Designation: E1441 − 11

Standard Guide for

Computed Tomography (CT) Imaging1
This standard is issued under the fixed designation E1441; the number immediately following the designation indicates the year of
original adoption or, in the case of revision, the year of last revision. A number in parentheses indicates the year of last reapproval. A
superscript epsilon (´) indicates an editorial change since the last revision or reapproval.
This standard has been approved for use by agencies of the U.S. Department of Defense.

1. Scope* 1.4 Standard practices and methods are not within the
1.1 Computed tomography (CT) is a radiographic method purview of this guide. The reader is advised, however, that
that provides an ideal examination technique whenever the examination practices are generally part and application
primary goal is to locate and size planar and volumetric detail specific, and industrial CT usage is new enough that in many
in three dimensions. Because of the relatively good penetra- instances a consensus has not yet emerged. The situation is
bility of X-rays, as well as the sensitivity of absorption cross complicated further by the fact that CT system hardware and
sections to atomic chemistry, CT permits the nondestructive performance capabilities are still undergoing significant evo-
physical and, to a limited extent, chemical characterization of lution and improvement. Consequently, an attempt to address
the internal structure of materials. Also, since the method is generic examination procedures is eschewed in favor of
X-ray based, it applies equally well to metallic and non- providing a thorough treatment of the principles by which
metallic specimens, solid and fibrous materials, and smooth examination methods can be developed or existing ones
and irregularly surfaced objects. When used in conjunction revised.
with other nondestructive evaluation (NDE) methods, such as 1.5 The principal advantage of CT is that it nondestructively
ultrasound, CT data can provide evaluations of material integ- provides quantitative densitometric (that is, density and geom-
rity that cannot currently be provided nondestructively by any etry) images of thin cross sections through an object. Because
other means. of the absence of structural noise from detail outside the thin
1.2 This guide is intended to satisfy two general needs for plane of inspection, images are much easier to interpret than
users of industrial CT equipment: (1) the need for a tutorial conventional radiographic data. The new user can learn quickly
guide addressing the general principles of X-ray CT as they (often upon first exposure to the technology) to read CT data
apply to industrial imaging; and (2) the need for a consistent set because the images correspond more closely to the way the
of CT performance parameter definitions, including how these human mind visualizes three-dimensional structures than con-
performance parameters relate to CT system specifications. ventional projection radiography. Further, because CT images
Potential users and buyers, as well as experienced CT are digital, they may be enhanced, analyzed, compressed,
inspectors, will find this guide a useful source of information archived, input as data into performance calculations, com-
for determining the suitability of CT for particular examination pared with digital data from other NDE modalities, or trans-
problems, for predicting CT system performance in new mitted to other locations for remote viewing. Additionally, CT
situations, and for developing and prescribing new scan pro- images exhibit enhanced contrast discrimination over compact
cedures. areas larger than 20 to 25 pixels. This capability has no
classical analog. Contrast discrimination of better than 0.1 % at
1.3 This guide does not specify test objects and test proce-
three-sigma confidence levels over areas as small as one-fifth
dures for comparing the relative performance of different CT
of one percent the size of the object of interest are common.
systems; nor does it treat CT inspection techniques, such as the
best selection of scan parameters, the preferred implementation 1.6 With proper calibration, dimensional inspections and
of scan procedures, the analysis of image data to extract absolute density determinations can also be made very accu-
densitometric information, or the establishment of accept/reject rately. Dimensionally, virtually all CT systems provide a pixel
criteria for a new object. resolution of roughly 1 part in 1000 , and metrological
algorithms can often measure dimensions to one-tenth of one
pixel or so with three-sigma accuracies. For small objects (less
1
This guide is under the jurisdiction of ASTM Committee E07 on Nondestruc- than 100 mm (4 in.) in diameter), this translates into accuracies
tive Testing and is the direct responsibility of Subcommittee E07.01 on Radiology of approximately 0.1 mm (0.003 to 0.005 in.) at three-sigma.
(X and Gamma) Method.
For much larger objects, the corresponding figure will be
Current edition approved July 1, 2011. Published July 2011. Originally approved
in 1991. Last previous edition approved in 2005 as E1441 - 00(2005). DOI: proportionally greater. Attenuation values can also be related
10.1520/E1441-11. accurately to material densities. If details in the image are

*A Summary of Changes section appears at the end of this standard

Copyright © ASTM International, 100 Barr Harbor Drive, PO Box C700, West Conshohocken, PA 19428-2959. United States

1
 E1441 − 11
known to be pure homogeneous elements, the density values consuming. Thus, less than 100 % CT examinations are often
may still be sufficient to identify materials in some cases. For necessary or must be accommodated by complementing the
the case in which no a priori information is available, CT inspection process with digital radiographic screening. One
densities cannot be used to identify unknown materials partial response to this problem is to use large slice thicknesses.
unambiguously, since an infinite spectrum of compounds can This leads to reduced axial resolution and can introduce partial
be envisioned that will yield any given observed attenuation. In volume artifacts in some cases; however, this is an acceptable
this instance, the exceptional density sensitivity of CT can still tradeoff in many instances. In principle, this drawback can be
be used to determine part morphology and highlight structural eliminated by resorting to full volumetric scans using planar
irregularities. detectors instead of linear detectors (see (1) under 6.5.1.5).
1.7 In some cases, dual energy (DE) CT scans can help 1.11 Complete part examinations demand large storage
identify unknown components. DE scans provide accurate capabilities or advanced display techniques, or both, and
electron density and atomic number images, providing better equipment to help the operator review the huge volume of data
characterizations of the materials. In the case of known generated. This can be compensated for by state-of-the-art
materials, the additional information can be traded for im- graphics hardware and automatic examination software to aid
proved conspicuity, faster scans, or improved characterization. the user. However, automated accept/reject software is object
In the case of unknown materials, the additional information dependent and to date has been developed and employed in
often allows educated guesses on the probable composition of only a limited number of cases.
an object to be made. 1.12 Units—The values stated in SI units are to be regarded
1.8 As with any modality, CT has its limitations. The most as standard. The values given in parentheses are mathematical
fundamental is that candidate objects for examination must be conversions to inch-pound units that are provided for informa-
small enough to be accommodated by the handling system of tion only and are not considered standard.
the CT equipment available to the user and radiometrically 1.13 This standard does not purport to address all of the
translucent at the X-ray energies employed by that particular safety concerns, if any, associated with its use. It is the
system. Further, CT reconstruction algorithms require that a responsibility of the user of this standard to establish appro-
full 180 degrees of data be collected by the scanner. Object size priate safety and health practices and determine the applica-
or opacity limits the amount of data that can be taken in some bility of regulatory limitations prior to use.
instances. While there are methods to compensate for incom-
plete data which produce diagnostically useful images, the 2. Referenced Documents
resultant images are necessarily inferior to images from com- 2.1 ASTM Standards:2
plete data sets. For this reason, complete data sets and E1316 Terminology for Nondestructive Examinations
radiometric transparency should be thought of as requirements. E1570 Practice for Computed Tomographic (CT) Examina-
Current CT technology can accommodate attenuation ranges tion
(peak-to-lowest-signal ratio) of approximately four orders of
magnitude. This information, in conjunction with an estimate 3. Terminology
of the worst-case chord through a new object and a knowledge 3.1 Definitions—CT, being a radiographic modality, uses
of the average energy of the X-ray flux, can be used to make an much the same vocabulary as other X-ray techniques. A
educated guess on the feasibility of scanning a part that has not number of terms are not referenced, or are referenced without
been examined previously. discussion, in Terminology E1316. Because they have mean-
1.9 Another potential drawback with CT imaging is the ings or carry implications unique to CT, they appear with
possibility of artifacts in the data. As used here, an artifact is explanation in Appendix X1. Throughout this guide, the term
anything in the image that does not accurately reflect true “X-ray” is used to denote penetrating electromagnetic radia-
structure in the part being inspected. Because they are not real, tion; however, electromagnetic radiation may be either X-rays
artifacts limit the user’s ability to quantitatively extract density, or gamma rays.
dimensional, or other data from an image. Therefore, as with 3.2 Acronyms:
any technique, the user must learn to recognize and be able to 3.2.1 BW—beam width.
discount common artifacts subjectively. Some image artifacts 3.2.2 CDD—contrast-detail-dose.
can be reduced or eliminated with CT by improved engineering
practice; others are inherent in the methodology. Examples of 3.2.3 CT—computed tomography.
the former include scattered radiation and electronic noise. 3.2.4 CAT—computerized axial tomography.
Examples of the latter include edge streaks and partial volume 3.2.5 DR—digital radiography.
effects. Some artifacts are a little of both. A good example is 3.2.6 ERF—edge response function.
the cupping artifact, which is due as much to radiation scatter
(which can in principle be largely eliminated) as to the 3.2.7 LSF—line spread function.
polychromaticity of the X-ray flux (which is inherent in the use
of bremsstrahlung sources). 2
For referenced ASTM standards, visit the ASTM website, www.astm.org, or
contact ASTM Customer Service at [email protected]. For Annual Book of ASTM
1.10 Depending on the technology of the CT system, Standards volume information, refer to the standard’s Document Summary page on
complete three-dimensional CT examinations can be time the ASTM website.

2
 E1441 − 11
3.2.8 MTF—modulation transfer function. problem (1)3 and is the result of several different interaction
mechanisms. For industrial CT systems with peak X-ray
3.2.9 NDE—nondestructive evaluation.
energy below a few MeV, all but a few minor effects can be
3.2.10 PDF—probability distribution function. accounted for in terms of the sum of just two interactions:
3.2.11 PSF—point spread function. photoelectric absorption and Compton scattering (1). The
photoelectric interaction is strongly dependent on the atomic
number and density of the absorbing medium; the Compton
4. Summary of Guide
scattering is predominantly a function of the electron density of
4.1 This guide provides a tutorial introduction to the tech- the material. Photoelectric attenuation dominates at lower
nology and terminology of CT. It deals extensively with the energies and becomes more important with higher atomic
physical and mathematical basis of CT, discusses the basic number, while Compton scattering dominates at higher ener-
hardware configuration of all CT systems, defines a compre- gies and becomes more important at lower atomic number. In
hensive set of fundamental CT performance parameters, and special situations, these dependencies can be used to advantage
presents a useful method of characterizing and predicting (see 7.6.2 and references therein).
system performance. Also, extensive descriptions of terms and 5.2.1 One particularly important property of the total linear
references to publications relevant to the subject are provided. attenuation coefficient is that it is proportional to material
density, which is of course a fundamental physical property of
4.2 This guide is divided into three main sections. Sections all matter. The fact that CT images are proportional to density
5 and 6 provide an overview of CT: defining the process, is perhaps the principal virtue of the technology and the reason
discussing the performance characteristics of CT systems, and that image data are often thought of as representing the
describing the basic elements of all CT systems. Section 8 distribution of material density within the object being in-
addresses the physical and mathematical basis of CT imaging. spected. This is a dangerous oversimplification, however. The
Section 8 addresses in more detail a number of important linear attenuation coefficient also carries an energy dependence
performance parameters as well as their characterization and that is a function of material composition. This feature of the
verification. This section is more technical than the other attenuation coefficient may or may not (depending on the
sections, but it is probably the most important of all. It materials and the energies of the X-rays involved) be more
establishes a single, unified set of performance definitions and important than the basic density dependence. In some
relates them to more basic system parameters with a few instances, this effect can be detrimental, masking the density
carefully selected mathematical formulae. differences in a CT image; in other instances, it can be used to
advantage, enhancing the contrast between different materials
5. Significance and Use of similar density.
5.2.2 The fundamental difference between CT and conven-
5.1 This guide provides a tutorial introduction to the theory tional radiography is shown in Fig. 1. In conventional
and use of computed tomography. This guide begins with a radiography, information on the slice plane “P” projects into a
overview intended for the interested reader with a general
technical background. Subsequent, more technical sections
3
describe the physical and mathematical basis of CT technology, The boldface numbers in parentheses refer to the list of references at the end of
this standard.
the hardware and software requirements of CT equipment, and
the fundamental measures of CT performance. This guide
includes an extensive glossary (with discussion) of CT termi-
nology and an extensive list of references to more technical
publications on the subject. Most importantly, this guide
establishes consensus definitions for basic measures of CT
performance, enabling purchasers and suppliers of CT systems
and services to communicate unambiguously with reference to
a recognized standard. This guide also provides a few carefully
selected equations relating measures of CT performance to key
system parameters.
5.2 General Description of Computed Tomography—CT is a
radiographic inspection method that uses a computer to recon-
struct an image of a cross-sectional plane (slice) through an
object. The resulting cross-sectional image is a quantitative
map of the linear X-ray attenuation coefficient, µ, at each point
in the plane. The linear attenuation coefficient characterizes the
local instantaneous rate at which X-rays are removed during
the scan, by scatter or absorption, from the incident radiation as
it propagates through the object (See 7.5). The attenuation of
the X-rays as they interact with matter is a well-studied FIG. 1 A CT Image Versus a Conventional Radiograph

3
 E1441 − 11
single line, “A-A;” whereas with the associated CT image, the which the projection data were collected. The backprojections,
full spatial information is preserved. CT information is derived when enough views are employed, form a faithful reconstruc-
from a large number of systematic observations at different tion of the object. Even in this simple example, with only four
viewing angles, and an image is then reconstructed with the aid projections, the concentration of backprojected rays already
of a computer. The image is generated in a series of discrete begins to show the relative size and position of features in the
picture elements or pixels. A typical CT image might consist of original object.
a 512 by 512 or 1024 by 1024 array of attenuation values for
5.3 System Capabilities—The ability of a CT system to
a single cross-sectional slice through a test specimen. This
image thin cross-sectional areas of interest through an object
resultant two-dimensional map of the slice plane is an image of
makes it a powerful complement to conventional radiographic
the test article. Thus, by using CT, one can, in effect, slice open
inspections. Like any imaging system, a CT system can never
the test article, examine its internal features, record the
duplicate exactly the object that is scanned. The extent to
different attenuations, perform dimensional inspections, and
which a CT image does reproduce the object is dictated largely
identify any material or structural anomalies that may exist.
by the competing influences of the spatial resolution, the
Further, by stacking and comparing adjacent CT slices of a test
statistical noise, and the artifacts of the imaging system. Each
article, a three dimensional image of the interior can be
of these aspects is discussed briefly here. A more complete
constructed.
discussion will be found in Sections 8 and 9.
5.2.3 From Fig. 1, it can be appreciated readily that if an
internal feature is detected in conventional projection 5.3.1 Spatial Resolution—Radiographic imaging is possible
radiography, its position along the line-of-sight between the because different materials have different X-ray attenuation
source and the film is unknown. Somewhat better positional coefficients. In CT, these X-ray coefficients are represented on
information can be determined by making additional radio- a display monitor as shades of gray, similar to a photographic
graphs from several viewing angles and triangulating. This image, or in false color. The faithfulness of a CT image
triangulation is a rudimentary, manual form of tomographic depends on a number of system-level performance factors,
reconstruction. In essence, a CT image is the result of trian- with one of the most important being spatial resolution. Spatial
gulating every point in the plane from many different direc- resolution refers to the ability of a CT system to resolve small
tions. details or locate small features with respect to some reference
5.2.4 Because of the volume of data that must be collected point.
and processed with CT, scans are usually made one slice at a 5.3.1.1 Spatial resolution is generally quantified in terms of
time. A set of X-ray attenuation measurements is made along a the smallest separation at which two points can be distin-
set of paths projected at different locations around the periph- guished as separate entities. The limiting value of the spatial
ery of the test article. The first part of Fig. 2 illustrates a set of resolution is determined by the design and construction of the
measurements made on a test object containing two attenuating system and by the amount of data and sampling scheme used
disks of different diameters. The X-ray attenuation measure- to interrogate the object of interest. The precision of the
ment made at a particular angle, φ1, is referred to as a single mechanical system determines how accurately the views can be
view. It is shown as fφ1(x'), where x' denotes the linear position backprojected, and the X-ray optics determine the fineness of
of the measurement. The second part of Fig. 2 shows measure- the detail that can be resolved. The number of views and the
ments taken at several other angles fφi(x'). Each of the number of single absorption measurements per view determine
attenuation measurements within these views is digitized and the size of the reconstruction matrix that can be faithfully
stored in a computer, where it is subsequently conditioned (for reconstructed. Reducing pixel size can improve spatial resolu-
example, normalized and corrected) and filtered (convolved), tion in an image until the inherent limit set by these constraints
as discussed in more detail in Section 7. The next step in image is reached. Beyond this limit, smaller pixels do not increase the
processing is to backproject the views, which is also shown in spatial resolution and can induce artifacts in the image.
the second part of Fig. 2. Backprojection consists of projecting However, under certain circumstances, reconstructing with
each view back along a line corresponding to the direction in pixels smaller than would otherwise be warranted can be a
useful technique. For instance, when performing dimensional
inspections, working from an image with pixels as small as
one-fourth the sample spacing can provide measurable benefit.
5.3.1.2 It can also be shown that a given CT image is
equivalent to the blurring (convolution) of the ideal represen-
tation of the object with a smooth, two-dimensional Gaussian-
like function called the point-spread-function (PSF). The
specification of the PSF of a system is an important character-
ization of a CT system and can be derived fairly accurately
from the parameters of the CT system. The effect of the PSF is
to blur the features in the CT image. This has two effects: (1)
small objects appear larger and (2) sharp boundaries appear
diffuse. Blurring the image of small objects reduces resolution
since the images of two small point-like objects that are close
FIG. 2 Schematic Illustrations of How CT Works together will overlap and may be indistinguishable from a

