INSTITUTE OF PHYSICS PUBLISHING PHYSICS IN MEDICINE AND BIOLOGY

Phys. Med. Biol. 47 (2002) R123R166 PII: S0031-9155(02)24746-6

TOPICAL REVIEW

The physics of computed radiography

J A Rowlands
Sunnybrook and Womens College Health Sciences Centre, University of Toronto,
2075 Bayview Avenue, Toronto, Ontario, Canada M4N 3M5

Received 23 July 2002

Published 20 November 2002
Online at [Link]/PMB/47/R123

Abstract
Cassette-based computed radiography (CR) systems have continued to evolve
in parallel with integrated, instant readout digital radiography (DR) systems.
The image quality of present day CR systems is approaching its theoretical
limits but is significantly inferior to DR. Further improvements in CR image
quality require improved concepts. The aim of this review is to identify the
fundamental limitations in CR performance. This will provide a background
for the development of new approaches to improve photostimulable phosphor
CR systems. It will also guide research in designing better CR systems to
possibly compete with DR systems in terms of image quality parameters such
as detective quantum efficiency and yet maintain CR convenience in being
portable and more economical.

1. Introduction

Present day computed radiography (CR) is based on the use of photostimulable phosphors,
which are also known as storage phosphors (Sonoda et al 1983). They are commercially
the most successful detectors for digital radiography. The phosphors used are most often in
the barium fluorohalide family (Barnes 1993) in powder form and deposited onto a substrate
to form an imaging plate or screen. X-ray absorption mechanisms are identical to those of
conventional phosphor screens used with film. They differ in that the useful optical signal
is not derived from the light emitted in prompt response to the incident radiation, but rather
from subsequent emission when the latent image, consisting of trapped charge, is optically
stimulated and released from metastable traps. This triggers a process called photostimulated
luminescence (PSL) resulting in the emission of shorter wavelength (blue) light in an amount
proportional to the original x-ray irradiation. In CR, an imaging plate (IP) containing the
storage phosphor is positioned in a light-tight enclosure, exposed to the x-ray image and then
read out by raster scanning with a laser to release the PSL. The blue PSL light is collected with
a light guide and detected with a photomultiplier tube (PMT). The PMT signal is digitized to
form the image on a point-by-point basis (Fujita et al 1989).
The broad acceptance of CR has been due to its large dynamic range, digital nature, easy
portability and uniqueness rather than its intrinsic image quality. CR systems have improved in
the almost 20 years that they have been available. They are now technological and engineering
0031-9155/02/230123+44$30.00 2002 IOP Publishing Ltd Printed in the UK R123
R124 Topical Review

marvels, but have been unable to transcend their inherent image quality weaknesses. CR
based on the use of storage phosphor screens in a cassette is seen to be complementary to
rather than directly competitive with integrated readout digital radiography (DR) systems.
There are numerous possible approaches to DR (Yaffe and Rowlands 1997) but to date the
most successful are flat panel systems based on active matrix arrays (Rowlands and Yorkston
2000). Flat panels are very close to being perfect in terms of quantitative measures of the
efficiency in which the x-ray aerial image is transformed to a digital image. In this regard DR
has far outstripped the present ability of CR. Thus even to equal the image quality obtainable
with screenfilm and to match DR, change is needed in the basic approach to CR. Several very
promising new approaches suggest that CR can rise to the challenge presented by flat panel
based DR systems.
1.1. Historical background
The fundamental innovation in the development of CR was by Kodak (Luckey 1975) who
conceived the storage of an x-ray image in a phosphor screen. It required significant technical
steps and conceptualization of the application by Fuji (Kotera et al 1980) to produce the first
medical x-ray images. Fuji, the main developer of CR in the eighties, used BaFBr:Eu2+
phosphor and a cassette-based approach. During this time, Agfa and Kodak performed
research and development on the same method but were constrained from commercialization
by patent issues and ambivalence due to the fear of damaging their installed base of screenfilm,
respectively. In this era the storage effect was also being observed in screenfilm applications
where it caused the unwanted effect of print through, i.e. a ghost image of a prior exposure to
the screen that appears on a subsequent film exposed in the same cassette. The storage effect
is related to the phenomenon of thermally induced luminescence of irradiated materials, i.e.
thermoluminescence. Both photoluminescence and thermoluminescence have a long history
that can be traced back to 1603 (McKeever 1985, Kato 1994, Seibert 1997) and forward
to present day applications in medicine (e.g. radiation dosimetry), biology (e.g. readout of
radioactively tagged electrophoresis gels) and elsewhere (e.g. archaeological dating).
1.2. Outline of review
There have been two previous comprehensive reviews of CR science and technology (Kato
1994, Seibert 1997). A recent text (Beutel et al 2000) though not specifically targeting CR
is highly relevant. A new review is timely because the science and technology of existing
systems have plateaued and new concepts are being actively investigated.
In the present review the fundamental operation of photostimulable phosphors is first
outlined. Then the two components of present day CR systems, the screens (or imaging
plates) and the most commonly used reader type (flying spot), are described. The combination
of a CR plate and reader forming a fully functioning CR system follows. From this the
limitations of present day CR systems are extracted and novel, new approaches are identified.
The current capabilities of clinical systems are described and potential areas of improvement
are identified.

