Principles of CT and CT Technology*

Lee W. Goldman

Department of Radiation Therapy and Medical Physics, Hartford Hospital, Hartford, Connecticut

Inefficient x-Ray Absorption

This article provides a review of the basic principles of CT within Before the introduction of rare earth–intensifying screens
the context of the evolution of CT. Modern CT technology can be 20–25 y ago, the x-ray absorption efficiency of typical par-
understood as a natural progression of improvements and in- speed/calcium tungstate film/screen cassettes was only about
novations in response to both engineering problems and clinical
25%. Thus, 75% of the available x-ray beam as well as 75%
requirements. Detailed discussions of multislice CT, CT image
quality evaluation, and radiation doses in CT will be presented
of the information was wasted.
in upcoming articles in this series.
High Scatter–to–Primary x-Ray Ratios
Key Words: CT; CT technology; imaging; reconstruction
Because of large beam areas, scattered photons repre-
J Nucl Med Technol 2007; 35:115–128 sented 50% or more of the x-rays absorbed by the screens,
DOI: 10.2967/jnmt.107.042978
even with a grid able to remove high levels of scatter. Scat-
ter effectively reduces subject contrast by creating a back-
ground intensity unrelated to the overlying anatomy. The

T he use of CT in nuclear medicine imaging has been

growing, first with the introduction of PET combined with
amount of lost subject contrast is given by the contrast re-
duction factor, 1/[1 1 (S/P)], where S and P are the scatter
and the primary x-ray intensities at the receptor, respec-
CT (PET/CT) and, more recently, with the introduction tively (1). If 50% of the detected x-rays are scatter (so that
SPECT combined with CT (SPECT/CT). In addition, some S 5 P), then subject contrast is reduced by a contrast re-
of the basic ideas underlying CT, such as reconstruction duction factor of 0.5.
from projections (i.e., data measured at many positions and
angles), are very similar to those underlying the nuclear med- Superimposition and Conspicuity
icine cross-sectional imaging modalities PET and SPECT. Conspicuity is the ease of finding an image feature dur-
This article is the first of 3 to deal with the principles of CT ing a visual search. A feature may be visible if one knows
and CT technology. This first article covers the fundamental where to look but may be missed—that is, it may be
principles of CT, including the basic geometry of the CT inconspicuous—if the image is complex. Radiography ren-
scan process, the nature of the measurements made by CT ders a 3-dimensional volume onto a 2-dimensional image;
detectors, a qualitative explanation of the image reconstruc- as a consequence, over- and underlying tissues and struc-
tion process, the evolution of CT technology (the 4 genera- tures are superimposed, generally resulting in reduced con-
tions of CT) from the EMI first-generation scanner through spicuity as well as subject contrast.
modern slip ring designs, and the process of helical CT. The
next 2 articles will cover image quality and radiation doses Receptor Contrast Versus Latitude
in CT and multislice CT. Clinically useful radiographic films must provide suffi-
cient exposure latitude (i.e., a range of exposures yielding
clinically acceptable film densities) to record as much of
LIMITATIONS OF CONVENTIONAL RADIOGRAPHY the range of x-ray intensities exiting the patient as possible;
Film/screen radiography has several drawbacks that limit this feature necessarily limits receptor contrast. For exam-
its ability to visualize low-contrast tissues and structures with ple, for a typical radiographic film with an average gradient
acceptable levels of patient radiation exposure. These limi- (i.e., film contrast) of 2.5, an intensity (I) difference of
tations include the following. 1.0% would yield a film optical density (OD) difference of

OD2 2 OD1 5 2:5 · logðI2 =I1 Þ 5 2:5 · logð1:01Þ

Received Apr. 23, 2007; revision accepted May 7, 2007.  0:01 OD
For correspondence or reprints contact: Lee W. Goldman, Department of
Radiation Therapy and Medical Physics, Hartford Hospital, 80 Seymour St.,
Hartford CT 06102.
E-mail: [email protected]
and (in the usual absence of well-defined edges) would not
*NOTE: FOR CE CREDIT, YOU CAN ACCESS THIS ACTIVITY THROUGH be visible.
THE SNM WEB SITE (http://www.snm.org/ce_online) THROUGH SEPTEMBER
2009.
Although improved x-ray absorption efficiency is available
COPYRIGHT ª 2007 by the Society of Nuclear Medicine, Inc. with modern radiographic technology and digital radiography

PRINCIPLES OF CT AND CT TECHNOLOGY • Goldman 115

has nearly eliminated the contrast–latitude trade-off, most of
these limitations of radiography still exist today.
These issues were recognized long before the develop-
ment of CT, which led investigators to consider improve-
ments. One such innovation was conventional (focal-plane)
tomography, described by Bocage (2) in 1921. Through si-
multaneous movement of the x-ray tube and the receptor
around a fulcrum, tomography improved conspicuity by blur-
ring over- and underlying structures.
Both conspicuity and contrast could be improved if irra-
diation and visualization were limited to individual cross-
sectional imaged slices, which could be displayed as
2-dimensional images without significant structure overlap.
Because the radiation source could be collimated to thin
(;1-cm) slices, significantly less scatter would be gener-
ated (because less tissue would be irradiated at any given
time) as well. One way to achieve this goal is the process of
reconstruction from projections. Interestingly, the theory of
image reconstruction from projections, which is central to
the basic concept of CT, was described in 1917 and was
proposed for medical imaging as early as 1940 (3).
The development of the first modern CT scanner was be-
gun in 1967 by Godfrey Hounsfield, an engineer at British FIGURE 1. CT arrangement. Axial slice through patient is
EMI Corp. Hounsfield was interested in situations in which swept out by narrow (pencil-width) x-ray beam as linked x-ray
large amounts of potential information may be inefficiently tube–detector apparatus scans across patient in linear trans-
used—an accurate description of conventional radiography lation. Translations are repeated at many angles. Thickness of
for the reasons cited earlier (4,5). He estimated that by tak- narrow beam is equivalent to slice thickness.
ing careful measurements of x-ray transmission through a
subject at many positions across the subject and at a suf-
ficient number of angles, it should be possible to determine subject corresponding to each measurement is called a ray.
attenuation differences of 0.5%—possibly sufficient to dis- The set of measurements made during the translation and
tinguish between soft tissues. After verification of his hy- their associated rays is a view. Hounsfield’s Mark I scanner
pothesis with a laboratory apparatus, the first clinical CT measured the transmission of 160 rays per view. The cor-
scanner was built and installed at Atkinson-Morley Hospi- responding number of measurements for today’s scanners is
tal in England in September 1971. To understand the basic typically over 750. After completion of the translation, the
principles of CT, one may begin with the operation of this tube–detector assembly is rotated around the subject by 1,
original EMI Mark I first-generation CT scanner. and the translation is repeated to collect a second view. If
the first translation is obtained with the tube above the
subject and the detector below (0), then the second trans-
BASIC PRINCIPLES OF CT: FIRST GENERATION OF CT lation is obtained with the tube–detector assembly at 1.
Hounsfield imagined the subject to be scanned as being The Mark I scanner repeated this process in 1 increments
divided into axial slices. The x-ray beam to be used was to collect 180 views over 180. Today’s scanners may typ-
collimated down to a narrow (pencil-width) beam of x-rays. ically collect 1,000 or more views over 360 (the reason for
The size of the beam was 3 mm within the plane of the slice 360 rather than 180 will be explained later).
and 13 mm wide perpendicular to the slice (along the axis The combination of linear translation followed by incre-
of the subject). In fact, it is this beam width that typically mental rotation is called translate–rotate motion. Data
specifies the slice thickness to be imaged. The x-ray tube is collection was accomplished with a single narrow beam
rigidly linked to an x-ray detector located on the other side and a single sodium iodide (NaI) scintillation detector. This
of the subject. Together, the tube and the detector scan across arrangement (single detector and single narrow beam with
the subject, sweeping the narrow x-ray beam through the slice translate–rotate motion) is referred to as first-generation CT
(6). This linear transverse scanning motion of the tube and geometry and required 5–6 min to complete a scan. To
the detector across the subject is referred to as a translation. minimize examination time, the Mark I scanner actually
The arrangement is diagramed in Figure 1. used 2 adjacent detectors and a 26-mm-wide x-ray beam (in
During translation motion, measurements of x-ray trans- the slice thickness direction) to simultaneously collect data
mission through the subject are made by the detector at for 2 slices. By the end of the scan, Hounsfield had 28,800
many locations (Fig. 2). The x-ray beam path through the measurements (180 views · 160 rays) for each slice, taken