4
 E1441 − 11
single feature. Blurring sharp edges reduces the perceptibility periphery. Artifacts occurring at the interfaces between differ-
of boundaries of different materials for the same reason. This ent density materials are more subtle. There is often an
effect is especially important at interfaces between materials, overshoot or undershoot in the density profile at such a density
where the possibility of separations of one type or another are boundary. The interface density profile must be well charac-
of the greatest concern. Thus, knowledge of the PSF of a CT terized so that delaminations or separations are not obscured. If
system is crucial to the quantitative specification of the the interface profile is not well characterized, false positive
maximum resolution and contrast achievable with that system. indications of defects or, more importantly, situations in which
5.3.1.3 It should be noted, since it is a common source of defects go undetected will result. Thus it is important to
misunderstanding, that the smallest feature that can be detected understand the class of artifacts pertinent to the inspection and
in a CT image is not the same as the smallest that can be to put quantitative limits on particular types of artifacts. Some
resolved. A feature considerably smaller than a single pixel can of the artifacts are inherent in the physics and the mathematics
affect the pixel to which it corresponds to such an extent that of CT and cannot be eliminated (see 7.6). Others are due to
it will appear with a visible contrast relative to adjacent pixels. hardware or software deficiencies in the design and can be
This phenomenon, the “partial-volume effect,” is discussed in eliminated by improved engineering.
7.6. The difference between the resolution of a small feature 5.3.3.2 The type and severity of artifacts are two of the
and the resolution of its substructure is of fundamental impor- factors that distinguish one CT system from another with
tance for CT. otherwise identical specifications. The user must understand
5.3.2 Statistical Noise—All images made from physical the differences in these artifacts and how they will affect the
interactions of some kind will exhibit intrinsic statistical noise. determination of the variables to be measured. For instance,
In radiography, this noise arises from two sources: (1) intrinsic absolute density measurements will be affected severely by
statistical variations due to the finite number of photons uncompensated cupping, but radial cracks can be visible with
measured; and (2) the particular form of instrumentation and no change in detectability.
processing used. A good example in conventional radiography
is film that has been underexposed. Even on a very uniform 6. Apparatus
region of exposure, close examination of the film will reveal 6.1 Modern CT systems, both industrial and medical, are
that only a small number of grains per unit area have been composed of a number of subsystems, typically those shown in
exposed. An example of instrumentation induced noise is the Fig. 3. The choice of components for these subsystems depends
selection of coarse- or fine-grain film. If the films are exposed on the specific application for which the system was designed;
to produce an image with a given density, the fine-grain film however, the function served by each subsystem is common in
will have lower statistical noise than the coarse-grain film. In almost all CT scanners. These subsystems are:
CT, statistical noise in the image appears as a random variation 6.1.1 An operator interface,
superimposed on the CT level of the object. If a feature is 6.1.2 A source of penetrating radiation,
small, it may be difficult to determine its median gray level and 6.1.3 A radiation detector or an array of detectors,
distinguish it from surrounding material. Thus, statistical noise 6.1.4 A mechanical scanning assembly,
limits contrast discrimination in a CT image. 6.1.5 A computer system,
5.3.2.1 Although statistical noise is unavoidable, its magni- 6.1.6 A graphical display system, and
tude with respect to the desired signal can be reduced to some 6.1.7 A data storage medium.
extent by attempting to increase the desired signal. This can be 6.2 Operator Interface—The operator interface defines what
accomplished by increasing the scan time, the output of the control the operator has over the system. From the perspective
X-ray source, or the size of the X-ray source and detectors. of the user, the operator interface is the single most important
Increasing the detector and source size, however, will generally subsystem. The operator interface ultimately determines every-
reduce spatial resolution. This tradeoff between spatial resolu- thing from the ease of use to whether the system can perform
tion and statistical noise is a fundamental characteristic of CT.
5.3.3 Artifacts—An artifact is something in an image that
does not correspond to a physical feature in the test object. All
imaging systems, whether CT or conventional radiography,
exhibit artifacts. Examples of artifacts common to conven-
tional radiography are blotches of underdevelopment on a film
or scattering produced by high-density objects in the X-ray
field. In both cases, familiarity with these artifacts allows the
experienced radiographer to discount their presence qualita-
tively.
5.3.3.1 CT artifacts manifest themselves in somewhat dif-
ferent ways, since the CT image is calculated from a series of
measurements. A common artifact is caused by beam harden-
ing and manifests itself as cupping, that is, a false radial
gradient in the density that causes abnormally low values at the FIG. 3 Typical Components of a Computed Tomography (CT)
interior center of a uniform object and high values at the System

5
 E1441 − 11
repetitive scan sequences. In short, the operator interface to the incident flux and coupling them to some type of device
determines how the system is used. that converts optical input to an electrical signal, sensors
6.3 Radiation Sources—There are three rather broad types suitable for CT can be engineered. The light-to-electrical
of radiation sources used in industrial CT scanners: (1) X-ray converter is usually a photodiode or photomultiplier tube, but
tubes, (2) linear accelerators, and (3) isotopes. The first two video-based approaches are also widely employed. Like ion-
broad energy spectra are (polychromatic or bremsstrahlung) ization detectors, scintillation detectors afford considerable
electrical sources; the third is approximately monoenergetic design flexibility and are quite robust. Scintillation detectors
radioactive sources. The choice of radiation source is dictated are often used when very high stopping power, very fast pulse
by precisely the same rules that govern the choice of radiation counting, or areal sensors are needed. Recently, for high-
source for conventional radiographic imaging applications. A resolution CT applications, scintillation detectors with discrete
majority of existing CT scanners use electrical bremsstrahlung sensors have been reported with array spacings on the order of
X-ray sources: X-ray tubes or linear accelerators. One of the 25 µm. Both ionization and scintillation detectors require
primary advantages of using an electrical X-ray source over a considerable technical expertise to achieve performance levels
radioisotope source is the much higher photon flux possible acceptable for CT.
with electrical radiation generators, which in turn allows 6.5 Mechanical Scanning Equipment—The mechanical
shorter scan times. The greatest disadvantage of using an X-ray equipment provides the relative motion between the test article,
source is the beam hardening effect associated with polychro- the source, and the detectors. It makes no difference, at least in
matic fluxes. Beam hardening results from the object prefer- principle, whether the test object is moved systematically
entially absorbing low-energy photons contained in the con- relative to the source and detectors, or if the source and
tinuous X-ray spectrum. Most medical scanners use for a detectors are moved relative to the test object. Physical
source an X-ray tube operating with a potential of 120 to 140 considerations such as the weight or size of the test article
kV. Industrial scanners designed for moderate penetrating should be the determining factors for the most appropriate
ability also use X-ray tubes, but they usually operate at higher motion to use.
potentials, typically 200 to 400 kV. Systems designed to scan 6.5.1 The majority of scan geometries that have been
very massive objects, such as large rocket motors, use high- employed can be classified as one of the following four
energy bremsstrahlung radiation produced by linear accelera- generations. This classification is a legacy of the early, rapid
tors. These sources have both high flux and good penetration, development of CT in the medical arena and is reviewed here
but they also have a broad continuous spectrum and the because these terms are still widely used. The distinctions
associated beam-hardening effect. Isotope sources are attrac- between these early scan geometries are illustrated in Fig. 4.
tive for some applications. They offer an advantage over X-ray 6.5.1.1 First-generation CT systems are characterized by a
sources in that problems associated with beam hardening are single X-ray source and single detector that undergo both linear
nonexistent for the monoenergetic isotopes such as Cesium- translation and rotational motions. The source and detector
137 and Cobalt-60. They have the additional advantages, assembly is translated in a direction perpendicular to the X-ray
which are important in some applications, that they do not beam. Each translation yields a single view, as shown in Fig. 2.
require bulky and energy-consuming power supplies, and they Successive views are obtained by rotating the test article and
have an inherently more stable output intensity. The intensity translating again. The advantages of this design are simplicity,
of available isotopic sources, however, is limited by specific good view-to-view detector matching, flexibility in the choice
activity (photons/second/gram of material). The intensity af-
fects signal-to-noise ratio, and, even more importantly, the
specific activity determines source spot size and thus spatial
resolution. Both of these factors tend to limit the industrial
application of isotopic scanners. Nevertheless, they can be used
in some applications in which scanning time or resolution is
not critical.
6.4 Radiation Detectors—A radiation detector is used to
measure the transmission of the X-rays through the object
along the different ray paths. The purpose of the detector is to
convert the incident X-ray flux into an electrical signal, which
can then be handled by conventional electronic processing
techniques. The number of ray sums in a projection should be
comparable to the number of elements on the side of the image
matrix. Such considerations result in a tendency for modern
scanners to use large detector arrays that often contain several
hundred to over a thousand sensors.
6.4.1 Scintillation Detectors—This type of transducer takes
advantage of the fact that certain materials possess the useful FIG. 4 Four Sketches Illustrating the Evolution of Medical CT
property of emitting visible radiation when exposed to X-rays. Scan Geometries. Each Embodiment is Representative of a Dis-
By selecting fluorescent materials that scintillate in proportion tinct Generation of Instrumentation

6
 E1441 − 11
of scan parameters (such as resolution and contrast), and ability around the X-ray anode. The detectors are positioned opposite
to accommodate a wide range of different object sizes. The the source circle and full rotational scans are created without
disadvantage is a longer scanning time. the need for object or system motion. Because the electron
6.5.1.2 Second-generation CT systems use the same beam can be moved very rapidly, this scanner can attain very
translate/rotate scan geometry as the first generation. The rapid image acquisition rates. This system has been referred to
primary difference is that second-generation systems use a fan variably as fifth generation and sixth generation. It has also
beam of radiation and multiple detectors so that a series of been described as a stationary-stationary scanner. The terms
views can be acquired during each translation, which leads to millisecond CT, ultrafast CT and electron beam CT have also
correspondingly shorter scan times. Like first-generation been used, although the latter can be confusing since the term
systems, second-generation scanners have the inherent flexibil- suggests that the object is exposed to an electron beam.
ity to accommodate a wide range of different object sizes, (2) In volume CT, a cone beam or highly-collimated, thick,
which is an important consideration for some industrial CT parallel beam is used rather than a fan beam, and a planar grid
applications. replaces the linear series of detectors. This allows for much
6.5.1.3 Third-generation CT systems normally use a rotate- faster data acquisition, as the data required for multiple slices
only scan geometry, with a complete view being collected by can be acquired in one rotation. It is computationally more
the detector array during each sampling interval. To accommo- intensive (although high speed computers are now making this
date objects larger than the field of view subtended by the approach practical) and corrections for scatter and hardening
X-ray fan, it is possible to include part translations in the scan effects may be required for sufficient image quality. Large cone
sequence, but data are not acquired during these translations as beam angles may lead to unsharpness at the outer volume
during first- or second-generation scans. Typically, third- elements.
generation systems are faster than their second-generation 6.5.2 A significant factor in driving medical CT systems to
counterparts; however, because the spatial resolution in a use rotate-only scan geometries was the requirement that
third-generation system depends on the size and number of scanning times be short compared to the length of time that a
sensors in the detector array, this improvement in speed is patient can remain motionless or that involuntary internal
achieved at the expense of having to implement more sensors motion can be ignored (that is, seconds). These considerations
than with earlier generations. Since all elements of a third- are not as important for industrial applications in which scan
generation detector array contribute to each view, rotate-only times for specific production-related items can typically be
scanners impose much more stringent requirements on detector much longer (that is, minutes) and the dose to the object is
performance than do second-generation units, where each view often not an important factor. A second-generation scan geom-
is generated by a single detector. etry is attractive for industrial applications in which a wide
6.5.1.4 Fourth-generation CT systems also employ a rotate- range of part sizes must be accommodated, since the object
only scan motion. The difference between third-generation and does not have to fit within the fan of radiation as it generally
fourth-generation systems is that a fourth-generation CT sys- does with third- or fourth-generation systems. A third-
tem uses a stationary circular array of detectors and only the generation scan geometry is attractive for industrial applica-
source moves. The test specimen is placed within the circle of tions in which the part to be examined is well defined and scan
detectors and is irradiated with a wide fan beam which rotates speed is important. To date, first- and fourth-generation scan
around the test article. A view is made by obtaining successive geometries have seen little commercial application, but there
absorption measurements of a single detector at successive may be special situations for which they would be well suited.
positions of the X-ray source. The number of views is equal to The ability of CT to image and quantify internal features makes
the number of detectors. These scanners combine the artifact it the nondestructive examination method of choice for inspect-
resistance of second-generation systems with the speed of ing parts containing complex internal structures or having
third-generation units, but they can be more complex and various internal layers. When 100 %, or a large area, of a part
costly than first-, second-, or third-generation machines, they needs to be inspected using CT, the most economical approach
require that the object fit within the fan of X-rays, and they are would be to use a volumetric CT system employing an area
more susceptible to scattered radiation. detector, assuming the desired image quality and uniformity
6.5.1.5 Several other CT scanner geometries that have been can be obtained.
developed and marketed do not precisely fit the above catego-
6.6 Computer Systems—The computer system(s) performs
ries. However, there is no agreed-upon generation designation
two major tasks: (1) controlling the scan motion, source
for them.4
operation, and data acquisition functions; and (2) handling the
(1) The cine CT system has no mechanical scanning
reconstruction, image display and analysis, and data archival
motion. In this system both the X-ray detector and the X-ray
and retrieval functions.
tube anode are stationary. The anode, however, is a very large
semicircular ring that forms an arc around the Object scan 6.7 Image Display and Processings—Image display and
circle, and is part of a very large, non-conventional X-ray tube. processing are subfunctions of the computer system that
The source of X-rays is moved around the same path as a provide a degree of image interaction not available with
fourth generation CT scanner by steering an electron beam conventional radiography. The mapping between the pixel
linear attenuation coefficient and the displayed intensity of the
pixel can be changed to accommodate the best viewing
4
Medcyclopaedia.com: GE Healthcare, Bio-Sciences, Europe. conditions for a particular feature. Image processing functions

7
 E1441 − 11
such as statistical and densitometric analyses can be performed their means of production and is dependent only on their
on an image or group of images. The digital nature of the image energy. For this reason, this document refers to penetrating
allows major advances in the way data are processed, analyzed, radiation in the energy range from a few keV to many MeV as
and stored. This process of mapping reconstructed pixel values X-rays, regardless of how they are produced.
to displayed pixel values is shown in Fig. 5. 7.2.1 X-rays can in theory interact with matter in only four
6.8 Archival Data Storage—Information such as image ways: they can interact with atomic electrons; they can interact
data, operating parameters, part identification, operator with nucleons (bound nuclear particles); they can interact with
comments, slice orientation, and other data is usually saved electric fields associated with atomic electrons, or atomic
(archived) in a computer-readable, digital format on some type nuclei, or both; or they can interact with meson fields surround-
of storage medium. The advantage of saving this material in ing nuclei. In theory, an interaction can result in only one of
computer-readable format rather than in simple hardcopy form three possible outcomes: the incident X-ray can be completely
is that it would take dozens of pictures of each slice at different absorbed and cease to exist; the incident X-ray can scatter
display conditions to approximate the information contained in elastically; or the incident X-ray can scatter inelastically. Thus,
a single CT image. Also, images of samples made with old and in principle, there are twelve distinct ways in which photons
new data sets can be compared directly, and subsequent can interact with matter (see Fig. 6). In practice, all but a
changes in reconstruction or analysis procedures can be reap- number of minor phenomena can be explained in terms of just
plied to saved data or images. a few principal interactions; these are highlighted in Fig. 6.
6.9 These elements are the basic building blocks of any CT Some of the possible interactions have yet to be physically
system. Each CT system will have its own particular set of observed.
features. It is the responsibility of the user to understand these 7.2.2 The photon-matter interactions of primary importance
differences and to select the system most appropriate for the to radiography are the ones which dominate observable phe-
intended application. nomenon: photoelectric effect, Compton scattering, and pair
production. Their domains of relative importance as a function
7. Theoretical Background of photon energy and material atomic number are shown in Fig.
7.1 Background—This section will cover the theoretical 7. At energies below about 1 MeV, pair production is not
background associated with CT. First, the means of penetrating allowed energetically; and X-ray interactions with matter are
radiation interaction will be discussed. Second, the specifics of dominated by processes involving the atomic electrons. Of the
CT will be delineated. other possible interactions (see Fig. 6), Rayleigh scattering is
typically small but non-negligible; the rest are either energeti-
7.2 X-Ray Interactions—Penetrating radiation is classified cally forbidden or insignificant. At energies above 1 MeV, pair
according to its mode of origin. Gamma rays are produced by production is energetically allowed and competes with Comp-
nuclear transitions and emanate from the atomic nucleus. ton scattering. Of the other possible interactions, photo-
Characteristic X-rays are produced by atomic transitions of disintegration is typically negligible in terms of measurable
bound electrons and emanate from the electronic cloud. Con- attenuation effects, but at energies above about 8 MeV can lead
tinuous X-rays, or bremsstrahlung, are produced by the accel- to the production of copious amounts of neutrons. The rest of
eration or deceleration of charged particles, such as free the interactions are either energetically forbidden or insignifi-
electrons or ions. Annihilation radiation is produced by the cant.
combination of electron-positron pairs and their subsequent
7.2.3 The three principle interactions are schematically
decomposition into pairs of photons. All evidence suggests that
illustrated in Fig. 8. With the photoelectric effect (see Fig. 8),
the interaction of these photons with matter is independent of
an incident X-ray interacts with the entire atom as an entity and
is completely absorbed. To conserve energy and momentum,
the atom recoils and a bound electron is ejected. Although the
subsequent decay processes lead to the generation of charac-
teristic X-rays and secondary electrons, these are not consid-
ered part of the photoelectric effect. As can be seen in Fig. 7,
the photoelectric effect predominates at low energies. Photo-
electric absorption depends strongly upon atomic number,
varying approximately as z raised to the 4th or 5th power.
7.2.4 With Compton scattering (see Fig. 8), an incident
X-ray interacts with a single electron (which, practically
speaking, is almost always bound) and scatters inelastically,
meaning the X-ray loses energy in the process. This type of
scattering is often referred to as incoherent scattering, and the
FIG. 5 Conceptual Illustration of the Process of Mapping a Large terms are used interchangeably. To conserve energy and
Range of Image Values Onto a Much Smaller Range of Display- momentum, the electron recoils and the X-ray is scattered in a
able Values. Two Important Cases are Shown: the One on the
Left Illustrates the Case of Maximum Image Latitude; the One on
different direction at a lower energy. Although the X-ray is not
the Right Illustrates the Case of Maximum Contrast Over a Nar- absorbed, it is removed from the incident beam by virtue of
row Range of Contrast having been diverted from its initial direction. The vast

8
 E1441 − 11

FIG. 6 X-Ray Interactions with Matter

basic principles of CT, the discussion limits itself to the

examination problem of determining a single image plane
through an object. Separate sections focus on (1) what consti-
tutes an acceptable CT data set, (2) one way in which such a
data set can be collected, and (3) some of the competing effects
that limit performance in practice. The discussion of the
companion task of image reconstruction limits itself to the
problem of reconstructing a single two-dimensional image;
three-dimensional reconstructions are not discussed. The treat-
ment includes the goal of the reconstruction process and one
way in which CT data can be reconstructed.
7.3.2 The task of obtaining a usable data set is reviewed in
7.4 – 7.6. The companion problem of how these data are then
reconstructed to produce an image of the object is reviewed in
FIG. 7 Principal X-Ray Interactions 7.7 and 7.8.
7.4 Radon Transform—The theoretical mathematical foun-
majority of background radiation in and around radiographic dation underlying CT was established in 1917 by J. Radon (4).
equipment is from Compton-scattered X-rays. As can be seen Motivated by certain problems of gravitational physics, Radon
in Fig. 7, Compton scattering predominates at intermediate established that if the set of line integrals of a function, which
energies and varies directly with atomic number per unit mass. is finite over some region of interest and zero outside it, is
7.2.5 With pair production (see Fig. 8), an incident X-ray known for all ray paths through the region, then the value of
interacts with the strong electric field surrounding the atomic the function over that region can be uniquely determined. A
nucleus and ceases to exist, creating in the process an electron- particular function and its associated set of line integrals form
positron pair. Energy and momentum are conserved by the a transform pair; the set of integrals is referred to as the Radon
emerging pair of particles. Although the positrons eventually transform of the function. Radon demonstrated the existence of
interact with electrons, generating annihilation radiation, this an inverse transform for recovering a function from its Radon
secondary effect is not considered part of the pair production transform, providing an important existence theorem for what
process. As can be seen in Fig. 7, pair production predominates later came to be called CT. Over the years, the process of
at high energies. Pair production varies approximately with recovering a function from its Radon transform has been
atomic number as z (z + 1). rediscovered numerous times (5-9).
7.3 CT Technical Background—CT is the science of recov- 7.4.1 In a classic example of the old principle that “like
ering an estimate of the internal structure of an object from a equations have like solutions,” tomography has been demon-
systematic, nondestructive interrogation of some aspect of its strated using many different physical modalities to obtain the
physical properties. Generally, but not always (2), the problem necessary line integrals of some physical parameter. Objects
is kept manageable by limiting the task to a determination of a ranging in size from bacteriophages (10) to supernova (11)
single image plane through the object. If three-dimensional have been studied tomographically using a wide variety of
information is required, it is obtained by comparing and, if physical probes, including X-rays (medical CAT scanners or
necessary, resectioning (3) contiguous cross-sections through simple X-ray CT) (12, 13), sound waves (ultrasonic imaging)
the object of interest. (14, 15), electromagnetic fields (NMR, or, more commonly
7.3.1 In its most basic form, the CT inspection task consists now, MR imaging) (16), ionizing particles (17, 18), and
of measuring a complete set of line integrals involving the biologically active isotopes (SPECT and PET scanners) (19-
physical parameter of interest over the designated cross-section 21). These methods have been used to study many types of
and then using some type of computational prescription, or material properties, such as X-ray attenuation, density, atomic
algorithm, to recover an estimate of the spatial variation of the number, isotopic abundance, resistivity, emissivity, and, in the
parameter over the desired slice. In order to best illustrate the case of living specimens, biological activity.

