by increasing the total radiation imparted to the patient. This is controlled through the milliampere seconds. By increasing the milliampere seconds, we deposit more photons in each voxel of tissue, which results in less statistical vanation, therefore less noise in the image.
The mathematical used
to control
teristics.
The
crease
noise
visibility
filters noise.
Many
smoothing
in the image
CT systems
filters
by using
of detail.
selected are
a smoothing
The
so-called
used
have
reconstruction
10 or i2 filters
to reduce
filter,
process
we
noise.
increase
edge-enhancing
or
the
with
can also be
different
Unfortunately,
blurring
characas we
and
detail-enhancing
de-
decrease filter
has
essen-
be selected when the primay goal is visualization of small detail. CT equipment manufacturers generally recommend which filters are useful for specific clinical procedures. Unfortunately, noise is a form of image detail. Anything we do to sharpen an image and increase detail, such as use of small voxels or edge-enhancing filters, will tially
also CONCLUSION
an
opposite
characteristic.
visibility
increase
A filter
of this
type
might
of noise.
We are concerned with the visibility ofobjects within the body, primarily lesions. As we consider lesions with respect to their size and inherent contrast, we have found that blurring of the image from several sources limits the visibility of the smaller objects and that visual noise limits visibility of the low-contrast objects. We can improve visibility of detail by using small voxels. However, when we use small voxels to push up the curtain of invisibility on the right and increase visibility of detail, the curtain will drop down on the left because those small voxels increase the noise (Fig 2). In controlling image quality, we must make compromises. An optimized protocol for a specific clinical study must take these physical principies into consideration and be adjusted to give a proper balance among detail (visibility of small posure.
SUGGESTED
READINGS
TH, Potts ofcomputed
Newton
pects
Mosby,
1046
U
RadioGraphics
U
objects),
1981.
Sprawls
low
DG, eds. tomography.
noise
(visibility
Technical
as-
St Louis:
oflow-contrast
objects),
and
patient
ex-
Sprawls P. The physical principles of medical imaging. Rockville, Md: Aspen,
1987.
Volume
12
Number
5