116 JOURNAL OF NUCLEAR MEDICINE TECHNOLOGY • Vol. 35 • No. 3 • September 2007

 FIGURE 3. Reconstruction matrix. Hounsfield envisioned
scanned slice as being composed of matrix of small boxes of
tissue called voxels, each with attenuation coefficient m. x-Ray
FIGURE 2. x-Ray transmission measurements. Measurements transmission measurements (Ni) can be expressed as sum of
are obtained at many points during translation motion of tube and attenuation values occurring in voxels along path of ray for Ni.
detector. x-Ray path corresponding to each measurement is
designated a ray, and set of rays measured during translation
is designated a view. Views are collected at many angles (in (X and Y) are often referred to as pixels; however, the sizes
1 increments in this example) to acquire sufficient data for image
reconstruction.
of the pixels in the displayed image (referred to as the
image matrix) are not necessarily the same as those in the
reconstruction matrix but rather may be interpolated from
at many angles (180) and positions (160). How an image is the reconstruction matrix to meet the requirements of the
generated from these data is discussed in the next section. display device or to graphically enlarge (zoom) the image.
To carry out reconstruction, consider the row of voxels
through which a particular ray passes during data collection
CT IMAGE RECONSTRUCTION (bottom of Fig. 3). Ni is the transmitted x-ray intensity for
this ray measured by the detector. No is the x-ray intensity
Hounsfield envisioned dividing a slice into a matrix of entering the subject (patient) for this ray. It can be shown
3-dimensional rectangular boxes (voxels) of material (tissue) (Appendix) that a derived measurement Xi can be related to
(Fig. 3). Conventionally, the X and Y directions are within a simple sum of the attenuation values in the voxels along
the plane of the slice, whereas the Z direction is along the the path of the ray; for the row of voxels in Figure 3, this
axis of the subject (slice thickness direction). The Z dimen- relationship is
sion of the voxels corresponds to the slice thickness. The
X and Y voxel dimensions (‘‘W’’ in Fig. 3) depend on the
size of the area over which the x-ray measurements are ob- Xi 5 u1 1 u2 1 u3 1 u4 1 . . . . . . : : 1 un ; Eq. 1
tained (the scan circle) as well as on the size of the matrix
(the number of rows and columns) into which the slice is where Xi 5 2ln(Ni/No) and ui 5wimi is the attenuation of
imagined to be divided. For example, suppose that each trans- voxel i. Similarly, measurements for all rays at all positions
lation covers 250 mm. After collection of all of the views, and angles can be expressed as sums of the attenuation
the measurements cover a scan circle with a diameter of values in voxels through which each ray has passed. Note
250 mm. If this scan circle is divided into a matrix of 250 that the quantity Xi is known; it is calculated from each
rows · 250 columns, each voxel is 1 · 1 mm. If a 512 · 512 detector measurement Ni with the known X and Y voxel
matrix is used (as is common today), each voxel is approxi- dimensions (W) and the known x-ray intensity entering the
mately 0.5 · 0.5 mm. This matrix is referred to as the re- patient (No). In Hounsfield’s scanner, No was directly mea-
construction matrix. sured by a reference detector sampling the x-ray intensity
The objective of CT image reconstruction is to determine exiting the x-ray tube. Modern scanners determine No from
how much attenuation of the narrow x-ray beam occurs in routine calibration scans.
each voxel of the reconstruction matrix. These calculated For example, consider a simple 2-row by 2-column recon-
attenuation values are then represented as gray levels in a struction matrix (Fig. 4A). Views are collected at 4 angles,
2-dimensional image of the slice (in a manner described 0 (left to right), 90 (top to bottom), 45 (diagonal), and 135
later). The 2 voxel dimensions lying in the plane of the slice (diagonal), and each measurement is expressed as the sum

PRINCIPLES OF CT AND CT TECHNOLOGY • Goldman 117

 u3, and u4). In fact, Hounsfield originally did exactly that,
using an 80 · 80 matrix to keep reconstruction times rea-
sonable. Shortly thereafter, a more computationally efficient
iterative approach was introduced. This iterative algorithm,
referred to as the algebraic reconstruction technique (ART),
is demonstrated for a 2 · 2 matrix slice in Figures 4B–4E.
Figure 4B shows the actual voxel attenuation values in pa-
rentheses (which are, of course, unknown before reconstruc-
tion) and each derived measurement. The process begins
by taking measurements of the first view (X1 5 10 and X2 5
10) and assuming that these attenuation values occurred
uniformly along the rays, yielding the first image estimate
(Fig. 4C). Next, measurements of the second view are taken
(X3 and X4), and these values are compared with the
summed attenuation values that would have occurred if
the first estimate were correct. The estimate predicted that
X3 and X4 should each equal 10 (Fig. 4C); the actual mea-
sured values were 14 and 6, respectively. Next, the first
estimate is adjusted to make it match the actual values of
X3 and X4 as follows: an adjustment for each measure-
ment of the second view is calculated by subtracting the
value predicted for it by the first estimate from its actual
value. In this case, these adjustments are

14 2 10 5 14 for X3

and
FIGURE 4. ART. (A) ART algorithm for 4-voxel ‘‘patient.’’ (B)
Attenuation measurements. (C) Starting estimate is constructed 6 2 10 5 2 4 for X4 :
by dividing measurements from first view equally along their
ray paths. (D–F) This estimate is iteratively adjusted to match Next, these adjustments are divided equally along the
measurements for each consecutive view, stopping when rays, adding 2 to u1 and u3 (total adjustment of 14) and
transmission measurements predicted by current estimate
match all actual measurements to within some preset tolerance. subtracting 2 from u2 and u4 (total adjustment of 24),
yielding the second estimate (Fig. 4D). The process con-
tinues in this manner, with the actual values for each view
being compared in turn with the values predicted by the
of the voxel attenuation values along each ray. In this case, latest estimate. For the third view (measurement 5), the
there are 6 equations: second estimate predicted an attenuation value of 10 (7 1
3); the actual value was 12. Thus, the 12 total adjustment is
X1 5 u1 1 u3 ; divided equally along the ray for X59, subtracting 1 each
from u1 and u4, to generate the third estimate (Fig. 4E). The
X2 5 u2 1 u4 ; required adjustment for the fourth view (a total of 22) is
divided along the ray (subtracting 21 from u2 and u3) to
X3 5 u1 1 u2 ; generate the fourth and last estimate (Fig. 4F). Because the
attenuation values predicted by this last estimate match all
X4 5 u3 1 u4 ; actual measurements, the image is accepted as the true re-
constructed image (and one can see that it matches the pa-
X5 5 u1 1 u4 ; renthetical values in Fig. 4B).
ART images were typically available within 20 s of scan
and completion. In practice, however, ART was sensitive to
quantum noise and could result in poor image quality if
X6 5 u2 1 u3 : transmitted x-ray intensities were low. As in conventional
radiography, noise (graininess) in CT images arises from a
limited number of x-ray photons contributing to each mea-
In principle, these equations may be solved by use of si- surement (because a limited patient radiation dose is used).
multaneous equations to calculate the 4 unknowns (u1, u2, This noise, known as quantum mottle, appears as unavoidable