9
 E1441 − 11

FIG. 8 X-Ray Interaction Mechanisms

7.4.2 The essential technological requirement, and that point on the ray path. In X-ray CT, the fractional transmitted
which these various methods have in common, is that a set of intensity, I/Io, is measured for a very large number of ray paths
systematically sampled line integrals of the parameter of through the object being inspected and is then logged to obtain
interest be measured over the cross-section of the object under a set of line integrals for input to the reconstruction algorithms.
inspection and that the geometrical relationship of these Specifically, the primary measurements, I and Io, are processed,
measurements to one another be well known. Within this often “on the fly,” to obtain the necessary line integrals:
constraint, many different methods of collecting useful data
exist, even for the same imaging modality. However, the * µ ~ s ! ds 5 2ln~ I/I o ! (4)
quality of the resulting reconstruction depends on at least three 7.5.2 To obtain an adequate measure of the line integrals,
major factors: (1) how finely the object is sampled, (2) how highly collimated pencil beams of X-rays are used to make the
accurately the individual measurements are made, and (3) how measurements of the fractional transmittance. In the terminol-
precisely each measurement can be related to an absolute frame ogy of CT, the set of line integrals resulting from a scan of an
of reference. object can be grouped conceptually into subsets referred to as
7.5 Sampling the Radon Transform—Given this general views. Each view corresponds to a set of ray paths through the
background, the discussion here now focuses on the specific object from a particular direction (see Fig. 9). The views are
task of tomographic imaging using X-rays as the inspection also referred to as projections or profiles, while each individual
modality. For monoenergetic X-rays, attenuation in matter is datum within a given projection is referred to as a sample or
governed by Lambert’s law of absorption (22), which holds often simply a data point.
that each layer of equal thickness absorbs an equal fraction of 7.5.3 As previously indicated, the reconstruction problem
the radiation that traverses it. Mathematically, this can be places a number of severe constraints on the data. First, the set
expressed as the following: of line integrals must represent a systematic sampling of the
dI
entire object. If the circle of reconstruction is inscribed in an M
I
5 2µdx (1) by M image matrix, this implies (π/4) M2 unknowns and a need
for at least (π/4) M2 linearly independent measurements. Refs
where: (23-25) have examined the minimum number of views and
I = the intensity of the incident radiation, samples per view necessary to reconstruct an arbitrary object
dI/I = the fraction of radiation removed from the flux as it from data in which the dominant source of noise is photon
traverses a small thickness, dx, of material, and statistics. Since the presence of random noise corrupts the data,
µ = the constant of proportionality. one would expect the minimum sampling requirements to be
In the physics of X-ray attenuation, µ is referred to as the greater than they are for noise-free data as well as to be
linear absorption coefficient. Eq 1 can be integrated easily to
describe X-ray attenuation in the following perhaps more
familiar form (1):
I 5 I o e 2µx (2)

where:
Io = the intensity of the unattenuated radiation, and
I = the intensity of the transmitted flux after it has traversed
a layer of material of thickness x.
7.5.1 If X-rays penetrate a non-homogeneous material, Eq 2
must be rewritten in the more general form:
I 5 I o e 2*µ ~ s ! ds (3)

where the line integral is taken along the direction of FIG. 9 Schematic Illustration of Basic CT Scan Geometry Show-
propagation and µ(s) is the linear absorption coefficient at each ing a Single Profile Consisting of Many Discrete Samples

10
 E1441 − 11
sensitive to the algorithm employed. Surprisingly, most algo- one atom participating at any given moment in time is remote,
rithms in use today can provide stable, high-quality reconstruc- but the sheer numbers of atoms typically involved guarantees
tions for data sets approaching the theoretical minimum a finite emission rate. The number of photons produced per unit
sampling requirements. Typically, data set sizes are on the time varies because of the statistical nature of the radiation
order of one to three times the minimal amount, depending on emission process. The variations have well-defined
the system and the application. Arbitrarily complex objects characteristics, which can be described by what are referred to
require more data than objects with simple geometrical shapes mathematically as Poisson statistics. This ubiquitous radio-
or highly developed symmetries. graphic problem of photon statistics is handled in CT by
7.5.4 The number of views and samples needed depends on integrating (or counting) long enough to keep statistical noise
the approach used and the amount of data required; however, to a diagnostically acceptable level (27, 33). What constitutes
independent of approach, the number of samples per view is an acceptable noise level is defined by the application and can
generally more important than the number of views, and the vary widely.
relative proportion of views and samples should reflect this 7.6.2 Beam hardening is a problem encountered with poly-
principle. Predicting the amount of noise in a CT image chromatic X-ray sources, such as X-ray tubes or linear accel-
reconstructed with an adequate number of samples and views erators (linacs). Such bremsstrahlung sources, as opposed to
is a well-studied problem (23-26); predicting the amount of monoenergetic (that is, isotopic) sources, produce a flux whose
noise when an insufficient number of samples or views, or average radiation energy becomes progressively higher as it
both, is used is more difficult and less well studied (24, 27). propagates through an object because the lower-energy pho-
7.5.5 Second, each line integral must be accurately known. tons are preferentially absorbed with respect to the more
It has been found that errors in the measurement of the energetic ones. This effect compromises the validity of Eq 4
fractional transmittance of even a few tenths of one percent are since µ is no longer associated with a single energy but rather
significant (28). This places strict requirements on the data with an effective energy that is constantly changing along the
acquisition system. As a result, the radiation detectors used in ray path. Although this effect can be partially controlled by
standard X-ray CT systems, along with their associated conscious engineering choices, it is generally a significant
electronics, represent some of the most sophisticated X-ray problem and must be corrected for at some stage in the
sensor technology developed to date. A typical CT system can reconstructive processing (see Refs (34-36) and references
handle a dynamic range (the ratio of peak signal-strength-to- therein).
rms noise) on the order of a million-to-one (29, 30), with a 7.6.3 Another source of difficulties is with the finite width of
linearity of better than 0.5 % (30, 31). the individual pencil beams. A pencil beam of X-rays is
7.5.6 Third, each sample must be referenced accurately to a geometrically defined by the size of the focal spot of the X-ray
known coordinate system. It is useless to have high-precision source and the active area of each detector element. Because
transmission measurements if the exact ray path through the these are finite, each source-detector line-of-sight defines a thin
object to which it corresponds is unknown. This places strict strip rather than an infinitely thin mathematical line. As a
demands on the mechanical equipment. Studies have shown result, each measurement represents a convolution of the
that the angle of each view must be known to within a few desired line integral with the profile of the pencil beam. In
hundredths of a degree, and the linear position of each sample general, the width of the strip integrals is small enough that
within a given projection must be known to within a few tens although some loss of spatial information occurs, no distracting
of micrometres (28). artifacts are generated. The exception occurs when there are
7.5.7 CT equipment has evolved to the stage at which each sharp changes in signal level. The error then becomes signifi-
of these performance requirements can be reasonably well cant enough to produce artifacts in the reconstructed image
satisfied. A state-of-the-art scanner routinely collects millions which manifest themselves in the form of streaks between
of measurements per scan, with each one quantified accurately high-contrast edges in the image. These edge artifacts (32,
and referenced precisely to a specific line-of-sight through the 37-39) are caused by the mathematical fact that the logarithm
object of interest. Once collected, the data are then passed to of the line integral convolved with the profile of the pencil
the reconstruction algorithm for processing. beam (which is what is measured) does not equal the convo-
7.6 Physical Limitations on the Sampling Process—The lution of the beam profile with the logarithm of the line integral
quality of the reconstructed image depends on the quality of the (which is what the reconstruction process desires).
data generated by the scanner. In actual practice, equipment 7.6.4 Unfortunately, edge artifacts cannot be eliminated by
and methods are limited in their ability to accurately estimate simply reducing the effective size of the focal spot or the
line integrals of the attenuation through an object (32). Some of detector apertures, or both, through judicious collimation. As
the more prominent sources of inaccuracy are the following: the strip integrals are reduced to better approximate line
photon statistics, beam hardening, finite width of the X-ray integrals and reduce susceptibility to edge artifacts, count rates
pencil beams, scattered radiation, and electronic and hardware become severely curtailed, which leads to either much noisier
nonlinearities or instabilities, or both. Considerable attention is images or much longer scan times, or both. In practice, the
devoted to managing these problems. pencil beams are engineered to be as small as practicable, and
7.6.1 The penetrating radiation used by CT systems is if further reductions in edge-artifact content are required, these
produced in a number of ways, all of which involve random are handled in software. However, software corrections entail
atomic or subatomic processes, or both. The probability of any some type of deconvolution procedure to correct for the beam

11
 E1441 − 11
profile (32, 37-39) and are complicated by the fact that the transmittance) have evolved. They can be grouped broadly into
intensity profile of the pencil beam has a complex geometrical three classes of algorithms: (1) matrix inversion methods, (2)
shape that varies along the path of the X-rays. finite series-expansion methods, and (3) transform methods.
7.6.5 The same problem occurs when the structure of the The general features of each are described in 7.7.2 – 7.7.8.
object undergoing inspection changes rapidly in the direction 7.7.2 Matrix inversion methods follow naturally from a very
normal to the plane of the scan. When the change is sizeable direct approach to the problem of reconstructing an M by M
over the thickness of the slice, the same mathematics that lead image matrix. At the outset, an M by M matrix consists of a
to the edge artifact produce what in this case is commonly blank matrix of M2 unknown attenuation values; while, on the
referred to as a partial-volume artifact (32, 37-39). It manifests other hand, each measurement can be described in terms of a
itself as an apparent reduction in attenuation coefficient in
linear combination of some fraction of these unknown attenu-
those parts of the image where the transverse structure is
ation values. Thus, from elementary algebraic considerations, a
changing rapidly. In the absence of a priori information,
set of M2 linearly independent measurements can in principle
nothing is known about the spatial variation of object structure
within the plane of the scan, and software corrections are much be solved for the unknown attenuation values. Further, because
more difficult to implement. a set of linear equations can be solved very generally using
7.6.6 Still another source of problems arises from the matrices, one class of algorithms focuses on matrix methods
presence of scattered radiation. When multiple detector ele- (42).
ments are employed, there is always the chance that radiation 7.7.3 Unfortunately, solving for N unknowns using matrices
removed from the incident flux by Compton interactions will involves determining and inverting an N by N matrix. If N is a
be registered in another detector. This scattered radiation, large number, such as M2, the size of the matrix and the
which becomes more severe with higher energies, cannot be inversion task becomes completely intractable with current
easily distinguished from the true signal and corrupts the computer technology. This is not to say that matrix inversion
measurements. This problem can be reduced (40), but not methods are not valuable, but that they should not be judged on
eliminated, through the use of proper collimation. the basis of contemporary commercial merits. Basic research in
7.6.7 The last type of inaccuracy is electronic and mechani- this area is an ongoing enterprise and provides valuable insight
cal nonlinearities and instabilities. These may result from into CT problems (24). However, such methods must await the
correctable engineering deficiencies or basic physical limita- further evolution of computer technology to make their way
tions of the available components. The validity of the data is into commercial CT systems.
compromised in either case. In some cases, the problem can be 7.7.4 When the first CT instruments were introduced in the
corrected (or reduced) in software; in others, it can be fixed
early 1970s, reconstructions were performed with what are
only by reengineering the offending subsystem. Because the
now classified as finite series-expansion algorithms. The origi-
bulk of existing information on this crucial topic is commer-
nal EMI scanner invented by G. N. Hounsfield used such an
cially sensitive and therefore proprietary, the literature is
approach (43). These methods, which included so-called alge-
relatively sparse. All that can be said on these issues here is that
considerable effort is required to keep these types of errors braic reconstruction techniques (44), simultaneous iterative
small compared to other less manageable sources of error, such reconstruction techniques (45, 46), and maximum entropy
as those discussed above. algorithms (47, 48), are rooted in a completely different branch
of mathematics from the transform methods described next.
7.7 Inverting the Radon Transform—The reconstruction Stated simply, these methods iteratively alter the reconstruction
task can be defined as follows: given a set of systematic matrix until a grid of values is obtained which produces line
transmission measurements corrupted by various known and integrals that match the measured data as nearly as possible.
unknown sources of error, determine the best estimate of the Obviously, a large number of figures of merit can be used to
cross-section of the object associated with that data. Cormack determine what constitutes the best match, given the statistical
(8) and, earlier, Radon (4) showed that it is possible to “find a
fluctuations in the data; in addition, great latitude exists in the
real function in a finite region of a plane given its line integrals
implementation of the iterative procedure (see Ref (42) and
along all straight lines intersecting the region.” Cormack later
references therein).
extended this result in a companion paper (41) that described
“a method for determining a variable gamma-ray absorption 7.7.5 While commercial CT systems no longer use iterative
coefficient in a sample from (a finite set of) measurements methods because of their inherent slowness, they offer numer-
made outside the sample.” Although Cormack’s algorithm ous advantages that suggest they could experience a rebirth of
never lent itself well to digital processing, at the time it popularity as computer technology continues to develop: they
provided a valuable existence theorem: it was possible to can be adapted readily to a far broader range of physical
recover a useful estimate of the internal structure of an object modalities and geometries (see, for instance, Refs (49) and
from a finite number of measurements of the X-ray transmis- (50)), they are reported to be less susceptible to edge artifacts
sion through an object of interest. (51), they are the preferred method for handling the complex
7.7.1 Over time, a large number of methods (that is, reconstruction problems of emission CT (18, 52, 53), they are
algorithms) for recovering an estimate of the cross-section of the best way of dealing with limited-angle data (48) or
an object (that is, reconstructing a CT image) from its Radon underdetermined data (too few views or samples) (54, 55), and
transform (that is, the set of measurements of the fractional they can be used when full three-dimensional reconstructions

12
 E1441 − 11
are performed (56, 57), as opposed to merely stacking adjacent eliminates its use in commercial CT systems) but has the virtue
slices. (See the review article by Censor (42) for further of lending itself nicely to optical implementation (68, 69), a
information.) technique that could someday be used to process most CT data.
7.7.6 Transform methods, the third class of restorative These methods are reviewed, along with several tutorials, in
algorithms, are based on analytical inversion formulas. Be- the article by Lewitt (63).
cause they are easy to implement, are fast in comparison to the 7.7.8 For the sake of completeness, it should be mentioned
other methods, and can produce high-quality images, they are that there is also a small class of reconstruction algorithms that
universally used by commercial CT systems. The two primary are a hybrid of transform and series-expansion methods and
types of transform methods are (1) the convolution- hence do not fit logically into either of these two broad groups.
backprojection algorithm (58-60) and (2) the direct Fourier Some examples are described in Ref (63).
algorithm (4, 61), but the so-called ρ-filtered layergram method
has also been used in special situations (62). They are based on 7.8 Convolution-Backprojection Methods—In order to give
the underlying fact that the one-dimensional Fourier transform the user a more intuitive feeling for the reconstruction process,
of a CT projection of an object corresponds to a spoke in the convolution-backprojection algorithm is described. It is
Fourier space of the two-dimensional transform of that object provided to give a sense of how such large amounts of data can
(the so-called Central-Section Theorem or Projection-Slice be processed efficiently into a high-quality image. No effort is
Theorem (63)). Thus, in theory, all that is required in order to made to be mathematically rigorous; the interested reader is
obtain an image by this method is to transform each projection referred to Ref (70) for a particularly readable account and to
as it is collected; place it along its proper spoke in two- Ref (63) for a more detailed, but still lucid, treatment of this
dimensional Fourier space; and when all the views have been algorithm.
processed, take the inverse two-dimensional Fourier transform 7.8.1 First, consider the sequence of steps shown in Fig. 10.
to obtain the final image. This method is called the direct Frame A shows a point object being scanned and the idealized
Fourier transform algorithm. response of a single detector as it traverses the field of view.
7.7.7 Within this general framework, there is considerable Frame B shows each of the many profiles collected during this
latitude concerning which of the steps to conduct in Fourier scan backprojected across an initially blank circle of recon-
space and which to conduct in direct space. The advantages of struction. Backprojection can be thought of as reversing the
each must be weighed against the disadvantages. The direct data collection process. Each sample within a given projection
Fourier algorithm is potentially the fastest method; however, represents the fractional transmittance of a narrow beam of
due to interpolation problems, X-ray CT images have not yet X-rays through the object, which is assumed to be sufficiently
been reported with the same quality as those obtained with the well approximated by small, discrete pixels of constant attenu-
convolution-backprojection method (61, 63, 64). Although ation. During backprojection, the value of each sample in the
some recent work has showed promising results (65), direct profile is numerically added to all of the image pixels that
Fourier techniques are used primarily in applications that participated in the attenuation process for that sample.
collect Fourier transforms of the projections directly, such as Conceptually, backprojection can be thought of as smearing
radio astronomy and magneticresonance (MR) imaging (66, each profile back across the image in the direction of the
67). The convolution-backprojection method (or its twin, the radiation propagation.
filtered-backprojection method) is theoretically not as fast as 7.8.2 Frame C shows the net result of this operation. For a
the direct Fourier method, but it produces excellent images and point object, the profiles superimpose to produce a central
with special-purpose hardware is capable of acceptable recon- spike with a broad skirt that falls off as 1/r (at any radius, the
struction times. The ρ-filtered layergram is impracticable when number of backprojected rays radiating from the center is a
dealing with large amounts of digital data (a deficiency that constant). It is implicitly assumed here that a large number of