118 JOURNAL OF NUCLEAR MEDICINE TECHNOLOGY • Vol. 35 • No. 3 • September 2007

random fluctuations (random errors) in detector measure- roundings (supplemental Fig. 1A) (supplemental materials
ments and is described by the Poisson statistical distribu- are available online only at http://jnm.snmjournals.org).
tion. The magnitude of the fluctuations is given by the The attenuation of the cylinder is highest through its center
square root of the number of photons contributing to the (where it is thickest) and decreases toward its edges. Back-
measurements. For example, if 100 x-rays were included in projection builds a cylinder image whose intensity decreases
measurement X3 in Figure 4B, then the size of the fluctu- from the maximum at the center toward the edges (supple-
ations (noise) would be the square root of 100, or 10. If mental Fig. 1B), exhibiting the classic appearance of blur-
repeated measurements of X3 were obtained for the same ring. To reconstruct a ‘‘deblurred’’ image, a filter function is
subject with the same apparatus, then the measurements mathematically applied to each view before backprojection.
would randomly fluctuate around an average of 100, with The mathematic operation is called convolution, but the pro-
approximately two thirds of the trials being between 90 and cess is referred to here as filtering. Examples of an original
110 (i.e., 100 6 10). view, a filter function, and a filtered (deblurred) view are
The problem with ART can now be explained: No estimate shown in supplemental Figure 2. The filtered view exhibits
that ART generates will ever match all measurements exactly, ‘‘edginess,’’ which the original view did not. For any scan,
because the measurements include random errors. Thus, ART once all views are filtered and then backprojected, the
must be discontinued after an estimate that matches all mea- resulting image no longer exhibits the original blurriness.
surements only within some tolerance is found; for high noise This reconstruction algorithm, known as filtered backpro-
(low intensities), that tolerance may not be met even after jection (FBP), is efficient, yields excellent results, and is
many iterations. Fortunately, the early success of CT created still the algorithm of choice today for slice-by-slice CT.
a wealth of research and development activities (discussed Figure 6 compares simulated phantom scans reconstructed
later). One such innovation was a much better reconstruction with and without filtering.
algorithm based on the idea of backprojection. Interestingly, the filter functions used in most CT image
Backprojection is illustrated in Figure 5. Suppose that a reconstructions are deliberately chosen to reconstruct par-
simple test object containing 3 objects with different atten- tially blurred images. The reason for this choice is to sup-
uation values is scanned as shown in Figure 5A and that press the visibility of quantum mottle in images (i.e., image
views (attenuation measurements) are obtained at 3 angles. noise). For typical scan technique factors (discussed later),
For each view at each angle, the reconstruction process is fully deblurred images exhibit unacceptably high image
very straightforward. The attenuation measurements of each noise (graininess); that is, high resolution makes noise easy
view are simply divided evenly along the path of the ray; to see. Noise significantly interferes with the perception of
this process is called backprojection. An example of back- low-contrast structures. Because a primary application of CT
projection for the view measured at 0 is shown in Figure is the imaging of low-contrast soft tissues and because soft
5B. Note that after backprojection of only 4 views, an tissues do not have well-defined edges (for which higher
image of the test object is beginning to appear in Figure 5C. resolution would be important), reduced noise is more im-
Backprojection is efficient (each measurement is pro- portant for soft-tissue image quality than is higher resolu-
cessed just once) and involves relatively simple calcula- tion. The ‘‘standard’’ filter function normally used is chosen
tions but has a serious flaw: The resulting images are blurry to deliver the best compromise between resolution and im-
(i.e., they exhibit poor spatial resolution). This problem is age noise. However, if the operator (or image reader) prefers,
evident even in the partial reconstruction shown in Figure a sharper (less blurry but noisier) or smoother (more blurry
5C. This blurring is a natural result of the scanning and but less noisy) filter function may be selected during ex-
backprojection processes, but there is a solution: The blur- amination prescription. These selectable reconstruction fil-
ring can be reversed by a mathematic process known as ters may be given mnemonic names, such as standard, smooth,
filtering. Consider a scan of a phantom containing a single detail, or bone, or numbers, such as B10 or B20, depending
cylinder with an attenuation higher than that of its sur- on the manufacturer.

FIGURE 5. (A) Backprojection recon-

struction for simple phantom containing 3
objects with different attenuation values.
(B) For each view, attenuation values are
simply divided evenly along their ray
paths. Summing backprojected views
from several angles builds image. (C) Four
views of phantom are summed. Although
this method is efficient, images recon-
structed with backprojection exhibit con-
siderable blurriness.