FIG. 10 Straight Backprojection

13
 E1441 − 11
profiles have been used; hence the smooth, featureless falloff.
One of the earliest attempts at reconstruction used this ap-
g~r! 5
π2
2 S ~! D
δ r
r
1
2 2
r
(7)
proach (6). The product was a blurry but diagnostically useful
where:
image, at least in the absence, at that time, of a viable
alternative. δ(r) = the Dirac delta function.
7.8.3 Fig. 11 shows an improved version of this basic There is an obvious problem with expressing g (r) in this
approach. Frame A shows the same scan situation depicted in form when working with digital computers. A severe discon-
Fig. 10. In Frame B, however, each profile has been convolved tinuity exists near the origin where, loosely speaking, the delta
with a function that preserves the essential response of the function must in some way be attached to the − 1 ⁄r2 tail.
detector to the presence of the point object but adds a negative However, this expression is presented only to give the reader
tail to beat down the 1/r falloff that occurs with pure back- an idea of the behavior of g(r); the rigorous mathematics of
projection. The result of back-projecting these modified pro- how such functions are handled digitally in practice are treated
files is schematically illustrated in Frame C, where the point in the literature (see Refs (63) and (70) and references therein).
object is shown reconstructed in much sharper detail. This 7.8.6 In words, Eq 6 and Eq 7 say that µ(x, y) can be
so-called convolution-back-projection method is the method recovered from a complete set of line integrals, P(θ, ρ), by first
used by virtually all commercial CT systems. It is easy to convolving each projection with a special function, g (that is,
implement with digital techniques, is numerically robust, and is the integral over η in Eq 6) and then backprojecting each
adaptable to special-purpose computer equipment, such as convolved view to obtain the final image (that is, the integral
array processors or hardwired back-projectors. over θ in Eq 6). Convolving the views with the function, g,
7.8.4 To obtain an idea of how this appears mathematically, given in Eq 7 accomplishes two tasks: (1) the first term is just
the results of Eq 4 are rewritten in the following form: the polar-coordinate version of the delta function and serves to
preserve the basic profile of each view; and (2) the second term
P ~ θ, ρ ! 5 2ln@ I ~ θ,ρ ! /I o # 5 * µ ~ x, y ! ds (5)
corrects for the blurring introduced by the back-projection
As before, I represents a single ideal measurement, but it has algorithm. In CT terminology, if the convolution is conducted
been rewritten to explicitly recognize that the detector is in direct space (that is, the inner integral in Eq 6 is evaluated
oriented with respect to the object at some angle, θ, and some directly), the method is called convolution-backprojection; if it
position, ρ, as indicated in Fig. 9. Io is the unattenuated signal is conducted in Fourier space (which is generally a much faster
level, µ(x, y) is the two-dimensional distribution of the linear way to do it), the method is called filtered-backprojection. This
attenuation coefficient of the object, and ds is an element of distinction is frequently overlooked, and the two terms are
distance along the X-ray path through the object at angle θ and often used interchangeably.
position ρ. The values of I(θ, ρ) are normalized to unity and
logged to yield a set of estimated line integrals through the 8. Interpretation of Results
object, P(θ, ρ).
8.1 Technical Objectives—The goal of a CT X-ray imaging
7.8.5 With this notation, the convolution-backprojection
system is to nondestructively produce internal images of
process schematically shown in Fig. 11 can be written as
objects with sufficient detail to detect crucial features. The task
follows:
of the CT user is to specify the system that will satisfy a
`
π particular need and to verify that the specification is met. The
µ ~ x, y ! 5 * * P ~ θ, ρ ! g ~ ρ 2 η ! dηdθ
o
2` (6) visibility of a feature in a CT image depends on the difference
where: in X-ray attenuation between the feature and its background,
size of the feature, size of the background object, X-ray optics,
g = the convolution function of the shape-theoretical form: number of samples collected, X-ray exposure, and numerous

FIG. 11 Convolution Backprojection

14
 E1441 − 11
other factors. To predict accurately the performance of a given flux than lower energy systems. The optimum tradeoff clearly
system in specific application requires a very complicated depends, to a great extent, on the specific application.
modeling process. However, many researchers have shown that 8.2.2 Contrast in CT has been defined historically as the
detectability obeys some fairly simple rules and can be percent difference of a feature from a background material.
expressed as a function of system noise, system resolution, size
and composition of the background object, and size and Contrast, % 5
? µ 2 µ ? 3 100 %
f b
(8)
µb
composition of the feature.
8.1.1 It will be shown in the following sections how these This expression has the disadvantage of being infinite for a
rules can be used to help specify a CT system as well as how feature in air, for which µb is effectively zero, but it is
they can be used to verify a specification. First, some back- convenient for comparing the contrast of different materials in
ground is presented to help the user understand the roles of CT a given background. It should be noted that this definition for
system resolution and noise in detectability. Contrast is defined contrast assumes that the feature in question extends through-
in 8.2. The effect of system resolution on contrast is discussed out the thickness of the CT slice. If the feature has thickness h
in 8.3. The effect of system noise on contrast is discussed in but is imaged with a slice of larger thickness t, the contrast is
8.4. The findings of various researchers that relate contrast further reduced by the factor h/t.
detectability (with a 50 % confidence level) to object size and 8.2.3 If the CT imaging system did not introduce
system noise are presented in 8.5. The results of the previous degradation, a profile through the center of the feature shown
sections are combined in 8.6 to aid the user in specifying a in Fig. 13(a) would have the crisp shape shown in Fig. 13(b).
system for a particular need. The user is shown how to measure Probability-distribution functions PDF(µf) and PDF(µb), which
the performance of an existing system in 8.7. describe the probabilities of finding a given value µ inside the
feature and inside the background, respectively, are plotted in
8.2 Contrast—The quantity that is reconstructed in X-ray
Fig. 13(c). In the absence of degradation, only the value µb
CT imaging is the linear attenuation coefficient, µ, usually
appears in the background, and only the value µf appears in the
within a two-dimensional slice defined by the thickness of the
feature, with each normalized to unit probability. The contrast
X-ray beam. It is measured in units of cm−1 and is directly
difference, ∆µ, is simply given by:
proportional to the electron density of the material. To be
distinguished, a feature must have a linear attenuation ?
∆µ 5 µ f 2 µ b ? (9)
coefficient, µf, that is sufficiently different from the linear
attenuation coefficient of its background material, µb.
8.2.1 Linear attenuation coefficients are functions of the
incident X-ray energy, E. Fig. 12 shows the functional energy
dependence of the X-ray linear attenuation coefficients of two
hypothetical materials, µf and µb. It is seen that the degree of
contrast, ∆µ, between two materials varies greatly as a function
of the energy. (For simplicity in these discussions, the X-rays
used are assumed to have a single energy, E, or to be
approximated by some mean energy, Ē, if a spectrum of
energies is used.) The X-ray energy is an important parameter
that must be chosen for a given scan specification. It would
seem advantageous to choose a low energy to maximize
contrast; however, the attenuation coefficient is large for low
energies, and this results in poor X-ray transmission and high
system noise, which is detrimental to good detectability. Also,
higher-energy systems usually have significantly higher X-ray

FIG. 13 (a) A Sketch Illustrating a CT Reconstruction of a Small

Feature of Attenuation Coefficient µf Embedded in the Center of
a Background Material of Attenuation Coefficient µb; (b) A Plot of
the CT Density Profile Through the Feature in (a); and (c) A
FIG. 12 A Sketch Illustrating the Dependence of Contrast Differ- Probability-Distribution Function (PDF) for the Attenuation Coef-
ence ∆µ Upon the Energy of Incident X-Rays ficients Found in (a)

15
 E1441 − 11
As resolution and noise are introduced into the discussion,
the effect of each on the profile of Fig. 13(b) and the PDF of
Fig. 13(c) will be monitored.
8.3 Resolution—The finite number and width of the X-ray
beams causes the blurring of a feature, which can alter both the
shape of the feature and the resolvability of multiple features.
This blurring also affects the perceived contrast, especially of
small features. To a first approximation, it is possible to derive
a two-dimensional blurring function that can be convolved
with an object to produce the equivalent of a CT image. This
blurring function, called the point-spread function (PSF), is the
response of the system to an ideal point object. In this
discussion, it will be assumed that the PSF has circular
symmetry and is uniform throughout the image. In this case,
the modulus of the one-dimensional Fourier transform of a
profile through the PSF gives the modulation-transfer function
(MTF) (71) of the system, which describes the differential
ability of the system to reproduce spatial frequencies. In
general, low frequencies (large, homogeneous features) are
reproduced more faithfully than high frequencies (small fea- FIG. 14 A Sketch Illustrating the Geometry of the X-Ray Beam of
tures). a CT Scanner, Where d is the Detector Width, a is the X-Ray
Source Width, L is the Distance Between the Source and the
8.3.1 First, a simple approximation to the PSF is discussed,
Detector, and q is the Distance Between the Source and the Im-
and its effect on the profile of Fig. 13(b), on the PDF of Fig. aging Point
13(c), and on the effective contrast of small features is
illustrated. Three methods of obtaining the MTF are then
discussed: one theoretical and two experimental. It should be
emphasized that the MTF is not merely a computational
curiosity; it is used both to predict and to measure system
performance.
8.3.2 For the purpose of illustration, the PSF can be ap-
proximated by a cylinder of diameter BW (72) that approxi-
mates the beam width. BW is a function of the detector width
d, the X-ray source width a, the distance between the source
and the detector L, and the distance between the source and the
imaging point q as follows:
=d 2 1 @ a ~ M 2 1 ! # 2
BW. (10)
M

where:
L FIG. 15 A Sketch Illustrating the Two-Dimensional Convolution
M5 (11)
q of a Point-Spread Function (PSF) of Diameter BW with Features
of Varying Diameters: (a) SW < BW, (b) BW = BW, and (c) LW >
The quantities d, a, L, and q are illustrated in Fig. 14. A BW. The Symbol (*) Represents the Convolution of Two Func-
justification for Eq 10 will be given in the discussion of the tions
MTF that follows (see also Ref (72)).
8.3.3 Fig. 15 shows the convolution of the PSF with features
that are smaller than, equal to, and greater than the PSF. Fig. PSF of diameter BW with a larger feature of diameter LW and
15(a) shows the result of convolving the PSF of diameter BW contrast difference ∆µ. The imaged feature will be a truncated
with a smaller feature of diameter SW and contrast difference cone of base (BW + LW) and contrast difference ∆µ. Thus, the
∆µ. The imaged feature will be a truncated cone with base diameters of features much wider than the PSF are affected
(BW + SW) and contrast difference ∆µ(SW/BW).2 Thus, the only slightly, and the contrast in their centers is not altered.
system PSF reduces the contrast of features smaller than the These results will prove very useful later in the discussion on
beam width by the ratio of their areas and increases the width the relationship between detectability and feature size.
of the imaged feature to approximately that of the PSF. Fig. 8.3.4 The fact that the CT imaging process is discrete rather
15(b) shows the result of convolving the PSF of diameter BW than continuous has been ignored thus far. In fact, the projec-
with a feature of diameter BW and contrast ∆µ. The imaged tion data is sampled at some discrete spatial increment, s.
feature will be a cone of base 2BW and maximum contrast Sampling theory dictates that s be BW/2 at most. The presen-
difference ∆µ. Fig. 15(c) shows the result of convolving the tation of the reconstructed image is also discrete. Again,

16
 E1441 − 11
sampling theory holds that pixel size, ∆p, in the reconstructed
image should be equal to or smaller than s to preserve spatial
resolution. In terms of the convolution of Fig. 15, the smallest
feature will occupy at least four (22) pixels, and possibly more.
8.3.5 Fig. 16(a) shows the effects on image fidelity that
convolution with the PSF and discrete sampling has had on the
ideal image of Fig. 12(a). The profile through the feature is
now rounded at the edges. Fig. 16(b), which is a plot of the new
probability-distribution functions (PDFs), shows that the PDF
of the background now has values larger than µb and that the
PDF of the feature has values smaller than µf.
8.3.6 The convolution of multiple features in the image with
the PSF of the system illustrates the concept of the modulation-
transfer function (MTF). Fig. 17 shows a central, one-
dimensional profile of the convolution of the PSF of width
BWwith periodic features of diameter D whose centers are
separated by 2D. These periodic features are dominated by
spatial frequencies of value 1/(2D). Notice that as long as D ≥
BW, the effective contrast (∆µ)e is not reduced; whereas for D
< BW, the effective contrast is reduced. Furthermore, there is FIG. 17 An Illustration of a One-Dimensional Profile Through the
no contrast at all at approximately D = BW/2. This spatial Center of Periodic Features of Varying Diameters Which Have
frequency at approximately 1/BW is called the cut-off fre- Been Convolved With a CT PSF: (a) D > BW, (b) D = BW , (c)
BW/2 < D < BW, and (d) D = BW/2
quency and represents the effective resolution limit of the
system because frequencies above this value are significantly
altered by the system and cannot contribute to a faithful
representation of the object. Fig. 18(d) shows a plot of the ratio However, this method is open to interpretation and is not
of the effective contrast, (∆µ)e, to the true contrast, ∆µ, as a recommended for an impartial system analysis.
function of the spatial frequency 1/(2D). This is the MTF 8.3.7 The following discussion describes how to obtain a
curve. It can be measured experimentally for a real system theoretical expression for the MTF of a hypothetical system.
from scans of spatial gauges similar to those of Fig. 17. The formalium applies to a parallel-beam method of data
collection, but the expressions for fan-beam data collection are
analogous. The method is attributed to Glover and Eisner (71),
who show that the MTF is approximately equal to the
one-dimensional Fourier transform (FT) of a circularly sym-
metric PSF and is given by the following expression:
F CON~ f !
MTF~ f ! 5 F BW~ f ! F MOV~ f ! F INT~ f ! F PIX~ f ! (12)
f

where:
F CON(f) = the FT of the convolution function,
FBW(f) = the FT of the effective beam width,
FMOV(f) = the FT of the data integration factor,
F INT(f) = the FT of the linear interpolation function in the
image reconstruction,
FPIX(f) = the FT of the display function, and
f = the spatial frequency variable. Each of these
factors will be described briefly.
8.3.8 The factor FCON(f)/f is the convolution filter factor,
assuming a reconstruction process of parallel-beam convolu-
tion and backprojection. (The reconstruction process is beyond
the scope of the present discussion.) The interested reader is
referred to Ramachandran (73) or Shepp and Logan (74). The
filter factor due to Ramachandran is used when high resolution
is desired and the contrast is large enough that noise is not an
issue. The filter factor due to Shepp and Logan is used when
noise is high, contrast is low, and high resolution is not the
FIG. 16 (a) A One-Dimensional Profile Through the Center of a
primary objective. The factors for these two filters are given
Feature Convolved with a CT PSF and Pixelized; and (b) A below, where s is the linear spacing between samples in a
Probability-Distribution Function for the Profile in (a) profile:

17
 E1441 − 11
8.3.11 Since data values are computed at discrete points and
the reconstruction process requires values at intermediate
points, some form of interpolation must be conducted. One
common form is linear interpolation whose FT has the follow-
ing form:
sin2 ~ πfs!
F INT~ f ! 5 (17)
~ πfs! 2
8.3.12 Finally, the interpolated data are displayed on a
square grid of width ∆p. Since this representation is equivalent
to a convolution, the MTF is also multiplied by the following
factor:
sin ~ πf∆p !
F PIX~ f ! 5 (18)
πf∆p
8.3.13 Eq 12 is useful for predicting the MTF of a hypo-
thetical system. The relationship between the PSF and the MTF
also suggests a superior method for measuring the MTF of an
existing system. Since the PSF is ideally the system response to
a point function, a point can be imaged, and the PSF can be
obtained directly as a profile of this point image. In practice
however, point objects always have some width. Fortunately, it
can be shown that the one-dimensional profile of a circularly
symmetric PSF is roughly equivalent to a profile taken perpen-
dicular to the two-dimensional response of the system to a line,
the line-spread function (LSF) (75). Although a line is equally
difficult to image, the LSF is well approximated by the first
derivative of the response of the system to an edge, the
FIG. 18 An Illustration of the Procedure for Obtaining the MTF so-called edge-response function (ERF), which is obtained
From a CT Image of a Small Cylinder: (a) Sketch indicating Rela- easily.
tive Orientation of Three Different Line Profiles Through the Cen- 8.3.14 Fig. 18 illustrates the process of obtaining the MTF
ter of the Imaged Cylinder; (b) The Result of Aligning and Aver- experimentally from the image of a simple cylinder. The use of
aging Many Edge Profiles, the Edge-Response Function, ERF; (c)
The System Line-Spread Function, LSF, Obtained by Differentia-
a cylinder (Fig. 18(a)) is preferred because, once its center of
tion of the ERF; and (d) The System Modulation-Transfer mass is determined, profiles through this point are perpendicu-
Function, MTF, Obtained by Discrete Fourier Transformation of lar to the cylinder edge. Many profiles can be aligned and
the LSF averaged to reduce system and quantization noise on the
edge-response function (ERF) (Fig. 18(b)). The LSF is esti-
F R CON~ f ! mated by taking the discrete derivative of the ERF (Fig. 18(c)),
5 1 @ Ramachandran# (13) and its discrete FT is taken to obtain the MTF (Fig. 18(d)).
f
S&L
(Note that, by convention, the height of the MTF is normalized
F CON ~ f ! sin ~ πfs! to unity.) This procedure is easy to execute and not open to
f
5
πfs
@ Shepp and Logan# (14)
misinterpretation.
8.3.9 Yester and Barnes (72) describe the FT of an arbitrary 8.4 Noise—In the previous section, the extent to which the
beam shape as follows, where these quantities are defined in Eq system PSF degrades contrast and resolution has been inves-
10 and Eq 11: tigated. However, no factor has been introduced thus far that

F BW~ F ! 5
sin F G F
πfd
M
sin
πfa~ M 2 1 !
M G (15)
would prevent detection of a feature (except at the cutoff
frequency). In this section, system noise is added to the model
πfd πfa~ M 2 1 ! of system behavior, and its impact on detectability is explored
M M in terms of basic system performance parameters.
8.4.1 It is not possible to build an X-ray CT imager without
They also note that this function can be approximated to a noise. Even if electronic noise and scatter noise are minimized,
good approximation by the FT of a square beam whose width quantum statistics dictates that there will be variation in the
is BW, given previously by Eq 10. number of X-rays detected from the source. The photon noise
8.3.10 Collecting discrete signals from a moving X-ray on the X-ray signal is known to obey Poisson statistics; that is,
source is equivalent to convolution by a square function whose it is characterized by the fact that the variance of the signal is
width is the linear sampling increment s. Its FT is given by the equal to its mean. It is customary to specify noise as the
following: standard deviation, which is the square root of the variance.
sin ~ πfs! This means that if an average of n photons is detected in a
F MOV~ f ! 5 (16)
πfs given sampling period, the number actually recorded in any