PRINCIPLES OF CT AND CT TECHNOLOGY • Goldman 119

 represents a range of 10 CT numbers. However, it can be seen
from Equation 2 that a 0.5% attenuation difference—well
within the ability of CT to distinguish—corresponds to a CT
number difference of 5 Hounsfield units. Many soft-tissue
structures would be displayed with the same gray level as their
surroundings and thus would be invisible. In fact, a material
must differ from its surroundings by at least 1% to ensure a
different gray level. Tissues of most common interest in CT
generally lie in the CT number range of 2100 to 1100 (7). If
instead discernible gray levels were limited to the CT number
FIGURE 6. FBP. Mathematic phantom image reconstructed range of 2100 to 1100, then no bone details (all white) and no
without (A) and with (B) filtering. FBP effectively reconstructs lung details (all black) would be visible. The solution, which is
high-quality images. Adapted from S. Napel. taken for granted today, is windowing, illustrated in supple-
mental Figure 3. Aviewer may (interactively) decide how gray
levels are to be allocated by specifying a window width (the
CT IMAGE PRESENTATION range of CT numbers to display, e.g., 250 to 200) and a
window level (or window center, e.g., 0). Adjustable window
A convention that has existed from the earliest days of
settings may then be altered to view other CT number ranges.
CT is to replace the attenuation value calculated for each
Universally today, windowing can be interactively performed
voxel of the reconstruction matrix with an integer (CT
with a mouse or trackball.
number) calculated as follows:
FIRST-GENERATION EMI CT SCANNER
CT number ðin Hounsfield units; HUÞ
The original EMI device was a dedicated head scanner in
5 ½K · ðuvoxel 2 uwater Þ=uwater : Eq. 2
which the patient’s head was recessed via a rubber mem-
brane into a water-filled box (supplemental Figs. 4 and 5).
In this equation, uvoxel is the calculated voxel attenuation The device was designed such that the water-filled box ro-
coefficient, uwater is the attenuation coefficient of water, and tated (in 1 increments) along with the single-narrow-beam,
K is an integer constant. In the original EMI scanner, K was single-detector assembly, resulting in a fixed path length
500, but K later become standardized as 1,000 (or some- through patient plus water for all rays and transmission mea-
times 1,024). The reason for calculating CT numbers rela- surements.
tive to water is discussed in the next section. Scanners today The results obtained with this first clinical EMI scanner
determine uwater from periodic calibration scans of water (installed in September 1971) were presented at a British
or water-equivalent phantoms. Because attenuation coeffi- radiologic society meeting in April 1972. The results left no
cients are affected by x-ray beam energy, proper x-ray gen- doubt as to the revolutionary clinical value of the process.
erator calibration is important for accurate and reproducible The historic success of the scanner created enormous inter-
CT numbers. Note that the quantity is referred to as CT est and led to an explosion of research and development by
number, but the units are Hounsfield units (to honor the in- many groups and corporations. One such development was
ventor). Thus, one would say, for example, that a particular FBP reconstruction, described earlier. However, a technol-
tissue in an image ‘‘has a CT number of 40 Hounsfield ogy race to improve and expand the CT process was also
units.’’ under way.
Some examples of CT number calculations are as follows: Of course, one important application was body scanning,
A voxel actually containing water would have a CT number of but the water-filled box had to be eliminated. However, a
0 (for a well-calibrated scanner), because uwater 2 uwater 5 0 discussion of why it was used in the first place is worthwhile.
(the CT number of water will always be approximately 0 but It served 2 purposes, both of which allowed Hounsfield to
not necessarily exactly 0 because of quantum mottle); if the maximize the accuracy of attenuation coefficient measure-
voxel contained air (for which uair  0), the CT number would ments. These 2 reasons were as follows.
be approximately –1,000; and for a voxel containing dense
cortical bone (for which u  2 uwater), the CT number would be Limitation of Dynamic Range
approximately 11,000. The water-filled box greatly reduced the range of inten-
Displaying the full possible CT number range (2500 to 500 sities over which the detector needed to accurately respond,
for the original EMI scanner and 21,000 to 11,000 for thus allowing optimization of the detector sensitivity.
scanners introduced soon after that) presents a problem: Even
today, displays often show only about 250 shades of gray, of Beam-Hardening Correction
which fewer (perhaps as few as 100) may be visually discern- x-Rays produced in x-ray tubes are mostly bremsstrah-
ible. If the full CT number range of 2,000 (21,000 to 11,000) lung x-rays which, unlike the discrete photon energies emitted
is spread evenly over 200 discernible gray levels, each level by radioactive isotopes, cover a broad continuum of energies

120 JOURNAL OF NUCLEAR MEDICINE TECHNOLOGY • Vol. 35 • No. 3 • September 2007

(up to a maximum numerically equal to the x-ray tube replaced. In part, this goal is accomplished by use of a
kilovoltage). Such beams are referred to as polychromatic. ‘‘bow-tie’’ filter through which the beam passes when exit-
Beam hardening refers to a gradual increase in the effective ing the x-ray tube. First, the filter (usually made of alu-
energy of polychromatic x-ray beams as they penetrate deeper minum) is thicker where the path through the patient is
into attenuating materials. It is caused by preferential atten- shorter (toward the edges), effectively limiting the range of
uation of the lower-energy (and thus less penetrating) photons intensities reaching the detector. Second, the filter prehard-
in the beam by each successive layer of attenuating material ens the beam (i.e., it removes the lower-energy x-rays), so
(1). Because attenuation coefficients depend on both the ma- that less hardening occurs in the patient. The remainder of
terial and beam energy, the same tissue at a greater depth the beam-hardening correction is done with software and
has a lower attenuation coefficient (because the deeper tis- periodically performed calibration scans of uniform phantoms
sue attenuates less of the hardened x-ray beam). of various sizes. Although quite sophisticated and effective,
During scanning of a uniform object (e.g., a cylindric software beam-hardening corrections are only approximate,
water-filled phantom), beam hardening causes lower atten- and nonuniformities can in fact often be observed in scans
uation coefficients to be reconstructed for deeper voxels, of uniform phantoms when a very narrow display window
producing an undesired ‘‘cupping’’ artifact in which the same is used.
material appears darker in the center of the image than in the
periphery. To understand how the water-filled box allowed
Hounsfield to correct for this beam-hardening effect, con- REDUCING SCAN TIME: SECOND GENERATION OF CT
sider the path of one of the measurements through the water- The first waterless full-body CT scanner was developed
filled box and patient. As in Equation 1, the measurement is and installed by Ledley et al. at Georgetown University in
expressed as a sum of the attenuation coefficients of the February 1974 (8). This device introduced several innova-
voxels through which it passes: tions now standard in CT (table movement through the gan-
try, gantry angulation, and a laser indicator to position slices)
Xa 5 u1 1 u2 1 u3 1 u4 1 . . . . . . :: 1 un : as well as a Fourier-based reconstruction algorithm math-
ematically equivalent to FBP (R.S. Ledley, personal com-
Now consider a measurement made along an identical path munication) but still used first-generation design, with scan
length but passing only through water: times of 5–6 min. As a result, body scans were unavoidably
corrupted by patient motion.
X w 5 uw 1 uw 1 uw 1 uw 1 . . . . . . : : 1 uw : A step toward reduced scan times was taken with the
introduction of second-generation CT geometry in late 1974
In fact, it is evident from supplemental Figure 5 that water- (9). Second-generation CT used multiple narrow beams
only measurements can be made at the beginning and end and multiple detectors and, as in the first generation, used
of every view, when the narrow beam passes outside the rotate–translate motion. A second-generation scanner with
edges of the patient’s head. Because each voxel along the 3 narrow beams and 3 detectors is shown in supplemental
path of Xw contains water, each voxel can be corrected for Figure 6 and diagramed in Figure 7 (the 3 photomultiplier
beam hardening (because all of the voxels must have the tubes associated with the 3 detectors can be seen to the right
same attenuation coefficient—that of water). The next step of the patient aperture in the photograph in supplemental Fig.
is to subtract Xw from Xa: 6). It may seem at first that second-generation CT accelerated
data collection via simultaneous measurements at each point
along a translation. In fact, if the set of rays measured by each
X9a 5 ðXa 2 Xw Þ 5 ðu1 2 uw Þ1ðu2 2 uw Þ1ðu3 2 uw Þ
detector is considered (Figs. 7B–7D for the left, center, and
1ðu4 2 uw Þ1 . . . . . . 1ðun 2 uw Þ: Eq. 3 right detectors, respectively), it can be seen that the detectors
actually acquire their own separate, complete views at dif-
The attenuation coefficients of soft tissues differ by only ferent angles. If there is a 1 angle between each of the
a few percentage points from that of water. Therefore, the 3 narrow beams, then one translation collects data for views
terms of Equation 3 represent small corrections for the at, for example, 30, 31, and 32. Only one third as many
attenuation coefficient of each voxel relative to that of translations are required, and scan time is reduced 3-fold;
water—the latter can be corrected (from the water-only what took 6 min now takes 2 min. In general, second-
path). By comparing each measurement with a water-only generation CT can reduce scan time by a factor of 1/N,
measurement, Hounsfield was actually calculating attenu- where N is the number of detectors.
ations as ui – uw (rather than ui), a factor that explains the Second-generation designs that used 20 or more narrow
numerator of Equation 2: CT numbers were calculated from beams and detectors (collecting equivalent numbers of si-
ui – uw because those were the values that Hounsfield multaneous views) were soon introduced clinically, reducing
actually reconstructed. scan times to 20 s or less. At that point, body scan quality
To eliminate the water-filled box and expand the useful- leapt forward dramatically, because scans could be performed
ness of CT, the functions that the box served must be within a breath hold for most patients. However, further speed