18
 E1441 − 11

particular interval will be in the range of n6 =n approximately 8.4.4 Fig. 19(a) shows the effect that noise has on the
70 % of the time. blurred, pixelized image of Fig. 16(a). The noise appears as a
jitter superimposed on the profile of the feature. Fig. 19(b),
8.4.2 The effect that noise has on a CT image is complicated
which shows the new PDFs, illustrates that the spread of
by the reconstruction process. For a parallel-beam scanner
attenuation values has increased and that the two distributions
geometry, Barrett and Swindell (76) show that the noise at the
may overlap. The photon noise on any one sample is Poisson
center of a reconstructed cylinder of radius Ro irradiated by
distributed, but the combination of independent samples is
X-rays of effective energy Ē is given by the following formulas
approximated better by the normal distribution given by the
for the Ramachandran (σR) and Shepp and Logan (σS&L)
following expression:
convolution filters:

σ R.
0.91
σ d @ Ramachandaran# (19)
PDF~ µ ! 5
1
=2πσ
F
exp 2
~ µ 2 µ̄ ! 2
2σ 2 G (24)
s =V
0.71 where:
σ S&L . σ d @ Shepp and Logan# (20) µ̄ = the mean of the distribution, and
s =V σ = the standard deviation.
where: Eq 24 has the advantage of being computationally simple.
V = the number of views or orientations, Fig. 19(c) shows the PDFs redrawn as smooth curves. The
s = the spatial sampling increment, and figure illustrates the fact that 70 % of the values are within 6σ
σd = the standard deviation of the noise on the samples in the of the mean.
profile data. 8.4.5 Fig. 19(c) shows that the contrast will be degraded.
The computation of the noise on the profile data, σd, is The difference in contrast can now be defined to be the
complicated by the fact that the profile data is the natural difference between the mean of the feature, µ̄ f, and the mean of
logarithm of the ratio of the intensity of the unattenuated the background, µ̄ b. However, if a detection threshold is placed
radiation, n, and the detected signal. Also, there is likely to be between the two distributions and they overlap, there will be
additional noise from the detector electronics and scattered instances when pixels within the feature will be counted as
radiation. In a detailed analysis, these contributions must be background and pixels within the background will be counted
included, and they will increase the noise. However, in the as features.
approximation that photon noise dominates, the minimum
possible data noise, σd, is given by the following expression,
where µo(Ē) is the linear attenuation coefficient of the cylinder:

σ d. F 1
nexp@ 22µ o ~ Ē ! R o #
1
1
n G ½
(21)

Notice that the noise decreases with increasing n and

increases with increasing Ro or µo.
8.4.3 Experimentally, the usual process for determining the
standard deviation, σ, for a homogeneous area of a recon-
structed image containing m pixels, each with some value µi, is
to first find the mean value of the set of m pixels:
m
1
µ̄ 5
m (µ
i51
i (22)

and then compute σ as:

F G
m ½

( ~ µ 2 µ̄ !
i51
i
2

σ5 (23)
m21

where:
Σ = summation over the region of interest, and
σ = a measure of the spread of the values of µi about the
mean µ̄.
Hanson (77) shows that σ is not very sensitive to the number
of pixels averaged if m is in the range of 25 ≤ m ≤ 100. The
noise in a reconstructed image does have a positional
dependence, especially near the edges of an object, so ex- FIG. 19 (a) A Line Profile Through the Center of a Noisy Feature
Convolved With a CT PSF and Pixelized: (b) A Probability-
tremely large regions should not be used. Hanson (78) has also Distribution Function for the Profile in (a); and (c) A Continuous
shown that the noise in CT images is not completely Representation of the PDFs of (b) With the Means, Standard
uncorrelated, but the effect on σ is small. Deviations, and Contrast Difference Indicated

19
 E1441 − 11
8.5 Contrast-Detail-Dose (CDD) Curve—In practice, detec- The plot of the contrast required for 50 % discrimination of
tion is not based solely on threshold criteria. Human beings use pairs of features as a function of their diameters in pixels is
visual integration when detecting features, and even computer called a contrast-detail-dose (CDD) curve.
detection processes are likely to use pattern recognition tech- 8.6 Performance Prediction—The detectability limits de-
niques. Thus, detection criteria should be based on the obser- fined by Eq 27 and Eq 29 can be used to estimate the detection
vations of human beings. Several investigators (77, 79, 80) ability of a proposed CT system to detect an object of a given
have reported that the effective contrast, (∆µ)e, which human size and composition. In the interest of simplicity, detectability
beings can detect with a 50 % probability of success, depends will be computed at the center of a uniform cylinder. The noise
on the image noise, σ, and the object diameter, D, according to in a reconstructed cylinder is highest at its center so that this
the following relationship: represents a worst case. Also, many complex objects can be
cσ∆p approximated by a cylinder of the same material and cross-
~ ∆µ ! e . D (25)
sectional area.
8.6.1 The contrast given in Eq 27 and Eq 29 is a function of
where:
µb, which in turn depends on the X-ray source effective energy
∆p = the pixel width, and Ē, the pixel size ∆p, the size of the feature relative to the X-ray
c = a constant in the range of 2 ≤ c ≤ 5. beam width BW, and the noise σ. Many references list linear
8.5.1 It is seen from Fig. 15 that the contrast difference of attenuation coefficients as functions of Ē (81, 82). BW is
features larger than the effective beam width BW is not affected defined by Eq 10 and Fig. 14 in terms of the source width,
by beam convolution. Thus, for large features: detector width, and position of the object. For a parallel-beam
cσ∆p CT scanner, σ is given by Eq 19 and Eq 20 in terms of the
~ ∆µ ! e 5 ∆µ. D @ D..BW# (26) sampling increment s, the number of views V, the cylinder
radius R o, and the number of unattenuated photons incident in
Dividing Eq 26 by µb and multiplying by 100 % gives the each sample n. Once these parameters are specified, it is
formula for percent contrast: possible to plot a detectability graph that will predict the
? µ 2 µ ? 100 % 5 cσ∆p 100 % @ D..BW#
f b
performance of the scanner.
(27) 8.6.2 Fig. 20 shows the detectability graph for an iron
µb Dµ b
cylinder 2.54 cm in radius that is irradiated with 0.8 MeV
8.5.2 Eq 25 has not been tested for features smaller than X-rays. The detectability line for objects of diameter D >> BW
BW. However, the results of Fig. 15 suggest a logical exten- (Eq 27) is represented by a solid line. For ease of analysis, the
sion. Features smaller than BW have effective diameter BW and log10 of the percent contrast has been plotted as a function of
have the contrast reduced by D2/(BW)2. Thus, the detectability the log10 of the feature diameter, D, measured in pixels. The
limit for smaller features can be approximated by the follow- detectability line for objects of diameter D << BW (Eq 29) is
ing: represented by a dashed line. To determine whether a feature of
2
∆µD cσ∆p given diameter, D, and linear attenuation, µf (0.8 MeV), will be
~ ∆µ ! e 5 ~ BW! 2 . BW @ D,,BW# (28)
detected in the center of this iron cylinder, plot the point whose
and the percent contrast is given by the following: ordinate is the percent contrast, 100 % × |µFE (0.8 MeV) − µf

? µ 2 µ ? 100 %. cσBW∆p 100 % @ D,,BW#

f b
(29)
µb D 2µ b

8.5.3 Detectability alone is often not sufficient; features

must be discriminated (detected and resolved). Eq 25 can also
be used to predict the discrimination of pairs of features of
diameter D separated by 2D. (This distance is used because it
is conventional to define resolvability in terms of the classical
Rayleigh sense, which stipulates a 2D separation.) From Fig.
17, it has been shown that in this case, (∆µ)e is given by the
product of the true contrast times the system modulation-
transfer function (MTF). In this case, the perceived contrast,
(∆µ)e, is given by the following expression:

~ ∆µ CDD! e 5 ∆µ CDD 3 MTF 2 D 5 D S D

1 cσ∆p
(30)

Solving Eq 30 for ∆µCDD, dividing by µb, and multiplying by

100 % gives an expression for the percent contrast for thresh-
old (50 %) discrimination: FIG. 20 A Plot Illustrating the Application of the Detectability
Lines and the CDD Curve for a Cylinder of Iron of Radius 2.54
? µ 2 µ ? 100 % 5
f b cσ∆p 3 100 %
@ CDD# (31) cm that is Irradiated By 0.8-MeV X-Ray Photons: A Value of 8.5
µb
MTF
1
2 S D
D Dµ b
was used for the Constant c in Eq 25 and its Derivative Equa-
tions

20
 E1441 − 11
(0.8 MeV)|/µFE (0.8 MeV) and whose abscissa is the diameter presented. Contrast has been defined and the degradation of
D in pixels. If this point falls well to the right of the lines, it contrast by the system point-spread function and the system
will be detected more than 50 % of the time. If it falls to the noise has been discussed. Finally, the use of a simple object has
left, it will not. Remember that the percent contrast must be been recommended to predict and verify the performance of
multiplied by the ratio h/t if the height of the feature h is less CT systems in the detection and discrimination of features in a
than the X-ray slice width t. background of specified size and composition. It must be
8.6.3 It is also possible to plot the theoretical CDD curve emphasized that this formalism is meant to be a simple
specified by Eq 31. The theoretical MTF has been given in Eq indicator of system capabilities and does not address such
12 as a function of specified scanner parameters. The theoreti- complications as detection in the presence of CT artifacts.
cal CDD curve for the iron cylinder is identified in Fig. 20 by
the short dashed curve. To determine whether two features of 9. Guidance for Precision and Bias
diameter D whose centers are separated by 2D can be discrimi-
nated at least 50 % of the time, plot a point whose ordinate is 9.1 Computed tomography (CT) images are well suited for
the percent contrast and whose abscissa is their common use in making quantitative measurements. The magnitude and
diameter in pixels. Determine the position of the point relative nature of the error in CT-based measurements depends very
to the curve. If it lies well to the right of the curve, the features strongly on the particulars of the scanner apparatus, the scan
will be discriminated with at least a 50 % probability. parameters, the object, and the features of interest. Among the
parameters which can be estimated from CT images are feature
8.7 Performance Verification—The detectability and CDD size and shape, feature density contrast, wall thickness, coating
curves for an existing CT scanner can also be plotted from Eq thickness, absolute material density, and average atomic num-
27, Eq 29, and Eq 31 for a cylinder of specified material and ber.
size. The quantity σ/µb is the noise-to-signal ratio at the center
of the cylinder as computed from Eq 22 and Eq 23. The 9.2 The use of such quantitative measurements requires that
function MTF(1⁄2 D) is computed experimentally from a small the errors associated with them be known.
cylinder as described in Fig. 18. Fig. 21 shows a comparison of
NOTE 1—This discussion addresses only the precision and bias of the
the predicted (solid line) and measured (dotted line) CDD measurements, not the noise or artifact in the images themselves.
curves from an existing CT scanner for an iron cylinder 2.54
cm in radius irradiated by an equivalent energy of 0.8 MeV. A 9.3 The precision of the measurements can best be mea-
comparison between the experimental and theoretical MTF sured by seeing the distribution of measurements of the same
curves is shown in Fig. 18d. The agreement between the theory feature under repeated scans, preferably with as much displace-
and experiment is quite good in this case. Because the cylinder ment of the object between scans as is expected in practice.
is relatively small, there is not a large contribution to the noise This ensures that all effects which vary the result are allowed
from scattered radiation. For a large cylinder, scatter will for; such as photon statistics, detector drift, alignment artifacts,
usually make the experimental noise larger than the predicted spatial variation of point-spread-function, object placement,
noise, and the curve will shift upwards. and so forth.
8.8 Conclusion—A simple formalism for the prediction and 9.4 One source of such variation in measurements is uncor-
evaluation of the performance of X-ray CT systems has been rected systematic effects such as gain changes or offset
displacements between different images. Such image differ-
ences can often be removed from the measurement computa-
tion by including calibration materials in the image, which is
then transformed so that the calibration materials are at
standard values. Since air is usually already present in the
image, a single additional calibration material (preferably
similar to the object material, and placed in a standard position
in the image) is often sufficient.
9.5 In addition to random variation, measurements of any
particular feature may also have a consistent bias. This may be
due to artifacts in the image or to false assumptions used in the
measurement algorithm. When determined by measurement of
test objects, such biases can be removed by allowing for them
in the algorithm.
9.6 Examination of the distribution of measurement results
from repeated scans of test objects with known features similar
to those which are the target of the NDE investigation is the
FIG. 21 A Comparison Between Predicted and Measured CDD best method of determining precision and bias in CT measure-
Curves for a Real Scanner; The Object Scanned is an Iron Cylin-
der of Radius 2.54 cm that is Irradiated By 0.8-MeV X-ray Pho-
ments. Once such determinations have been made for a given
tons; A Value of 8.5 was used for the Constant c in Eq 25 and its system and set of objects and scanning conditions; however,
Derivative Equations they can be used to give well-based estimates of precision and

21
 E1441 − 11
bias for objects intermediate in size, composition and form, as tion (MTF); nondestructive evaluation (NDE); pointspread
long as no unusual artifact patterns are introduced into the function (PSF); reconstruction; resolution; scan; X-ray attenu-
images. ation
10. Keywords
10.1 computed tomography (CT); contrast; contrast-detail-
dose (CDD); detectability; imaging; modulation transfer func-

APPENDIX

(Nonmandatory Information)

X1. GLOSSARY

X1.1.1 afterglow—Afterglow varies substantially between that property generated by a CT imaging process.
different types of scintillator, and is negligible in many CT Discussion—The term artifact is usually restricted to repeat-
measurement situations. able discrepancies, with other variations classed as noise. The
Discussion—Afterglow varies substantially between differ- most common tomographic artifacts result from undersampling
ent types of scintillator, and is negligible in many CT measure- the object (where there is object detail finer than the measure-
ment situations. ment spacing), uncorrected physical effects (such as cupping
X1.1.2 air measurement—a reference radiation-intensity from beam hardening), and incorrect calibration of detector
measurement made with no object in the examination region of response or apparatus position.
a tomograph. X1.1.7 attenuation coefficient, X-ray—a measure of the
Discussion—Air measurements are used with radiation- rate at which the material in a particular region attenuates an
intensity measurements through an object to infer opacity. An X-ray beam with a particular spectrum as it passes through.
air measurement is required for each detection element. If the Discussion—Of particular relevance to CT is the linear
radiation source moves relative to the detection elements, a set attenuation coefficient, which is the decrease in radiation
of air measurements will generally be required for each source intensity per unit of distance traveled, for a particular substance
position. and radiation-beam composition. Units for this coefficient are
X1.1.3 analytical reconstruction techniques—methods for typically cm−1. The linear attenuation coefficient is the mass
computing a map of internal CT density from opacity attenuation coefficient multiplied by the mass density of the
measurements, based on mathematical integration techniques substance. The CT density in each pixel of a tomogram is
for directly inverting the Radon transform, which the process basically a linear-attenuation-coefficient value (perhaps with a
of measurement approximates. scaling factor), although artifacts may cause local or global
Discussion—Contrasted to iterative reconstruction tech- deviations. This coefficient is the sum of the coefficients for
niques. Analytical techniques using Fourier transforms are the several physical attenuation processes (scattering, photoelec-
basis of almost all commercial CT reconstructions. tric absorption, and/or pair production), each of which varies
X1.1.4 aperture function, detector—a three-dimensional substantially with the X-ray photon energy and the elemental
function centered on the axis from the radiation source to a composition of the material. The integral of the linear attenu-
detector element, giving the sensitivity of the detector to the ation coefficients along a ray path gives the X-ray opacity for
presence of attenuating material at each position. that ray in the dimensionless natural units called attenuation
Discussion—The detector aperture function gives the extent lengths.
and intensity distribution of each ray around and along the X1.1.8 attenuation length—the dimensionless natural unit
length of its central line. The function is determined by the size of X-ray projection values along rays through an object, in
and shape of the radiation source and of the active region of the terms of the natural logarithm of intensity reduction.
detector, and by relative distance to the source and the detector. Discussion —An opacity of n attenuation lengths implies
The average width of this function in the region of the object that the fraction of a photon beam passing through the object
being examined is an important limit on the spatial resolution without interaction is 1/en. (For multienergetic beams, each
of a CT scan. energy group is weighted by the signal it generates.)
X1.1.5 area detector—an X-ray detection apparatus with X1.1.9 attenuation, X-ray—the process of reduction of
numerous individual elements arranged in a pattern spread over radiation-beam intensity due to interactions during passage
two dimensions, such as a fluoroscopic screen. through matter.
Discussion —This is in contrast to a linear detector array Discussion—Each of the penetrating photons emitted from
such as used in many tomographic systems an X-ray source has a probability of interaction with material
X1.1.6 artifact, CT—a discrepancy between the actual in its path, dependent on the photon energy and on the
value of some physical property of an object and the map of thickness, density, and elemental composition of the material.