PRINCIPLES OF CT AND CT TECHNOLOGY • Goldman 121

FIGURE 7. Second-generation data collection. (A) Transmis-
FIGURE 8. Third-generation geometry. Time-consuming and
sions of multiple narrow beams (3, in this case) were simulta-
mechanically complex translation motion was eliminated by
neously acquired by multiple detectors during each translation.
opening x-rays into fanbeam. Large array of detectors mea-
(B–D) Small angle between narrow beams allowed each
sured data across width of fan. Tube and detectors were rigidly
detector to acquire complete separate view at different angle.
linked and underwent single rotational motion.
Number of required translations was correspondingly reduced
by factor of 1/(number of detectors).
scan as quickly as one third of a second for cardiac appli-
improvements were limited by the mechanical complexity cations.
of rotate–translate geometry. Translations and rotations Third-generation CT requires extremely high detector
needed to be performed quickly and precisely while moving stability and matching of detector responses. First- and
heavy (lead-shielded) x-ray tubes and the associated gan- second-generation detectors were dynamically recalibrated
tries and electronics—all without causing significant vibra- at the beginning of each translation, before passing into the
tions. Even small deviations (because of vibrations or other patient’s shadow. Also, each detector measured rays passing
misalignments) of scanner hardware position relative to re- through all voxels, so that any detector error or drift was
construction matrix voxels would cause data to be back- spread evenly across the image and generally was not visible.
projected through the wrong voxels, creating severe artifacts Because third-generation CT tubes and detectors are rigidly
in the images. The mechanical tolerances and complexities linked, each detector measures rays passing only at a specific
involved indicated the need to eliminate translation motion. distance from the center of rotation, depending on the lo-
cation of the detector in the array (supplemental Fig. 7a). Any
error or drift in the calibration of a detector relative to the
THIRD GENERATION OF CT other detectors is backprojected along these ray paths and
Faster scans required the elimination of translation mo- reinforced along a ring where they cross. The result is an
tion and the use of smoother and simpler pure rotational image ‘‘ring artifact’’ (supplemental Fig. 7b). Furthermore,
motion. This goal is accomplished by widening the x-ray most detectors are in the patient’s shadow during the entire
beam into a fanbeam encompassing the entire patient width scan, preventing dynamic recalibration. Ring artifacts may
and using an array of detectors to intercept the beam (Fig. result from detector inaccuracies as small as 0.1%.
8). The detector array is rigidly linked to the x-ray tube, so A solution was provided by xenon detector arrays (sup-
that both the tube and the detectors rotate together around plemental Fig. 8) (10). A xenon array consisted of a long
the patient (a motion referred to as rotation–rotation). Many metal chamber subdivided into many small chambers by
detectors are used (;250 for the initial models and 750 or thin plates called septa. Every second septum was connected
more in later designs) to allow a sufficient number of mea- to a common positive bias voltage source. The alternate septa
surements to be made across the scan circle. This design, acted as collectors individually connected to an electronic
characterized by linked tube–detector arrays undergoing readout. Each small chamber thus formed an ionization
only rotational motion, is known as third-generation geom- detector (7). Xenon arrays were inherently stable and well
etry. The early third-generation CT scanners, installed in matched because factors affecting detector responses were
late 1975, could scan in less than 5 s; current designs can either uniform for the entire array (bias voltage or type and

122 JOURNAL OF NUCLEAR MEDICINE TECHNOLOGY • Vol. 35 • No. 3 • September 2007

density of gas) or constant over time (chamber volume). Scatter
Xenon (atomic number 54) under high pressure (10–25 A fundamental advantage of the CT process was the
atm, to increase density), rather than poorly absorbing air, potential to (almost) completely eliminate scatter. Scattered
was used to maximize the absorption of x-rays entering the photons are emitted in (nearly) random directions (i.e.,
detectors. Xenon arrays were eventually replaced by solid- nearly isotropic). First-generation CT was essentially scatter-
state detectors (discussed later). free, because only a very few scattered photons that hap-
It should be noted than ring artifacts in third-generation pened to be scattered in directions nearly parallel to the
CT images are never completely eliminated. Rather, they narrow beam could reach the detectors.
are minimized by high-quality detector design and frequent In comparison, it seems that the broad fanbeam used
(daily) calibration scans. Residual ring artifacts in images in third-generation CT would reintroduce scatter carefully
are then removed by image-processing algorithms, without avoided in earlier generations. It is certainly true that more
which rings would likely be seen in every CT image. De- scatter is produced at any instant (because more tissue is ir-
tector failure or drift may still cause ring artifacts to occa- radiated) and that scatter that would miss a first-generation
sionally appear. Despite this and other limitations described CT detector may be detected by a third-generation array (Fig.
in the next section (and despite the introduction of a faster 9). However, the amount of scatter produced is still far less
fourth generation of CT), third-generation CT was highly than that associated with conventional radiography, because
successful and remains the basic geometry of most CT scan- the CT beam is not more than 1 cm wide in the Z direction,
ners manufactured today. compared with up to 43 cm (17 in.) in radiography. In any
case, even this scatter can be (mostly) eliminated by use of the
equivalent of a scatter removal grid; x-ray-absorbing septa
CT PERFORMANCE CRITERIA
placed at the boundaries between detectors and aimed at the
A few concepts regarding CT radiation dose and image focal spot allow primary x-rays to pass while stopping most
quality are mentioned here to allow meaningful discussions scattered x-rays before they enter the detectors (Fig. 9B).
of the advantages and trade-offs of various CT designs.
Sampling and Spatial Resolution
Dose Efficiency A CT view was defined earlier as a set of measurements
Sensitivity (the ability to visualize low-contrast structures) (samples) collected at a particular angle during one transla-
is limited by noise (quantum mottle), which is caused by use tion. Later designs (third- and fourth-generation CT; dis-
of a finite, limited number of primary x-rays to form the cussed later) collect views in the form of fanbeams (Fig. 8)
image. Maximum sensitivity for any given radiation dose and perform FBP reconstruction from fanbeam data; how-
requires capturing and using as high a fraction of the primary ever, fanbeam data can always be resorted into—and are thus
x-rays exiting the patient as possible. This concept is referred
to as dose efficiency, and it involves 2 factors: geometric
efficiency (fraction of transmitted x-rays interacting with
active detector areas) and absorption efficiency (fraction of
x-rays interacting with active detector areas that are actually
captured). Geometric efficiency is reduced if some x-rays are
absorbed before detection (such as in the front housing of
detectors) or if some x-rays do not enter active areas of the
detectors (such as by passing between detectors or striking
inactive dividers between individual detectors). Absorption
efficiency is reduced if some x-rays entering the detectors
pass all the way through without being absorbed.
The scintillation detectors used in first- and second-
generation CT were sized to intercept the entire exiting nar-
row beam and absorbed essentially 100% of available x-rays
(;100% dose efficient).
In contrast, xenon detector absorption efficiency was
about 60%–70%. Geometric efficiency was reduced by x-ray
absorption in the housing and by x-rays striking the septa FIGURE 9. Scatter with CT fanbeam. Use of fanbeam
edge on, so that overall dose efficiency was typically about increases scatter production at any moment (more tissue is
50%–60%. By the mid-1980s, improvements in technology irradiated), and more scatter can reach detectors. Amount of
allowed more dose-efficient solid detector arrays to begin scatter produced is still much smaller than that in radiography
because fanbeam is only ;1 cm thick. Scatter can be
replacing the xenon arrays used in third-generation scan- eliminated in third-generation CT with scatter removal septa,
ners; solid detector arrays are the detectors of choice for all which act like nearly ideal grid. (A) First-generation CT. (B)
CT scanner manufacturers today. Third-generation CT.

PRINCIPLES OF CT AND CT TECHNOLOGY • Goldman 123

equivalent to—parallel-ray views of translate–rotate scan-
ners. Spatial resolution in the plane of the CT slice is ulti-
mately limited by the spacing of the measurements (referred
to here as samples) and by the sizes of the detectors (sampling
aperture). Fundamentally, the closer together the samples
and the smaller the aperture, the higher the maximum
resolution (although the actual resolution depends on the
reconstruction filter selected). Views must of course be taken
at a sufficient number of angles to provide sufficient data for
the reconstruction process.
A limitation of third-generation geometry is this sampling.
Although it is possible to collect views at any number of
angles, sample size and spacing are fixed by detector design;
samples cannot be closer together than the distance between
rays associated with each detector at the level of the center of
rotation (supplemental Fig. 9). Obtaining smaller and more
closely spaced measurements would require a detector array
with smaller—and more—detectors. A common technique
for circumventing this limitation is the ‘‘1/4-ray offset.’’
Suppose a scan is performed over 360 rather than 180. At
some point during the 360 scan, every ray is measured twice, FIGURE 10. Fourth-generation scan geometry. Fixed detector
but with the rays traveling in opposite directions. For exam- ring in original design was quite large, because tube rotated
ple, a detector at the center of the array measures the same ray inside ring. Later designs moved tube outside ring and tilted
(i.e., x-rays passing through the same voxels) when the tube is ring out of way of x-ray beam as x-ray tube swept by.
at 0 and the detector is at 180 and again 180 later when
their positions are reversed. Through a shift of the detector
array mounting by a small amount (1/4 of a detector width) rapidly measurements are made as the tube sweeps across
relative to the center of rotation, these opposing rays can be the field of view of the detector. Also, unlike third-generation
made to interleave, effectively doubling the sample density detectors, each fourth-generation detector can measure rays
(i.e., the samples are half as far apart) (supplemental Fig. 10). at any distance from the center of rotation and can be dy-
An extension of this idea is the dynamic focal spot; although namically calibrated before it passes into the patient’s shadow,
the tube and detectors are linked, the x-ray beam can move so that ring artifacts are not a problem. On the other hand,
relative to the detectors through electronic shifting of the the number of detectors strictly limits the number of views
location of the focal spot on the x-ray tube anode. If the focal that can be acquired.
spot is shifted by half of a detector width, then interleaved Drawbacks of early fourth-generation CT included size
samples are again obtained (11). and geometric dose efficiency. Because the tube rotated
inside the detector ring, a large ring diameter (170–180 cm)
was needed to maintain acceptable tube–skin distances. On
FOURTH-GENERATION SCANNERS the other hand, acceptable spatial resolution limited detector
By 1976, 1-s scans were achieved with a design incor- apertures to ;4 mm. Consequently, even allowing for ;10%
porating a large stationary ring of detectors, with the x-ray space between detectors, 1,200 or more detectors were
tube alone rotating around the patient (supplemental Fig. 11). needed to fill the ring, but cost considerations initially limited
This approach, known as fourth-generation geometry and de- the number to 600. The result was gaps between detectors and
veloped under contract with the National Institutes of Health, low geometric dose efficiency (,50%), as shown in supple-
is diagramed in Figure 10. Although it appeared later, fourth- mental Figure 11 (up to 4,800 detectors were eventually
generation technology was in fact developed (for the most used). A later, alternate design used a smaller ring placed
part) in parallel with—and partly in response to—engineering closer to the patient, with the tube rotating outside the ring;
challenges of the third-generation approach. during tube rotation, the part of the ring between the tube and
With regard to sampling in fourth-generation CT, the set the patient would tilt out of the way of the x-ray beam (the
of rays measured by one detector as the x-ray tube sweeps peculiar wobbling motion of the ring was called nutation).
across its field of view is analogous to one third-generation Another disadvantage of fourth-generation designs was
fanbeam view but with the roles of tube and detectors re- scatter. The scatter-absorbing septa used in third-generation
versed (supplemental Fig. 12). Each fourth-generation de- designs could not be used, because the septa would neces-
tector collects a complete fanbeam view (which may be sarily be aimed at the center of the ring, which was the
rebinned into parallel-ray views)—but a view in which sam- source of the scatter (patient’s location); that is, they would
ple spacing may be arbitrarily close, limited only by how preferentially transmit scatter rather than primary x-rays.

124 JOURNAL OF NUCLEAR MEDICINE TECHNOLOGY • Vol. 35 • No. 3 • September 2007

The removal of scatter was never truly solved in fourth- SLIP RING SCANNERS AND HELICAL CT
generation designs. After the fourth generation, CT technology remained
Although it was introduced later, fourth-generation CT stable (other than incremental improvements) until 1987.
was not more (or less) advanced than third-generation CT. By then, CT examination times were dominated by interscan
Both had advantages and disadvantages, and both types of delays. After each 360 rotation, cables connecting rotating
scanners were manufactured until relatively recently (al- components (x-ray tube and, if third generation, detectors) to
though third-generation CT had a larger market share). It the rest of the gantry required that rotation stop and reverse
was only the advent of multislice CT, for which fourth- direction. Cables were spooled onto a drum, released during
generation detector arrays would be prohibitively expen- rotation, and then respooled during reversal. Scanning,
sive, that led to the demise of fourth-generation CT. braking, and reversal required at least 8–10 s, of which only
1–2 were spent acquiring data. The result was poor temporal
resolution (for dynamic contrast enhancement studies) and
ELECTRON-BEAM CT (EBCT) long procedure times.
Eliminating interscan delays required continuous (non-
Besides obvious mechanical challenges, subsecond scans
stop) rotation, a capability made possible by the low-voltage
would burden x-ray tubes; because approximately 250 mAs
slip ring. A slip ring passes electrical power to the rotating
(i.e., tube current in milliamps times scan time in seconds)
components (e.g., x-ray tube and detectors) without fixed
per slice are required for acceptable body CT, a ‘‘mere’’
connections. The idea is similar to that used by bumper cars;
0.25-s scan would require sustaining 1,000 mA. However,
power is passed to the cars through a metal brush that slides
cardiac CT requires ultrafast scans (,50 ms) to freeze
along a conductive ceiling. Similarly, a slip ring is a drum
cardiac motion, a goal unreachable with conventional scan-
or annulus with grooves along which electrical contactor
ning even today (although cardiac CT is possible with today’s
brushes slide (supplemental Fig. 14). Data are transmitted
fastest multislice CT scanners assisted by gating and ad-
from detectors via various high-capacity wireless technolo-
vanced image-processing software). Various ideas were
gies, thus allowing continuous rotation to occur. A slip ring
proposed (including the use of multiple x-ray tubes), but
the goal was ultimately accomplished with a novel CT design allows the complete elimination of interscan delays, except
that had no moving parts and that was capable of performing for the time required to move the table to the next slice
complete scans in as little as 10–20 ms. position. However, the scan–move–scan sequence (known as
The idea behind ultrafast CT is a large, bell-shaped x-ray axial step-and-shoot CT) is still somewhat inefficient. For
tube (supplemental Fig. 13) (12). An electron stream emitted example, if scanning and moving the table each take 1 s, only
from the cathode is focused into a narrow beam and elec- 50% of the time is spent acquiring data. Furthermore, rapid
tronically deflected to impinge on a small focal spot on an table movements may introduce ‘‘tissue jiggle’’ motion arti-
annular tungsten target anode, from which x-rays are then facts into the images.
produced. The electron beam (and consequently the focal An alternate strategy is to continuously rotate and contin-
spot) is then electronically swept along all (or part) of the uously acquire data as the table (patient) is smoothly moved
360 circumference of the target. Wherever along the annular though the gantry; the resulting trajectory of the tube and
target the electron beam impinges, x-rays are generated and detectors relative to the patient traces out a helical or spiral
collimated into a fanbeam (various detector schemes were path (Fig. 11) (15,16). This powerful concept, referred to
envisioned). The concept is known as EBCT. synonymously as helical CT or spiral CT, allows for rapid
A cardiac EBCT scanner with a partial (semicircular, scans of entire z-axis regions of interest, in some cases within
;210) anode and opposing detectors was built by 1984 (13). a single breath hold. So significant were improvements in
To acquire rapid heart scans without table (patient) move- body CT quality and throughput that helical scanning became
ment, it included 4 anodes and 2 detector banks, each offset the de facto standard of care for body CT by the mid-1990’s.
along the z-axis so as to acquire 8 interleaved slices covering Certain concepts associated with helical CT are funda-
8 cm of the heart. Although the device was able to go faster, mentally different from those of axial scanning. One such
limited tube current (650 mA) required slower (50- to 100- concept is how fast the table slides through the gantry
ms) scans for acceptable mAs values and associated image relative to the rotation time and slice thicknesses being
quality (for a complete description of EBCT, see the chapter acquired. This aspect is referred to as the helical pitch and
by McCollough in Medical CT and Ultrasound: Current is defined as the table movement per rotation divided by the
Technologies and Applications (14)). Although still avail- slice thickness. Some examples are as follows: if the slice
able, EBCT was limited to a niche market (cardiac screen- thickness is 10 mm and the table moves 10 mm during one
ing), mostly because image quality for general scanning was tube rotation, then the pitch 5 10/10, or 1.0; if the slice
lower than that of conventional CT (because of low mAs thickness is 10 mm and the table moves 15 mm during one
values) and because of higher equipment costs. With progress tube rotation, then the pitch 5 15/10, or 1.5; and if the slice
being made in cardiac scanning by multislice CT, the future thickness is 10 mm and the table moves 7.5 mm during one
of EBCT is uncertain. tube rotation, then the pitch is 7.5/10, or 0.75. In the first