22
 E1441 − 11
Almost all of these interactions will result in the photon being energy is transferred to the electron as kinetic energy.
absorbed or scattered so that it will not reach the detector Discussion —The probability of this type of interaction is
toward which it was originally travelling. The photons remain- proportional to the local electron density. For the range of
ing in the primary beam are not reduced in energy or changed photon energies and objects used in normal CT scans, it
in any way; thus X-ray beam attenuation is an all-or-nothing decreases gradually with increasing photon energy, and is
process for the individual photons, unlike the gradual loss of generally the most likely mode of attenuation in light materials
energy by each particle in charged particle beams. or at intermediate (0.1 to 10 MeV) photon energies. Also called
X1.1.10 backprojection—the process of adding to each inelastic scattering.
pixel a contribution from a (possibly interpolated) value X1.1.16 computed tomography (CT)—a nondestructive
associated with a line through it, as part of the process of examination technique in which penetrating-radiation mea-
reconstruction of a CT-density map of an object from measure- surements of the X-ray opacity of an object along many paths
ments through it. are used to compute a cross-sectional CT-density map called a
Discussion—The values to be backprojected are derived tomogram.
from groups of measurements, which are usually organized in Discussion—In the original approach, the measurements are
views, each of which is a projection of the object from one planar views made up of overlapping measurements along rays
direction. For other than parallel-beam views, a weighting from many regularly-spaced directions, all centered on a slice
factor (based on distance from the radiation source for fan- plane. Approaches using a cone beam have also been devel-
beam views) is also used. oped.
X1.1.11 beam hardening—the shift in the proportions of X1.1.17 cone beam—the diverging radiation from a source
the energies in a multienergetic beam of penetrating radiation shaped by collimation into a pattern whose dimensions at any
resulting from the preferential attenuation of the less- given distance from the source are roughly the same in all
penetrating photons. directions, typically directed at an area detector.
Discussion—The beam photons which pass through an Discussion—This is in contrast to a fan beam or a pencil
object without interaction are, on the average, more penetrating beam.
or “harder” than the original set which entered the object.
X1.1.18 cone-beam CT—use of cone-beam X-ray opacity
(Since the penetrating power of X-rays generally increases
measurements from many directions to estimate CT density
with energy, hardening usually increases the average energy of
throughout a three-dimensional volume of an object.
the beam.) Failure to correct for the non-linearity in opacity
Discussion—Using two-dimensional area detectors, mea-
caused by this change in beam composition may cause char-
surements may be made rapidly through all of an object at
acteristic “cupping” or “diagonal” artifacts in tomograms.
once. Such measurements from many directions can be used to
Spectral shifts of this kind can be substantially reduced by
compute CT-density values throughout the volume. The speed,
suitable beam filtration to remove the least-penetrating portions
efficiency, X-ray energy range, resolution, artifacts, noise, and
of the beam.
scatter-rejection capabilities of systems utilizing such cone-
X1.1.12 beam width—the distance normal to the axis of a beam methods can differ substantially from systems using
ray of penetrating radiation over which changes in object fan-beam methods based on linear detector arrays.
opacity will substantially influence the signal generated.
X1.1.19 contrast—the extent to which a parameter of
Discussion —Typically an average value based on the
interest differs for some set of features.
aperture function in the region of the object is taken to
Discussion—Thus the contrast in linear attenuation coeffi-
characterize this parameter. The beam width may differ in
cient (“CT density”) of aluminum (0.33 cm−1) to iron (1.15
different directions due to the shape of the source spot or
cm−1) is − 0.82 cm−1, for photons of 200 KeV. Contrast is often
detector aperture. For fan beams, the beam width in the
stated as the percentage by which the value for one feature is
direction normal to the plane of the fan is called the slice
greater or less than the value of the other (“aluminum has a
thickness.
71 % CT-density contrast to iron at 200 KeV”). Contrasts in the
X1.1.13 CAT—Computed Axial Tomography, an earlier physical properties of different parts of an object may result in
term for what is now known as computed tomography (CT). contrasts in the image densities for tomograms or radiograms.
Discussion—The term “axial” was used to distinguish the Since CT density varies with energy quite differently for
method from focal-plane tomography. different materials, the contrast in tomograms can be strongly
X1.1.14 collimation—the restriction of the possible paths influenced by beam energy, usually increasing with lower
for radiation by placement of absorbing material. energy. Since image noise usually increases with lower energy
Discussion—Collimation near the radiation source is used to even more, image contrast is an incomplete measure of the ease
limit the radiation beam to correspond to the general shape of of distinguishing features; see density resolution and contrast-
the detection apparatus. In some cases, further collimation near detail diagram.
the detector bank or for each detector is used to reduce or X1.1.20 contrast-detail diagram—a diagram showing, for
eliminate scattered radiation from that which will ultimately be a given imaging situation, the contrast at which features of
measured. various sizes (and perhaps shapes) can be distinguished with
X1.1.15 Compton scattering—a type of interaction be- some specified confidence.
tween a photon and an electron, in which part of the photon’s Discussion—Such a CDD summarizes the impact of the

23
 E1441 − 11
noise and blurring in an image on a decision process. Such a X1.1.26 CT number—a quantitative value for CT density,
diagram is most dependable when it represents empirically- generally based on a linear scale between zero for air and a
verified tests conducted under actual operating conditions (thus standard value for a reference material.
including operator performance and effects specific to a par- Discussion—CT-number values for a given object depend on
ticular inspection task), but diagrams computed from measures the radiation spectrum as well as the object characteristics,
of spatial and density resolution can also be useful. especially for materials of different effective atomic number.
X1.1.21 contrast sensitivity—see density resolution. X1.1.27 cupping—an artifact in tomographic images, typi-
X1.1.22 convolution—the transformation of an ordered ar- cally due to uncorrected beam hardening, in which the CT-
ray of numbers (such as a tomographic view) such that, for density values in the interior of an object are reduced compared
each position, a new number is formed from the weighted sum to those near the outside.
of some of the original numbers, with the weighing factors X1.1.28 dark measurement—a calibration measurement
based only on the amount of difference in position. from each detection element made while the radiation source is
Discussion—The array of weighting factors is called a closed or turned off.
convolution kernel. In most cases the weights decrease with Discussion—Dark measurements are used to correct each
increasing distance; the typical tomographic-reconstruction measurement with that detector. Also called offset measure-
kernel is − 1 ⁄d2 for nonzero distances d, with a positive weight ment.
at d = 0 large enough to make the sum of the weights zero. A X1.1.29 density—amount per unit of volume (or, more
process equivalent to convolution can be accomplished effi- rarely, of area or length); especially, the amount of mass per
ciently for large kernels with the Fast Fourier Transform (FFT) unit of volume (mass density), but the term is also used for
by multiplying each value of the transform of the data by the analogous parameters such as electron density or CT density. A
corresponding point of a frequency-space filter which is the different use of the term is for the dimensionless parameters
transform of the kernel. An inverse FFT then converts this optical density and film density, measures of attenuation which
product array into the convolved data. The most common are the logarithms of transmission ratios.
methods of tomographic image computation use the FFT to Discussion—Both usages are relevant in work with radio-
convolve each view of opacity measurements, and then back- graphic measurements, so adding the appropriate modifier
project the resulting filtered line. The − 1 ⁄ d2 kernel transforms when first using the term or when changing meaning is
into a filter proportional to frequency up to a cutoff frequency recommended to avoid confusion. Mass density is often the
determined by the measurement spacing. The precise shape of physical parameter of interest to the investigator in a CT
the filter can be modified to minimize artifacts or to include any examination; electron density can often be directly inferred
other linear filtering desired (see smoothing and sharpening). from CT scans; CT density (closely related to the linear
X1.1.23 crosstalk—a condition in which activity in a mea- attenuation coefficient) is the parameter actually measured. In
surement channel causes spurious activity in another (usually all these cases, a density map over two or three dimensions is
adjacent) channel. used, approximated by values at discrete pixels or voxels.
Discussion—This may be due to scattering of radiation in Optical (or film) density refers to a projection along a ray rather
the detector, or to optical or electromagnetic coupling of the than a value at a single point; in fact, digital radiograms are
signals resulting from detection. Software correction for computed density projections in this sense (based directly on
known crosstalk patterns is often possible. the object’s transmission of X-rays, not a film’s transmission of
X1.1.24 CT—see computed tomography. light). These X-ray projection values which comprise a digital
X1.1.25 CT density—the parameter, related to the action of radiogram differ from film-density values in that high values
each region of an object cross-section in attenuating an X-ray mean less X-ray exposure (zero density is maximum
beam, which is computed for a two- or three-dimensional exposure), and in the more dependable relationship between
region by the computed tomography imaging process. the projection values and the amount of attenuating matter,
Discussion—(Note: the term “density” is used in several since exposure time is calibrated for and such sources of
related but different senses (often without explicit distinction) variation as film characteristics and development history are
in reference to CT: mass density, electron density, optical avoided.
density, and image density, for example, as well as the CT X1.1.30 density resolution, CT—a measure of the extent to
density defined here.) For monoenergetic beams, CT density is which a tomogram or radiogram can be used to detect
proportional to the linear attenuation coefficient of each area of differences in the physical parameter mapped by the image, for
the object for the penetrating radiation used for the through- features of a given size.
the-object X-ray measurements from which the image is Discussion—The limiting factor in CT density resolution is
computed. For multienergetic radiation, where the beam spec- generally the noise in the image averaged over areas of the
trum (and thus the attenuation coefficient) passing through each feature size; this may vary significantly between different
interior point varies with ray direction due to beam hardening, regions of the image. Another important factor is the contrast
CT densities are averages. In some cases, such as objects made that the features show under the scan conditions for this image.
of a single known material, the CT-density measurements or Taking the ratio of some multiple of the standard deviation of
images can be transformed to give values directly in mass the image noise to a typical image density value is a common
density or some other physical parameter which is independent method for quantifying density resolution. Image artifacts may
of the energy spectrum of the radiation used for measurement. also limit resolution in certain cases. Note that the size of the

24
 E1441 − 11
feature and all of the factors which influence image noise and beam energy spectra to separate the effects of a mixture of
contrast (beam energy, object size, scan time, etc.) must be materials.
specified for a comparison of density-resolution values to be Discussion—Because the variation with beam energy of the
meaningful. probability for each type of attenuation process is significantly
X1.1.31 detectability, CT—the extent to which the pres- different for most materials, it is possible to use two different-
ence of a feature can be reliably inferred from a tomographic energy measurements along the same path in an object to solve
inspection image. for energy-independent physical parameters such as electron
Discussion—CT detectability is dependent on the spatial density and average atomic number. A common technique is to
resolution and density resolution of the image, as well as the solve for the amounts of each of two predetermined basis
levels of confidence required that false positives and false materials whose mixture would give the pair of measurements
negatives will be avoided. Features may be detectable even if seen. The separated energy-independent values derived from
they are too small to be resolved, provided their contrast after the measurements can be used to form separate maps (either
blurring is still sufficient. tomograms or radiograms) of each basis material.
X1.1.32 detector—a device which generates a signal corre- X1.1.39 edge response function (ERF)—the graph of CT
sponding to the amount of radiation incident on it. density across an edge which shows how faithfully the image
Discussion—CT detectors are usually arranged in arrays in of a sharp edge is reproduced in a tomogram.
one or two dimensions. Discussion—The image of an edge proceeds in an “S-curve”
X1.1.33 detector spacing—the distance (linear or angular) from a background value through intermediate values (due to
between adjacent radiation collection elements in a detector partial volume effects or reconstruction artifacts) to a limiting
array. value (the interior CT density of the object). The width of the
Discussion—In most scanning systems this spacing deter- intermediate region is a good measure of the spatial resolution
mines one of the dimensions of the measurement spacing, of an image. For images with little edge artifact, such as
although some systems use measurement interlacing to over- tomograms of low-opacity cylinders, the derivative of the edge
come this limitation if their detector spacing is large. response function is a good approximation to the line-spread
function or point-spread function. The normalized Fourier
X1.1.34 digital radiography (DR)—formation of a map of transform of the point-spread function yields the modulation
projected X-ray opacity values through all or part of an object transfer function (MTF), which gives the relative frequency
by digitization of signals derived from measurements of response of the imaging process.
penetrating radiation.
Discussion—Such X-ray opacity maps can be produced by X1.1.40 elastic scattering—an interaction between a pho-
either a cone beam, used with an area detector such as a ton and a bound electron in an atom, in which the photon is
fluoroscopic screen or X-ray film, or by moving the object redirected with negligible loss of energy.
perpendicular to the plane of a fan beam directed at a linear Discussion—The electron is not affected, with the recoil
detector array. Differences in scatter rejection, detection momentum being transferred to the atom as a whole. The effect
efficiency, and total detector active area give these approaches is most pronounced at energies less than the binding energy of
quite different characteristics. All DR techniques benefit from the electron, and its probability decreases with increasing
the great precision and flexibility in display and analysis that energy. Also referred to as coherent scattering or Rayleigh
image-analysis software provides. Radiograms made by tomo- scattering.
graphic scanners are used both for direct object inspection and X1.1.41 electron density—the number of electrons per unit
as “preview scans” to select the slice planes of interest for CT volume.
scans. Discussion—The ratio of electron density to mass density is
X1.1.35 dimensioning accuracy—the extent to which the roughly constant, gradually decreasing from about 3.0 × 1023
actual dimensions of an object correspond to dimensions electrons/gram for light elements (except hydrogen, which is
calculated from an image, such as a tomogram. twice this value) to 2.4 × 1023 for the heaviest ones. Because
Discussion—For objects made of uniform-density materials Compton scattering (the dominant attenuation process in many
with smooth surfaces, it is usually possible to obtain dimen- tomographic scans) is directly proportional to electron density,
sions substantially more accurate than the spatial resolution of many tomograms are actually maps of electron density.
the image, especially if measurements can be averaged along a X1.1.42 false negative—an erroneous assertion of the non-
surface. existence of a condition (such as a defect) by a decision
X1.1.36 display matrix size—the number of horizontal and process, often due to the limited resolution of a tomographic
vertical pixels available for display of images. image.
Discussion—Display matrix size has no direct connection Discussion—See false positive.
with the spatial resolution of a tomographic system; however, X1.1.43 false positive—an erroneous assertion of the exis-
insufficient display matrix size may require the use of image- tence of a condition (such as a defect) by a decision process,
zooming techniques to show images at full resolution. often due to noise or artifact when interpreting tomographic
X1.1.37 DR—see digital radiography. images.
X1.1.38 dual-energy scanning—use of two sets of mea- Discussion—The incidence of false positives (“false
surements through an object taken with differing radiation- alarms”) depends on the decision criteria as well as the image;

25
 E1441 − 11
decreasing the sensitivity of the process will generally decrease X1.1.51 iterative reconstruction techniques—successive-
false positives, for example, but will increase false negatives. approximation methods using X-ray opacity measurements for
An analysis of the expected cost and incidence of each type of computing an object description (typically a map of some
error is required to choose optimal decision criteria for any density parameter), based on sequential adjustments of the
particular inspection process. description to make it consistent with the measurements.
X1.1.44 fan beam—penetrating radiation from a small Discussion—Algebraic reconstruction techniques (ART) are
source, typically directed at a linear detector array, which has of this type. Contrasted with analytical reconstruction tech-
been shaped by collimation into a pattern which is wide in one niques.
direction and narrow in the orthogonal direction. X1.1.52 kernel—the set of numerical weights used in the
Discussion—In fan-beam CT systems, each measurement convolution stage of the image-reconstruction process.
period gives a planar fan of measurements with a common Discussion—The kernel and the associated frequency-space
vertex at the beam spot. Depending on the pattern of object filter are Fourier transforms of each other.
motion, these measurements can be directly handled as fan- X1.1.53 keV—kilo-electron-volts, a measure of energy.
beam views or distributed into parallel-beam views. Contrasted Discussion—The photons used in industrial CT range in
to cone beam and pencil beam collimation. energy from a few keV to several thousand keV.
X1.1.45 field of view (FOV)—the physical size of the area X1.1.54 kV, kVp—kilovolts, a measure of electrical poten-
to be examined which must be subtended by the X-ray beam. tial.
Discussion—If the test object is larger than the FOV or Discussion—CT beams are often formed by accelerating
moves out of the FOV during scanning, unexpected and electrons onto a metal target over voltages ranging from a few
unpredictable artifacts or a measurable degradation of image tens of kV up to several thousand kV. In each such case, the
quality can result. Many methods have been devised to scan bremsstrahlung (braking radiation) photons formed by colli-
objects larger than the largest FOV for which an instrument sions in the target will range in energy from very small values
was designed. up to a value in electron volts equal to the accelerating
X1.1.46 filter, beam—uniform layer of material, usually of potential in volts.
higher atomic number than the specimen, placed between the X1.1.55 laminogram, computed—map of CT-density esti-
radiation source and the film for the purpose of preferentially mates of an object at positions on a two-dimensional surface,
absorbing the softer radiations. formed by backprojecting radiographic data (perhaps after
Discussion—Filters are used in CT scanners to reduce dose, mathematical filtering) onto the surface.
scattered radiation, and beam-hardening effects. Discussion—Typically some blurred off-surface features
X1.1.47 filter, mathematical—a function of spatial fre- remain in a laminographic image. The advantage of the
quency giving weighting factors to use to modify each point of technique is the ability to produce localized three-dimensional
Fourier-transformed functions or numeric arrays. Application CT-density estimates from substantially less data than would
of such a filter to X-ray projection values is usually a step in the be required for full three-dimensional reconstructions. Similar
process of reconstructing CT images. in many respects to the analog process of focal-plane tomog-
Discussion—Use of such a filter with an FFT is a common raphy.
way of implementing a convolution. The filter is the Fourier X1.1.56 limited-data reconstruction—a tomogram formed
transform of the corresponding convolution kernel. from an “incomplete” data set in which the object is sampled
substantially more in some areas or directions than in others.
X1.1.48 focal spot—the region at which the electrons are
Discussion—Many forms of data limitation have been dealt
focussed in an X-ray machine or linear accelerator.
with by special methods, including these types of reconstruc-
Discussion—The size of the resulting beam spot as seen
tions: few-angle (large angular steps between views), limited-
from the object region is an important determinant of the
angle (views missing over some range of directions, as when
aperture function, especially in the region near the radiation
scanning a wall), limited-field (some portion of some views
source. Since the spot does not generally have a sharp edge,
missing, typically due to high opacity or to positioning
quantitative values for spot size will reflect the method used to
constraints), and region-of-interest (views consist of measure-
define it, since the average radius of, for example, the mini-
ments through only a portion of the cross-section).
mum region from which 99 % of the emission comes will be
much larger than, say, the standard deviation of the intensity X1.1.57 line-spread function—see edge response function.
distribution. X1.1.58 linear attenuation coefficient—a measure of the
fractional decrease in radiation beam intensity per unit of
X1.1.49 gantry—the mechanical apparatus in a tomo- distance traveled in the material (cm−1).
graphic scanner which controls the relative movement of the Discussion—The value of this parameter at each point in an
object to be examined and the source and detector mechanisms. object being examined by penetrating radiation depends on the
X1.1.50 ionization detector—a radiation detector in which composition of both the material and of the radiation beam
the signal is produced by the collection of free electrons or ions passing through that region, as well as the density of the
directly produced by the radiation beam. material. Units for this coefficient are typically cm−1. The
Discussion —Examples include xenon gas detectors and linear attenuation coefficient is the mass attenuation coefficient
semiconductors such as mercuric iodide. multiplied by the mass density of the substance. The CT

26
 E1441 − 11
density in each pixel of a tomogram is basically a linear- Discussion—Used in specifications of spatial resolution to
attenuation-coefficient value, although artifacts may cause state the allowed amount of blurring at the specified line
local or global deviations. This coefficient is the sum of the spacing.
coefficients for several physical attenuation processes (see X1.1.66 modulation transfer function (MTF)—a function
scattering, photoelectric absorption, and pair production), each giving the relative frequency response of an imaging system.
of which varies substantially with the photon energy and the Discussion—The MTF is the normalized amplitude of the
elemental composition of the material. The integral of linear Fourier transform of the point spread function.
attenuation coefficient along a ray path gives the X-ray
projection value for that ray, which is measured in the X1.1.67 monitor detector—a detector used to measure
dimensionless natural units called attenuation lengths. variations in the intensity of the source of penetrating radiation
or some other system parameter.
X1.1.59 linear detector array—an array of radiation- Discussion—Also called reference detector.
sensing elements arranged in a one-dimensional sequence,
typically uniformly spaced along an arc or straight line. X1.1.68 monochromatic—another term for monoenergetic,
when applied to beams of X-rays.
X1.1.60 magnification—the increase in the distance be-
X1.1.69 monoenergetic—comprised of photons all having
tween rays as they proceed from the object to the detectors.
the same energy.
Discussion—Equal to the source-detector distance (SDD)
Discussion—X-rays and gamma rays produced by the decay
divided by the source-object distance (SOD). Large magnifi-
of a few radioisotopes, such as Americium-241 and Cesium-
cations are made practicable by the use of microfocus X-ray
137, are essentially monoenergetic. Many theoretical concepts
tubes, which give a very compact aperture function close to the
are defined in terms of monoenergetic beams. See also multi-
source.
energetic.
X1.1.61 mass attenuation coefficient—a measure of the
X1.1.70 multienergetic—comprised of photons with sev-
fractional decrease in radiation beam intensity per unit of
eral different energies.
surface density cm2·gm−1.
Discussion—Radiation produced by bremsstrahlung (sudden
Discussion—The value of this parameter at each point in an
stopping of fast-moving electrons) in X-ray tubes or linear
object being examined by penetrating radiation depends on the
accelerators has a continuous multienergetic spectrum. See
composition of both the material and of the radiation beam
beam-hardening for a discussion of one of the consequences of
passing through that region. This coefficient, which is typically
making opacity measurements with a multienergetic beam.
expressed in units of cm2/g, is independent of the density of the
substance; that is why it is generally what is given in tables X1.1.71 noise—the variation in a measurement (or in an
rather than the related linear attenuation coefficient, which is estimate or image derived from measurements) when it is
the mass attenuation coefficient multiplied by the mass density. repeated under nominally identical conditions.
Discussion—Noise is distinguished from consistent biasing
X1.1.62 mean free path—the average distance traveled by
effects, which are referred to as artifacts in CT images.
an X-ray photon before it is scattered or absorbed by the
Averaging n independent measurements of the same object
material through which it is passing.
generally reduces the noise by a factor of the square root of n,
Discussion—See linear attenuation coefficient for a discus-
as the random effects partially cancel each other. The noise in
sion of the factors involved.
measurements of penetrating radiation has (in addition to a
X1.1.63 measurement interlacing—a tomographic scan- usually-small instrumental component) a photon statistics
ning pattern of object motion and data reordering in which a component determined by the measurement time, object
finely-spaced fan-beam view is formed by interlacing a set of opacity, radiation beam intensity, and detector aperture. In
more-coarsely-spaced fans with a common vertex. radiograms, this noise is almost uncorrelated, and its average
X1.1.64 measurement spacing, CT—the angular and lin- value is inversely proportional to the square root of area for
ear separation between samples in each view and between view features which cover several pixels. Since the originally-
angles. independent measurements for a tomogram are mixed in the
Discussion—The measurement spacing is a basic limit on convolution and backprojection processes, the dependence of
spatial resolution, since it determines the scale at which noise on feature size in a tomographic image is more complex.
reconstruction artifacts become unavoidable. For a fixed num- X1.1.72 non-linearity correction—a function by which a
ber of measurements, artifacts are generally minimized when measured signal is transformed so that the result has a linear
the number of views is about equal to the number of measure- relationship to the property being measured.
ments in each view. The measurement spacing is usually Discussion—The response of a tomographic detector to the
matched with the width of the aperture function to give radiation incident on it will not necessarily be linear. If the
samples which partially overlap but are still mostly indepen- response is monotonic, however, it (or the X-ray projection
dent. values derived from it) can be transformed to linear by a
X1.1.65 modulation—the extent to which the imaged den- function or table computed from an appropriate set of calibra-
sities of adjacent features of a given size or spacing are tion measurements.
resolved in an image, expressed as a percentage of the actual X1.1.73 offset measurement—see dark measurement.
density contrast. X1.1.74 opacity—see X-ray opacity.