PRINCIPLES OF CT AND CT TECHNOLOGY • Goldman 125

 slice plane, interpolation between the measurements ob-
tained 5 mm below and those obtained 5 mm above is used. In
this case, because the 2 measurements are equidistant from
the slice, the 2 measurements are simply averaged. If instead
the first measurement is only 2 mm above the slice and the
second measurement is 8 mm below, then the first measure-
ment—being closer—is given greater weight.
It is distinctly possible, however, that the anatomy actually
FIGURE 11. Helical CT. Improved body CT was made lying between the 2 measurements in the example just
possible with advent of helical CT (or spiral CT). Patient table
is moved smoothly through gantry as rotation and data
described differs significantly from that at which the mea-
collection continue. Resulting data form spiral (or helical) path surements are obtained (there may be rapidly changing
relative to patient; slices at arbitrary locations may be anatomy or small structures). If so, then the estimate may
reconstructed from these data. be significantly in error, leading to image misrepresentation
or image artifacts. As a rule, the farther apart interpolated
example, the x-ray beams associated with consecutive data points are, the greater the chance of error. The process
helical loops are contiguous; that is, there are no gaps be- just described is called 360 interpolation, because data are
tween the beams and no overlap of beams. In the second interpolated between measurements obtained 360 (one full
example, a 5-mm gap exists between x-ray beam edges of rotation) apart by the same detector. However, this scheme
consecutive loops. In the third example, beams of consec- can be improved; as noted earlier, at some point during each
utive loops overlap by 2.5 mm, thus doubly irradiating the 360 rotation, every ray is measured twice, but with the
underlying tissue. The choice of pitch is examination de- x-rays traveling in opposite directions (i.e., 180 apart).
pendent, involving a trade-off between coverage and accu- These 180-opposed rays are closer together—and thus allow
racy. Pitch is discussed in greater detail later in the article. fewer chances for interpolation errors—than those that are
Another concept associated with helical CT is slice inter- 360 apart. The process that makes use of these rays is known
polation. For step-and-shoot CT, all data for slice reconstruc- as 180 interpolation and is the one that is generally used
tion are collected at the Z position of that slice before moving (Fig. 12). (Note that it is only for detectors at the center of the
to the next position. Reconstructed slices clearly correspond array that 180-opposed rays are exactly half as far apart as
to the Z positions at which the data were obtained. Helical those separated by 360; for detectors closer to the ends of the
CT, however, provides no obvious slice Z positions; relative array, measurements will not be as evenly spaced.)
to the continuous helical data path, any arbitrarily selected
slice plane (such as the vertical line in Fig. 11) is equivalent to
any other. Thus, once helical data are acquired, slices may be
reconstructed at any Z position. On the other hand, no slice
plane contains sufficient data to actually reconstruct a slice.
To see the problem, consider the slice plane shown as a
vertical line in Fig. 11; if the helix represents the path of the
detectors, then it is clear that the detectors (and thus the
locations of measured views) are above the slice plane at 0
but below the slice plane 180 later (or vice versa). Only a
small number of measurements actually lie exactly within the
plane of the slice. Recall, however, that image reconstruction
requires that a sufficient number of views over a sufficiently
large angle (at least 180) be obtained. To allow image
reconstruction, it is necessary to estimate from the measure-
ments lying above and below the slice what the measure-
ments would have been within the slice. The process known
as interpolation is used to make these estimates.
For example, suppose a scan is performed with a 10-mm
slice thickness and a pitch of 1 and then data (rays) measured FIGURE 12. Helical CT sample spacing and interpolation. If
by one particular detector are considered. Suppose that when data for desired slice of thickness d (dark gray bar in figure) are
this detector is at 50 (i.e., it measures a ray whose path interpolated between equivalent rays from adjacent helical
through the patient is at 50), the detector is 5 mm above the rotations (loops) with pitch of 1.5, samples will be 1.5 d apart
plane of the slice to be reconstructed. After one rotation with along z-axis (e.g., 10.5 mm apart for 7-mm thickness). Larger
spacing means greater chance that interpolated estimate is in
a pitch of 1, that detector again measures a ray at 50 but now error. If 180-opposed rays are included, measurements aver-
5 mm below the desired plane. To estimate the equivalent 50 age half as far apart (and are more likely to actually lie within
ray that would have been measured by this detector within the slice). det 5 detector.

126 JOURNAL OF NUCLEAR MEDICINE TECHNOLOGY • Vol. 35 • No. 3 • September 2007

 The selection of pitch is essentially a trade-off between pa- of a ray, consider the row of voxels in Figure 3, where No
tient coverage and accuracy; larger pitches allow more cov- is x-ray intensity entering the row of voxels, Ni is the
erage of a patient per unit of time (or per breath hold), but slice detector-measured intensity, wi is the path length of the ray
data must be interpolated between points that are farther apart, through the voxel, and mi is the attenuation coefficient of
allowing more chances for errors. For a 10-mm slice thickness, the material contained within that voxel.
a pitch of 1.5, and a 1-s rotation time, 30 cm of patient can be Consider the intensity N1 exiting the first voxel (atten-
covered in only 20 s, but interpolation points are, on average, uation m1). Using the expression for exponential attenua-
7.5 mm apart. Alternatively, for a pitch of 1, only 20 cm can be tion,
covered in 20 s, but the average interpolation point separation
will be only 5 mm. Pitches of between 1 and 1.5 are commonly N1 5 No e 2 ðw1 m1 Þ : Eq. 1A
used. Pitches of greater than 1.5 are uncommon, whereas
pitches of greater than 2 generally yield unacceptable results Similarly, given that the intensity N1 enters the second
and are not used. Pitches of less than 1 are not used in single- voxel, the intensity exiting the second voxel is given by the
slice helical CT because of the double irradiation mentioned following equation:
earlier but are common in multislice CT.
The definition of pitch described earlier (table movement
N2 5 N1 e 2 ðw2 m2 Þ 5 No e 2 ðw1 m1 Þ e 2 ðw2 m2 Þ : Eq. 2A
per rotation divided by the slice thickness) has been altered to
accommodate multislice CT. The definition has been updated
by replacing slice thickness in the denominator with the z-axis Given that N2 enters the third voxel, the intensity exiting
x-ray beam width. However, for conventional single-slice CT, the third voxel is calculated as follows:
slice thickness and z-axis x-ray beam width are equivalent.
N3 5 N2 e 2 ðw3 m3 Þ 5 No e2 ðw1 m1 Þ e 2 ðw2 m2 Þ e 2 ðw3 m3 Þ : Eq. 3A
CT FLUOROSCOPY
Proceeding in this fashion through the last voxel,
The potential for the high sensitivity of CT to guide per-
cutaneous aspirations and biopsies had long been recognized.
With continuous, dynamic scanning being made possible Ni 5 No e 2 ðw1 m1 Þ e 2 ðw2 m2 Þ e 2 ðw3 m3 Þ . . . . . . . . . ::e 2 ðwn mn Þ :
by slip ring CT, real-time CT fluoroscopy became feasible. Eq. 4A
A slip ring scanner is modified to allow real-time tableside
image viewing and table positioning via foot pedal– Because the product of exponential functions is equal to the
controlled acquisitions and joy stick–controlled table posi- sum of the exponents,
tioning. Because a slip ring scanner can continuously acquire
views at a fixed z-axis, temporal resolution can be consid- Ni 5 No e 2 ðw1 m1 Þ e 2 ðw2 m2 Þ e 2 ðw3 m3 Þ . . . . . . e 2 ðwn mn Þ :
erably better than expected on the basis of rotation speed; Eq. 5A
images can be updated several times per second by con-
tinually adding new data to the reconstruction dataset and Dividing both sides by No, taking the natural logarithm of
dropping the oldest data. For example, suppose a first im- each side, and dividing by 21 yields the following equa-
age is displayed after one full 360 rotation with a 1-s rota- tion:
tion time. After another 1/6 s, another 60 of views has been
acquired and is added to the dataset, and the original 60 of
2 lnðNi =No Þ 5 w1 m1 1w2 m2 1w3 m3 . . . . . . : :1wn mn :
data is dropped. A new image is displayed on the basis of
the latest 360 (in this case, from data collected between Eq. 6A
60 and 420) but is 1/6 s later than the first image. After
another 1/6 s, the image is again updated, from data col- Ni is the measurement obtained by the detector for this ray,
lected between 120 and 480, and so forth. Images are thus and No is known from in-air reference detector measure-
updated at 6 frames per second. ments or from prior calibration scans. The left side of
CT fluoroscopy has proven valuable for interventional Equation 6A is designated the processed data point Ni9.
procedures but creates the potential for significant radiation Each term wimi represents the attenuation occurring within
doses if too many images are acquired at a given location voxel i, which is designated ui, yielding the following equa-
(17). To minimize radiation exposure, a recent trend has been tion:
to replace continuous acquisition with a series of discrete,
rapidly acquired images to guide biopsies. N i 9 5 u1 1 u2 1 u3 1 u4 1 . . . . . . : : 1 un : Eq. 7A

APPENDIX For the ray shown in Figure 3, wi is the voxel width W and
To understand how each measurement can be reduced to is equal for all voxels. However, in the more general case
a sum of the attenuation values in the voxels along the path for other angles, a ray may pass through a voxel at an angle

PRINCIPLES OF CT AND CT TECHNOLOGY • Goldman 127

or only partially through a voxel, so that wi will differ for 9. Ketteringham J, Gempel P. History of computed tomography: 1967–1978—
excerpts from an ongoing study for the National Science Foundation.
each voxel. In any case, because voxel size as well as the Cambridge, MA: Arthur D. Little, Inc.; 1978.
path of each ray is known, wi can be calculated and the 10. Boyd D, Coonrod J, Denhart J, et al. A high pressure xenon proportional
attenuation coefficient mi of voxel can be determined. chamber for x-ray laminographic reconstruction using fan beam geometry. IEEE
Trans Nucl Sci. 1974;21:184–187.
11. Sohval AR, Freundlich D. Plural source computed tomography device with
improved resolution. U.S. patent 4 637 040. 1986.
REFERENCES
12. Iinuma TA, Tateno Y, Umegaki Y, Watanabe E. Proposed system for ultrafast
1. Seibert JA, Boone JM. X-ray imaging physics for nuclear medicine technolo- computed tomography. J Comput Assist Tomogr. 1977;1:494–499.
gists, part 2: x-ray interactions and image formation. J Nucl Med Technol. 13. Boyd DP, Gould RG, Quinn JR, et al. A proposed dynamic cardiac 3-D den-
2005;33:3–18. sitometer for early detection and evaluation of heart disease. IEEE Trans Nucl
2. Bocage AEM. Procede et dispositifs de radiographie sur plaque en mouvement. Sci. 1979;NS-26:2724–2727.
French patent 536464. 14. McCollough CH. Principles and performance of electron beam CT. In:
3. Webb S. Historical experiments predating commercially available computed Goldman LW, Fowlkes JB, eds. Medical CT and Ultrasound: Current
tomography. Br J Radiol. 1992;65:835–837. Technologies and Applications. Madison, WI: Advanced Medical Publishing;
4. Hounsfield GN. Historical notes on computerized axial tomography. J Can Assoc 1995:487–518.
Radiol. 1976;27:135–142. 15. Kalender WA, Seissler W, Klotz E, Vock P. Spiral volumetric CT with single-
5. Bull J. History of computed tomography. In: Newton TH, Potts DG, eds. Technical breath-hold technique, continuous transport and scanner rotation. Radiology.
Aspects of Computed Tomography. St. Louis, MO: C.V. Mosby Co.; 1981:3835– 1990;176:181–183.
3849. 16. Toki Y. Principles of helical scanning. In: Kimura K, Koga S, eds. Basic
6. Hounsfield GN. Computerized transverse axial scanning (tomography), part 1: Principles and Clinical Applications of Helical Scan: Application of Continuous
description of system. Br J Radiol. 1973;46:1016–1022. Rotation CT. Tokyo, Japan: Iryokagakusha; 1993:110–120.
7. Introduction to Computed Tomography. General Electric Co.; 1976. 17. Silverman SG, Tuncali K, Douglass FA, et al. CT fluoroscopy-guided abdominal
8. Ledley RS, Di Chiro G, Luessenhop AJ, Twigg HL. Computerized transaxial interventions: techniques, results and radiation exposure. Radiology. 1999;212:
x-ray tomography of the human body. Science. 1974;186:207–221. 673–681.

128 JOURNAL OF NUCLEAR MEDICINE TECHNOLOGY • Vol. 35 • No. 3 • September 2007