27
 E1441 − 11
X1.1.75 pair production—the process whereby a gamma Discussion—The likelihood of photoelectric absorption in-
photon with energy greater than 1.02 MeV is converted directly creases abruptly between energies just below and above the
into matter in the form of an electron-positron pair. Subsequent binding energies of electron shells, as more electrons become
annihilation of the positron results in the production of two available for emission. Other than at these absorption edges,
0.511 MeV gamma photons. the photoelectric effect becomes less likely as the photon
Discussion—The minimum photon energy required for energy increases, decreasing as roughly the cube of the energy.
electron-positron pair production is 1.022 MeV (the rest mass Photoelectric absorption is strongly dependent on atomic
of the two particles); any excess over this threshold goes into number, with interactions with higher-atomic-number elements
the kinetic energy of the particles. After the positron is stopped more likely by about the fourth power of the ratio of atomic
by interactions in the medium, it combines with an electron to numbers (on a per-atom basis; about the third power on a
form two 511 keV photons of annihilation radiation. The per-gram basis). For the photon energies used in industrial CT,
likelihood of pair production, which is directly proportional to photoelectric absorption is usually significant, and sometimes
the square of atomic number, increases with increasing energy, predominant, in both the process of attenuation in the object
unlike the likelihood of scattering or absorption. This results in and the process of measurement of beam intensity in the
minimums in the opacity of matter to photons at energies from detector. The primary competing process is Compton
300 MeV (hydrogen) to 9 MeV (iron) to 3.5 MeV (lead). scattering, with elastic scattering and pair production also
X1.1.76 parallel beam—a mode of arrangement of tomo- significant in some cases.
graphic opacity measurements into sets of measurements made X1.1.82 photon statistics—the variation in intensity of a
along parallel paths. beam of photons (and thus in measurements derived from it)
Discussion—Some scan geometries such as translate-rotate due to the randomness of emission of individual photons.
naturally produce a uniformly-spaced parallel-beam ray set for Discussion—This phenomenon sets a minimum level for the
each detector; other scanning patterns can be used to produce noise in the measurements, especially in cases where the total
parallel-beam views by reordering and/or interpolation. Con- number of photons detected is small. The mathematical theory
trasted to fan beam views. of Poisson statistics is relevant to this case; it predicts that such
X1.1.77 partial-volume artifact—an erroneous feature in a sampling variation will increase the noise in X-ray opacity
tomogram or radiogram due to inconsistent data caused by estimates by a term whose variance is the reciprocal of the
variation in the X-ray opacity on scales smaller than the width number of photons.
of the rays. X1.1.83 pixel (tomographic)—one of a group of discrete
Discussion—Such high-frequency variation will usually re- positions composing a tomogram or related image.
sult in an underestimate of the actual projected density, causing Discussion—In general usage, pixels are the points on an
characteristic artifacts such as low-density lines aligned with image-display surface at which various colors and/or intensi-
straight edges in the object. ties may be shown. In computed tomography, the term is also
X1.1.78 partial-volume effect—the deviation of measured often used to refer to the CT-density estimate computed for the
X-ray projection values from proportionality to average pro- corresponding physical position in the object being inspected.
jected mass along a ray path, when projected mass changes Because of interpolation or compression, the display pixels and
rapidly within the width of the ray. the CT pixels in CT images may not represent areas of equal
Discussion—This effect is most pronounced for rays along size. Pixels are usually arranged in rectangular arrays. The
flat edges. See partial-volume artifact. spatial resolution of a CT system is generally not determined
by CT-pixel spacing, but rather by measurement spacing or by
X1.1.79 pencil beam—beam of penetrating radiation colli-
the aperture function. (If pixel size is so large as to limit
mated so that its dimensions in directions perpendicular to the
resolution, the measured data can be used again to compute a
beam axis are small compared to the source-object distance.
tomogram of a subregion with pixels small enough that their
Discussion—Usually used with a single detector, as in
size is not limiting.) The three-dimensional entity correspond-
“first-generation” CT scans.
ing to the pixel is the voxel, or volume element.
X1.1.80 phantom—test object containing features of
X1.1.84 point-spread function (PSF)—the image of a
known size, spacing, and contrast, which can be scanned to
small isolated point under an image-formation process, nor-
determine spatial or density resolution.
malized to the total integrated density of the point.
Discussion—A variety of such standardized test objects are
Discussion—See spatial resolution, edge-response function,
used with CT scanners for alignment of the systems, for
and modulation transfer function.
calibration of image geometry and density mappings, and/or
for determining the spatial and density resolution which can be X1.1.85 polychromatic—another term for multienergetic,
achieved under specific scanning conditions. In some cases, when applied to beams of X-rays.
small phantoms are included in production scans as image X1.1.86 projection—the integral of a density function,
quality indicators. typically along a straight line or a set of lines.
X1.1.81 photoelectric absorption—a mode of interaction Discussion—The X-ray opacity measurements used in CT
between photons and matter in which a photon is absorbed by are projections of the linear attenuation coefficients along the
an atom, which then emits an electron whose kinetic energy is ray paths. These are called X-ray projection values. A radio-
the photon energy less the electron’s binding energy. gram of an object is a two-dimensional projection.

28
 E1441 − 11
X1.1.87 radiation intensity—the quantity of radiation per X1.1.96 reference detector—a detector used to measure
second passing through an area normal to the beam path. variation in the source of penetrating radiation or some other
X1.1.88 radiation source—the apparatus from which the system parameter.
penetrating radiation used in CT scans is emitted. Discussion—Also called monitor detector.
Discussion—Examples of industrial CT radiation sources X1.1.97 reformatting—use of three-dimensional CT-
are X-ray tubes, linear accelerators, radioisotopes, and syncho- density information to produce tomograms for surfaces other
tron radiation. The size of the radiation-emitting region, or than the slice planes in which the X-ray opacities are measured.
beam spot, is an important determinant of the aperture Discussion —Most common is the production of two sets of
function, and thus of the spatial resolution. planes orthogonal to the slice planes and to each other, but
X1.1.89 radiogram—a two-dimensional projection of tomograms can also be interpolated in this way onto curved or
X-ray opacity recorded as a digitized array of computed values. discontinuous surfaces.
Discussion—Analogous to radiographs recorded on X-ray X1.1.98 region-of-interest scan—a tomographic scan in
film. Such images, when produced by CT scanners, are also which only a limited portion of the object cross-section is
called by such names as preview scans or scout scans because included in all views.
of their use in selection of the desired slice plane. See digital Discussion—The amount and pattern of the artifacts due to
radiography. the missing data depend strongly on the object shape, but are
X1.1.90 radioisotope—an unstable isotype (type of atom), often acceptable within the boundaries of the fully-scanned
in which the nucleus will eventually change spontaneously, region.
with the emission of particles and/or gamma rays. X1.1.99 resolution—the ability to distinguish features in an
Discussion —Several radioisotopes manufactured in nuclear image.
reactors have been used as photon radiation sources for CT Discussion—See spatial resolution, density resolution, and
scanners, including isotopes of cobalt (60 Co), cesium (137 Cs), contrast-detail diagram.
iridium (142 Ir), and americium (241 Am). X1.1.100 scan, (CT or DR)—set of measurements from
X1.1.91 Radon transform—a transform of a density func- which a tomogram or radiogram is to be computed, or the
tion in two or more dimensions into its projections along process of acquiring such measurements.
straight lines. X1.1.101 scan geometry—the pattern and sequence of
Discussion—The Radon transform approximates the process opacity measurement for a tomographic scan.
of X-ray opacity measurement, and the basic image- Discussion—Scan geometry is often specified by reference
reconstruction problem of computed tomography is finding an to the sequence of “generations” (first, second, third, fourth) of
inverse to the projection transform. This problem was first medical CT scanners. Second generation (translate-rotate) and
solved by Radon in 1917. third generation (rotate-rotate) geometries are most used in
X1.1.92 ray—the path taken by the penetrating radiation industrial CT. Another important scanning approach is cone
used in a particular X-ray opacity measurement. beam geometry, in which an area detector is used.
Discussion—The aperture function gives the extent and X1.1.102 scatter—the redirection of radiation-beam pho-
intensity distribution of rays. tons due to interactions with matter in their path.
X1.1.93 ray spacing—the distance (linear or angular) be- Discussion—Photons of the penetrating radiation used in
tween adjacent rays in a tomographic view. tomography can interact with the electrons (or, at low energies,
Discussion—This distance is usually the most important the nuclei) in a material in a manner which changes the
determinant of spatial resolution. See measurement spacing. direction of the photon. This scattering process has several
X1.1.94 rebinning—production of a tomographic data set implications for tomographic measurements: (1) the photon
of a desired pattern (usually uniformly-spaced parallel-ray will generally not interact in the detector toward which it was
views) by redistribution or interpolation of the data set actually originally travelling (causing attenuation of the primary beam),
measured. (2) the photon will usually transfer some of its energy to an
electron in the material (causing dose in the object or signal in
X1.1.95 reconstruction—the process of computing a de-
the detector), and (3) such a scattered photon may interact
scription of an object from X-ray opacity measurements
some detector not on its original path (introducing error into
through it.
the associated opacity measurement). The contribution of
Discussion—Normal CT practice is to use CT-density values
scattered photons to measurements may be greatly reduced in
in a rectangular array of pixels or voxels to describe the object;
some cases by collimation of the detector array or separation of
such an array is called a tomogram. The standard method for
the detector from the object; it is sometimes compensated for
computing tomograms groups opacity measurements into
by subtracting an estimate of its value from the measurements.
views which are convolved using Fourier transforms to apply
See Compton scattering and elastic scattering.
a mathematical filter proportional to frequency. To each pixel
in the tomogram is added a backprojected value from each X1.1.103 scintillation detector—a device which converts
convolved view, chosen based on which ray went through that incident radiation into visible light that is subsequently mea-
pixel from that view. Other reconstruction techniques, notably sured after conversion to an electrical signal.
successive-approximation methods, have also been developed X1.1.104 SDD (source-to-detector distance)—distance
but are not now in general use. from radiation source to detector element.

29
 E1441 − 11
X1.1.105 second-generation scan—a sequence of tomo- tomograms, linear smoothing (or sharpening, the inverse pro-
graphic data acquisition in which the object being examined is cess) can be applied without additional computation by suitable
translated across a fan beam several times, with the object modification of the convolution filter used in image formation.
rotated by the detector-array fan angle after each pass. X1.1.112 SOD (source-to-object distance)—distance from
Discussion—The primary advantages of such translate- radiation-source beam spot to the center of rotation of the
rotate scans are the natural production of parallel-beam views object.
(which are easier to process), the use of a single detector for all
X1.1.113 source spot—the small region from which pen-
measurements in a view (which eliminates some types of
etrating radiation is emitted from radiation sources used in
artifacts), and more flexibility in the size of object which can
radiographic imaging.
be scanned. The primary disadvantage compared to third-
Discussion—The size of a source spot as seen from the
generation scans is the extra motion time required.
region of the object is an important limit on the sharpness of an
X1.1.106 sharpening—a transformation of a view or image image formed with its radiation, with smaller spots giving
in which the differences between points located near to each sharper images. However, smaller spots also give less intense
other are increased relative to differences between more (and thus noisier) radiation beams. For X-ray machines, which
separated points. produce source spots of from 2 mm (0.079 in.) down to 0.005
Discussion—An important class of such transforms are mm (0.0002 in.), the limit on maximum intensity is due to the
filters which increase the high-frequency content of images. melting point of the tungsten anode. For a radioisotope, the
Sharpening typically increases noise but may also increase intensity limit is due to a combination of its half-life,
spatial resolution, up to the limit imposed by measurement concentration, density, gamma-ray multiplicity, self-
spacing. The convolution step in tomographic image formation attenuation, and energy spectrum. The brightest radioisotopes
includes a particular type of sharpening. See also smoothing, used in industrial imaging, 192Ir and 60Co, are much less
which is the opposite process. intense than X-ray machines per unit of source-spot area.
X1.1.107 sinogram—the set of X-ray opacity values ob- X1.1.114 spatial resolution, CT—the extent to which a
tained during a tomographic scan, typically ordered as a tomogram or radiogram can be used to detect details of the
two-dimensional array (position within view vs. view angle). shape of image features whose contrast is substantially greater
X1.1.108 slice, CT—a tomogram or the object cross-section than the image noise.
corresponding to it. Discussion—CT spatial resolution is best characterized by
Discussion—The slice plane is the plane, determined by the the point-spread or line-spread functions of the image, or by
focal spot and the linear array of detectors, around which each the equivalent modulation transfer function (MTF) in fre-
measurement of a planar tomographic scan is centered. Each quency space. The spatial resolution is generally limited by the
such scan also has a slice thickness, which is the distance measurement spacing (which in turn is often influenced by the
normal to the slice plane over which changes in object opacity aperture function), not by the spacing of the pixel grid. For
will significantly influence the measurements; typically an assessing the information in images in which noise is
average value based on the aperture function is used to significant, a contrast-detail diagram should be used.
characterize this parameter. When three-dimensional CT- X1.1.115 third-generation scan—a sequence of tomo-
density maps have been reconstructed, a slice may be formed graphic data acquisition in which the object being examined is
on an arbitrary plane or other surface, not just on slice planes. rotated relative to the radiation source and detectors.
X1.1.109 slice plane—the plane, determined by the focal Discussion —This “pure rotation” methodology has the
spot and the linear array of detectors, around which each advantages of speed and efficient use of the detector array, but
measurement of a planar tomographic scan is centered. requires careful calibration to avoid circular artifacts. The most
Discussion —Not applicable to cone-beam scan geometries. common mode of scanning in medical CT; also common in
Using three-dimensional techniques such as reformatting, im- industrial CT.
ages may be formed on planes other than the original slice X1.1.116 tomogram—map of CT-density estimates of an
plane. object at positions on a two-dimensional surface, typically a
X1.1.110 slice thickness—the average distance normal to a square grid on a cross-sectional plane.
fan of radiation over which changes in object opacity will X1.1.117 translation—motion of the object being exam-
significantly influence the signal generated in detectors at ined relative to the tomographic measurement apparatus, gen-
which the fan is directed. erally in a straight line.
Discussion—Typically an average value based on the aper- Discussion—Contrasted to rotation of the object or appara-
ture function in the region of the object is used to characterize tus. Linear translation can be used to produce a parallel-ray
this parameter. See beam width. view with each detector (see second-generation scan). On fan
X1.1.111 smoothing—a transform of a tomographic view or beam systems, translation normal to the slice plane is used to
image in which the difference between nearby points is produce digital radiograms.
reduced. X1.1.118 view, CT—a set of X-ray opacity projection
Discussion—Smoothing typically reduces noise but also values (derived from measurements or by simulation) grouped
decreases spatial resolution. Smoothing may be linear (such as together for processing purposes, especially for the convolu-
local averaging) or nonlinear (such as median filtering). For tion and backprojection steps of computing a tomogram.

30
 E1441 − 11
Discussion—The rays forming a view will usually be dynamic range exceeds that of human visual inspection. The
parallel (parallel-beam views) or have a common vertex at the mapping of image-density values to colors or gray-scale
radiation source (fan-beam views). The pattern of rays in views intensities is usually linear, but can be changed to other modes,
need not be the same as that of the measurements, since they such as logarithmic compression or histogram equalization.
can be sorted or interpolated.
X1.1.121 X-ray opacity—the extent to which an object
X1.1.119 voxel—volume element; one of a group of posi- attenuates X-ray radiation passing through it.
tions or small volumes at which some density parameter of an Discussion—Because such attenuation is an exponential
object is estimated in a three-dimensional tomographic recon-
function of the amount of material penetrated, opacity is
struction.
measured in attenuation lengths, which are the normal loga-
Discussion—See pixel.
rithm of the ratio of the amount of radiation entering an object
X1.1.120 window (density)—the range of pixel values in a to the amount which passes through it without interaction. The
tomogram or other image which are presented to the viewer as opacity along many straight-line ray paths through an object is
varying intensities on a graphics display monitor. the information from which tomograms are computed. While
Discussion —Typically, values above or below the current
the opacity of a material to mono-energetic radiation is
density window are all shown as the same color, such as black
independent of detector efficiency and directly proportional to
or white. The width of the window can be a small fraction of
material thickness, beam-hardening corrections may be re-
the full range of the image density values; changing the
position and size of the density window within the full range quired to correctly infer thickness from opacity measurements
permits display of all features, even in images whose full with multienergetic X-ray beams.

REFERENCES

(1) Evans, R. D., The Atomic Nucleus, McGraw-Hill Book Company, Brain, Vol 5: “Technical Aspects of Computed Tomography,” C.V.
New York, NY, 1955. Mosby Company, 1981.
(2) Robb, R. A., et al, “High-Speed Three-Dimensional X-Ray Computed (14) Raviv, J., et al, Eds., Computer Aided Tomography and Ultrasonics
Tomography: The Dynamic Spatial Reconstructor,” Proceedings in Medicine, North-Holland Publishers, Amsterdam, The
IEEE, Vol 71, 1983, pp. 308–319. Netherlands, 1979.
(3) Glenn, W. V., et al, “Image Manipulation and Pattern Recognition,” (15) Greenleaf, J. F., “Computerized Tomography with Ultrasound,”
Radiology of the Skull and Brain, Vol 5: “Technical Aspects of Proceedings IEEE, Vol 71, 1983, pp. 330–337.
Computed Tomography,” C.V. Mosby Company, 1981, pp. (16) Lauterbur, P. C., “Medical Imaging by Nuclear Magnetic Resonance
4326–4354. Zeugmatography,” Transactions on Nuclear Science, IEEE, Vol
(4) Radon, J., “Über die Bestimmung von Funktionen durch Ihre Inter- NS-26, 1979, pp. 2808–2811.
gralwerte Längs Gewisser Mannigfaltigkeiten,” Saechsische Akad- (17) Altschuler, M. D., and Perry, R. M., “On Determining the Electron
emie der Wissenschaftten, Leipzig, Berichte über die Verhandlungen, Distributions of the Solar Corona from K-Coronameter Data,”
Vol 69, 1917, pp. 262–277. Solar Physics, Vol 23, 1972, pp. 410–428.
(5) Bracewell, R. N., “Strip Integration in Radio Astronomy,” Australian (18) Hanson, K. M., “Proton Computed Tomography,” Computer Aided
Journal of Physics., Vol 9, 1956, pp. 198–217. Tomography and Ultrasonics in Medicine, North-Holland
(6) Oldendorf, W. H., “Isolated Flying-Spot Detection of Radiodensity Publishers, Amsterdam, The Netherlands, 1979, pp. 97–106.
Discontinuities; Displaying the Internal Structural Pattern of a Com- (19) Budinger, T. F., et al, “Emission Computed Tomography,” Image
plex Object,” IRE Transactions on Bio-Medical Electronics, Vol Reconstruction from Projections: Implementation and Applications,
BME-8, 1961, pp. 68–72. Springer-Verlag, Berlin and New York, 1979.
(7) Kuhl, D. E., and Edwards, R. Q., “Image Separation Radioisotope (20) Knoll, G. F., “Single-Photon Emission Computed Tomography,”
Scanning,” Radiology, Vol 80, 1963, pp. 653–662. Proceedings IEEE, Vol 71, 1983, pp. 320–329.
(8) Cormack, A. M., “Representation of a Function by Its Line Integrals, (21) Ter-Pogossian, M. M., “Positron Emission Tomography,” Scientific
with Some Radiological Applications,” Journal of Applied Physics, American, Vol 243, 1980, pp. 171–181.
Vol 34, 1963, pp. 2722–2727. (22) Handbook of Chemistry and Physics, 45th ed., Chemical Rubber
(9) De Rosier, D. J., and Klug, A., “Reconstruction of ThreeDimensional Company, 1964–1965, p. F-47.
Structures from Electron Micrographs,” Nature, Vol 217, 1968, pp. (23) Logan, B. F., “The Uncertainty Principle in Reconstructing Func-
130–134. tions from Projections,” Duke Mathematical Journal, Vol 42, 1975,
(10) Smith, P. R., et al,“ Studies of the Structure of the T4 Bacteriophage pp. 661–706.
Tail Sheath. I. The Recovery of Three-Dimensional Information (24) Huesman, R. H., “The Effects of a Finite Number of Projection
from the Extended Sheath,” Journal of Molecular Biology, Vol 106, Angles and Finite Lateral Sampling of Projections on the Propaga-
1976, pp. 243–275. tion of Statistical Errors in Transverse Section Reconstruction,”
(11) Moore, W. E., and Garmire, G. P., “The X-Ray Structure of the Vela Physics Med. Biol., Vol 22, 1977, pp. 511–521.
Supernova Remnant,” The Astrophysics Journal, Vol 199, 1975, pp. (25) Snyder, D. L., and Cox, J. R.,“ An Overview of Reconstructive
680–690. Tomography and Limitations Imposed by a Finite Number of
(12) Brooks, R. A., and Di Chiro, G., “Principles of Computer Assisted Projections,” Reconstruction Tomography in Diagnostic Radiology
Tomography (CAT) in Radiographic and Radioisotopic Imaging,” and Nuclear Medicine, University Park Press, Baltimore, MD, 1977,
Physics in Medicine and Biology, Vol 21, 1976, pp. 689–732. p. 28.
(13) Newton, T. H., and Potts, D. G., Eds., Radiology of the Skull and (26) Gore, J. C., and Tofts, P. S., “Statistical Limitations in Computed

31
 E1441 − 11
Tomography,” Physics in Medicine and Biology, Vol 23, 1978, pp. (48) Minerbo, G., “MENT: A Maximum Entropy Algorithm for Recon-
1176–1182. structing A Source from Projection Data,” Computer Graphics and
(27) Seitz, P., et al, “The Influence of Photon Counting Statistics on Image Processing, Vol 10, 1979, pp. 48–68.
Picture Noise and Reproducibility in Quantitative Tomography,” (49) Tasto, M., and Schomerg, H., “Object Reconstruction from Projec-
Transcripts Nuclear Science, IEEE, Vol NS-32, 1985, pp. tions and Some Nonlinear Extensions,” Pattern Recognition and
1162–1168. Signal Processing, Sijthoff and Noordhoft, Amsterdam, The
(28) Joseph, P. M., “Artifacts in Computed Tomography,” Radiology of Netherlands, 1978, pp. 485–503.
the Skull and Brain, Vol 5: “Technical Aspects of Computed (50) Schomberg, H., “Nonlinear Image Reconstruction from Projection of
Tomography,” C.V. Mosby Company, 1981, pp. 3956–3992. Ultrasonic Travel Times and Electric Current Densities,” Mathemati-
(29) Dennis, M. J., “Technical Notes on the Use of the General Electric cal Aspects of Computerized Tomography (Lecture Notes in Medical
CT 9800,” General Electric Medical Systems, General Electric Informatics), Vol 8, Springer, Berlin, 1981, pp. 270–291.
Company, 1984. (51) Barton, J. P., et al, “Computerized Tomography for Industrial
(30) Haque, P., and Stanley, J. H., “Basic Principles of Computed Applications,” Technical Report IRT 4617-011, IRT Corporation,
Tomography Detectors,” Radiology of the Skull and Brain, Vol 5: San Diego, CA, Feb. 1979.
“Technical Aspects of Computed Tomography,” C.V. Mosby (52) Gullberg, G. T., The Attenuated Radon Transform: Theory and
Company, 1981, pp. 4096–4103. Application in Medicine and Biology , PhD Dissertation, Technical
(31) Peschmann, K., “An Upper Limit on the Nonlinearity of Scintillator- Report LBL-7486, Donner Lab., University of California, Berkeley,
Photodiode Detectors,” Report by Radiological Imaging Laboratory CA, March 1980.
to ARACOR on Subcontract W-0262 with Bureau of Alcohol, (53) Natterer, F., “The Identification Problem in Emission Computed
Tobacco and Firearms, 1981. Tomography,” Mathematical Aspects of Computerized Tomography
(32) Hounsefield, G. N., “Some Practical Problems in Computed Tomog- (Lecture Notes in Medical Informatics), Vol 8, 1981, pp. 45–56.
raphy Scanning,” Reconstruction Tomography in Diagnostic Radi- (54) Oppenheim, B. E., “Reconstruction Tomography from Incomplete
ology and Nuclear Medicine, 1977, pp. 217–223.
Projections,” Reconstruction Tomography in Diagnostic Radiology
(33) Hanson, K. M., “Detectability in Computed Tomographic Images,”
and Nuclear Medicine, 1977, pp. 155–183.
Medical Physics, Vol 6, 1979, pp. 441–451.
(55) Lewitt, R. M., and Bates, R. H. T., “Image Reconstruction from
(34) Brooks, R. A., and Di Chiro, G., “Beam Hardening in X-Ray
Projections, IV: Projection Completion Methods (Computational
Reconstructive Tomography,” Physics in Medicine and Biology, Vol
Examples),” Optik, Vol 50, 1978, pp. 269–278.
21, 1976, pp. 390–398.
(56) Ritman, E. L., et al, “Quantitative Imaging of the Structure and
(35) Herman, G. T., “Correction for Beam Hardening in Computed
Function of the Heart, Lungs, and Circulation,” Mayo Clinic
Tomography,” Physics in Medicine and Biology, Vol 24, 1979 , pp.
Proceedings, Vol 53, 1978, pp. 3–11.
81–106.
(36) Alvarez, R. E., and Macovski, A., “Energy-Selective Reconstruc- (57) Schlindwein, M.,“ Iterative Three-Dimensional Reconstruction from
tions in X-Ray Computerized Tomography,” Physics in Medicine Twin-Cone Beam Projections,” Transactions on Nuclear Science,
and Biology, Vol 21, 1976, pp. 733–744. IEEE, Vol NS-25, 1978, pp. 1135–1143.
(37) Bracewell, R. N., “Correction for Collimator Width (Restoration) in (58) Bracewell, R. N., and Riddle, A. C., “Inversion of Fan-Beam Scans
Reconstructive X-Ray Tomography,” Journal of Computer Assisted in Radio Astronomy,” Astrophysics Journal, Vol 150, 1967, pp.
Tomography, Vol 1 , 1977, pp. 6–15. 427–434.
(38) Joseph, P. M., and Spital, R. D., “The Exponential Edge-Gradient (59) Ramachandran, G. N., and Lakshminarayanan, A. V., “ThreeDimen-
Effect in X-Ray Computed Tomography,” Physics in Medicine and sional Reconstruction from Radiographs and Electron Micrographs:
Biology, Vol 26, 1981, pp. 473–487. Application of Convolutions Instead of Fourier Transforms,” Pro-
(39) Yester, M. W., and Barnes, G. T., “Geometrical Limitations of ceedings of the National Academy of Sciences USA, Vol 68, 1970, pp.
Computed Tomography (CT) Scanner Resolution,” ProceedingsAp- 2236–2240.
plied Optical Instrumentation in Medicine, SPIE, Vol VI 127, 1977, (60) Herman, G. T., and Rowland, S. W., “Three Methods for Recon-
pp. 296–303. structing Objects from X-rays: A Comparative Study,” Computer
(40) Stonestrom, J. P., and Macovski, A., “Scatter Considerations in Fan Graphics Image Processing, Vol 2, 1973, pp. 151–178.
Beam Computerized Tomographic Systems,” Transactions on (61) Stark, H. H., et al, “Direct Fourier Reconstruction in Computer
Nuclear Science IEEE, Vol NS-23, 1976, pp. 1453–1458. Tomography,” IEEE Transactions on Acoustics, Speech, and Signal
(41) Cormack, A. M., “Representation of a Function by Its Line Integral, Processing ASSP-29, 1981, 237–245.
with Some Radiological Applications, II,” Journal of Applied (62) Bates, R. H. T., and Peters, T. M., “Towards Improvements in
Physics, Vol 35, 1964, pp. 2908–2913. Tomography,” New Zealand Journal of Science, Vol 14, 1971, pp.
(42) Censor, Y., “Finite Series-Expansion Reconstruction Methods,” 883–896.
Proceedings IEEE, Vol 71, 1983, pp. 409–419. (63) Lewitt, R. M., “Reconstruction Algorithms: Transform Methods,”
(43) Hounsfield, G. N., “A Method and Apparatus for Examination of a Proceedings IEEE, Vol 71, 1983, pp. 390–408.
Body by Radiation such as X or Gamma Radiation,” Patent Spec. (64) Low, K. H., and Natterer, F., “An Ultra-Fast Algorithm in
1283915, The Patent Office, London, UK. Tomography,” Technical Report A81/03, Fachbereich Angewandte
(44) Gordon, R., et al, “Algebraic Reconstruction Techniques (ART) for Mathematik und Informatik, Universitat des Saarlandes, 6600
Three-Dimensional Electron Microscopy and X-Ray Photography,” Saarbrucken, West Germany, 1981.
Journal of Theoretical Biology, Vol 29, 1970, pp. 471–481. (65) Flannery, B. P., Roberge, W. G., and Deckman, H. W., “Computed
(45) Herman, G. T., et al, “ART: Mathematics and Applications,” Journal Tomography by Direct Fourier Inversion: Fast Methods for X-ray
of Theoretical Biology, Vol 42, 1973, pp. 1–32. Microtomography,” Exxon Research and Engineering Company, Vol
(46) Herman, G. T., and Lent, A., A Family of Iterative Quadratic 237, 1987, pp. 1439–1444.
Optimization Algorithms for Pairs of Inequalities, with Application (66) Bracewell, R. N., “Image Reconstruction in Radio Astronomy,”
in Diagnostic Radiology, Mathematical Programming Study 9, 1978, Image Reconstruction from Projections: Implementation and
pp. 15–78. Applications, 1979, pp. 81–103.
(47) Baba, N., and Murata, K., “Maximum Entropy Image Reconstruction (67) Hinshaw, W. S., and Lent, A. H., “An Introduction to NMR Imaging:
from Projections,” Technical Report, Department of Applied From the Bloch Equation to the Imaging Equation,” Proceedings
Physics, Hokkai-do University, Sapporo 060, Japan. IEEE, Vol 71, 1983, pp. 338–350.

32
 E1441 − 11
(68) Greivenkamp, J. E., et al, “Incoherent Optical Processor for X-Ray (76) Barrett, H. H., and Swindell, W., Radiological Imaging, Vol 2,
Transaxial Tomography,” Applied Optics, Vol 20, 1981, pp. 264–273. Academic Press, New York, NY, 1981.
(69) Sato, T., et al, “Tomographic Image Reconstruction from Limited (77) Hanson, K. M., “Detectability in the Presence of Computed Tomo-
Projections Using Coherent Optical Feedback,” Applied Optics, Vol graphic Reconstruction Noise,” SPIE Optical Instrumentation in
20, 1981, pp. 3073–3076. Medicine VI, Vol 27, 1977, pp. 304–312.
(70) Macovski, A., Medical Imaging Systems, Prentice-Hall, Inc., Engle- (78) Hanson, K. M., “Detectability in Computed Tomographic Images,”
wood Cliffs, NJ, 1983. Medical Physics, Vol 6, No. 5, 1979, pp. 441–451.
(71) Glover, G. H., and Eisner, R. L., “Theoretical Resolution of CT (79) Sekihara, K., Kohno, H., and Yamamoto, S., “Theoretical Prediction
Systems,” Journal of Computer Assisted Tomography, Vol 3, No. 1, of X-ray CT Image Quality Using Contrast Detail Diagrams,” IEEE
1979, pp. 85–91.
Transactions on Nuclear Science, Vol NS-29, 1982, pp. 2115–2121.
(72) Yester, M. W., and Barnes, G. T., “Geometrical Limitations of
(80) Cohen, G., and Di Bianca, F., “The Use of Contrast-Detail-Dose
Computed Tomography (CT) Scanner Resolution,” Proceedings
Evaluation of Image Quality in a CT Scanner,” Journal of Computer
Applications of Optical Instrumentation in Medicine, SPIE, Vol VI
Assisted Tomography, Vol 3, No. 2, 1979, pp. 189–195.
127, 1977, pp. 296–303.
(73) Ramachandran, G. N., and Lakshminarayanan, A. V., “ThreeDimen- (81) McMaster, W. H., Del Grande, N. K., Mallett, H. H., and Hubbell, J.
sional Reconstruction from Radiographs and Electron Micrographs: H., Compilation of X-ray Cross Sections, University of California
Application of Convolutions Instead of Fourier Transforms,” Pro- Radiation Laboratory, UCRL-50173-SEC 2-R-1.
ceedings of the National Academy of Science, USA, Vol 68, 1970, (82) Hubbell, J. H., Gimm, H. A., and Overbo, I., “Pair, Triplet, and Total
pp. 2236–2240. Atomic Cross-Sections (and Mass Attenuation Coefficients) for 1
(74) Shepp, L. A., and Logan, B. F., “Reconstructing Interior Head Tissue MeV-100 GeV Photons in Elements A = 1 to 100,” Journal of
From X-Ray Transmissions,” IEEE Transactions on Nuclear Chemical Physics Reference Data, Vol 9, No. 4, 1980, pp.
Science, Vol NS-21, 1974, pp. 228–236. 1023–1148.
(75) Judy, P. F., “Line-Spread Function and Modulation-Transfer Func- (83) Dennis, M. J., “Industrial Computed Tomography,” ASM Handbook
tion of a Computed Tomographic Scanner,” Medical Physics, Vol 3, Volume 17: Nondestructive Evaluation and Quality Control, ASM
No. 4, 1976, pp. 233–236. International, 1989, pp. 358–386.

SUMMARY OF CHANGES

Committee E07 has identified the location of selected changes to this standard since the last issue
(E1441 - 00(2005)) that may impact the use of this standard. (July 1, 2011)

(1) Deleted metrological limits in 1.6. Technology changes (5) Added descriptions of several additional scan geometries
have rendered these values obsolete in 6.5.1.5.
(2) Deleted timing reference in 1.10since it is was too limiting; (6) Added sentence to 6.5.2 describing use of volumetric CT.
and deleted sentence on limitation of volumetric CT. (7) Deleted obsolete information regarding computer systems
(3) Changed 1.12 to state that this standard is an SI standard in6.6.
with hard conversions to inch-pound units. (8) Deleted obsolete information on data storage media in 6.8.
(4) Deleted Ionization Detectors in 6.4.1since they are obso- (9) Add reference (83).
lete technology for CT.

ASTM International takes no position respecting the validity of any patent rights asserted in connection with any item mentioned
in this standard. Users of this standard are expressly advised that determination of the validity of any such patent rights, and the risk
of infringement of such rights, are entirely their own responsibility.

This standard is subject to revision at any time by the responsible technical committee and must be reviewed every five years and
if not revised, either reapproved or withdrawn. Your comments are invited either for revision of this standard or for additional standards
and should be addressed to ASTM International Headquarters. Your comments will receive careful consideration at a meeting of the
responsible technical committee, which you may attend. If you feel that your comments have not received a fair hearing you should
make your views known to the ASTM Committee on Standards, at the address shown below.

This standard is copyrighted by ASTM International, 100 Barr Harbor Drive, PO Box C700, West Conshohocken, PA 19428-2959,
United States. Individual reprints (single or multiple copies) of this standard may be obtained by contacting ASTM at the above
address or at 610-832-9585 (phone), 610-832-9555 (fax), or [email protected] (e-mail); or through the ASTM website
(www.astm.org). Permission rights to photocopy the standard may also be secured from the Copyright Clearance Center, 222
Rosewood Drive, Danvers, MA 01923, Tel: (978) 646-2600; http://www.copyright.com/

33