2. Photostimulable phosphors

2.1. Operation of photostimulable phosphors

The following sections will outline the types of photostimulable phosphors used in CR. The
physical phenomena arising in conventional phosphors and the new phenomena unique to
photostimulable phosphors will be discussed.
Topical Review R125

2.1.1. Types of photostimulable phosphors. The photostimulable phosphor first used for
CR was BaFBr:Eu2+. Its crystal structure is non-cubic, i.e. a layered structure that gives
rise to phosphor grains with a plate-like rather than the more desirable cubic morphology
(Blasse and Grabmaier 1994). BaFBr:Eu2+ is a good storage phosphor in that it can store a
latent image for a long time, e.g. the latent image 8 h after irradiation will still be 75% of
its original size (Kato 1994). The family of phosphors BaFX:Eu2+ where X can be any of
the halogens Cl, Br or I (or an arbitrary mixture of them) have been studied extensively. The
decay time after photostimulation of all these phosphors is now known to be approximately
the same (0.7 s) and so they can all be used in CR. In earlier literature there was a long
decay noted for BaFCl:Eu2+ which can now be eliminated. In recent years most manufacturers
have used BaFBr0.85I0.15:Eu2+ not for the marginal increase in x-ray absorption compared to
BaFBr:Eu2+, but rather for the better optical match of the wavelength of maximum stimulation
of the phosphor to diode lasers. Recently Konica has utilized pure BaFI:Eu2+ in commercial
systems (Nakano et al 2002) where the change in absorption is significant.
RbBr:Tl+ is cubic and has the advantage that it can be made into a needle-structured layer.
By guiding the light to the surface, even a thick layer can achieve high resolution. However
it has the disadvantage of a rapid (tens of seconds) loss of the latent image that makes it
unsuitable for cassette-based systems (Nakazawa et al 1990). Konica has used this material
in integrated readers where the CR plate can be rapidly read out in situ immediately after the
termination of the exposure. CsBr:Eu2+ is also cubic, can be made in the needle structure, has
a stable latent image and can be photostimulated. Agfa has proposed using this material in
both cassette-based and integrated readers (Leblans et al 2001).
The spectrum of light emitted by an efficient phosphor is controlled by a dilute (<1
atomic%) impurity called an activator. Such activated phosphors have a characteristic line
spectrum caused by the isolated atom in the host or matrix. In BaFX phosphors used
in CR the activator is Eu2+, which substitutes for Ba in the crystal lattice. Additionally
in a photostimulable phosphor there should be effective electron and hole traps at every
activator site so that the maximum number of x-ray induced excitations can be trapped. The
detailed mechanism of PSL is still controversial (Blasse and Grabmaier 1994, Seibert 1997)
and probably differs between specific photostimulable phosphors.
The importance of having an appropriate phosphor manufactured in a reproducible
manner cannot be overstated. The interested reader is referred to Blasse and Grabmaier
(1994) as an introductory and accessible source on the physics, chemistry and manufacture
of both conventional and photostimulable phosphors. We will first review the operation
of conventional phosphors before developing the additional concepts necessary for the
understanding of photostimuable phosphors.

2.1.2. Operation of conventional (non-photostimulable) phosphors. A key concept in

phosphors is the exciton, a hydrogen-like pseudo-atom consisting of a bound electron and
hole as shown in figure 1. The exciton is a neutral entity that can form in crystalline insulators
after radiation has created ionization. The exciton can move freely within the crystal since in
the band structure representation shown in figure 1(a), the electron is in the conduction band
of the crystal, the hole is in the valence band and their mutual attraction creates a localized
minimum preventing the exciton from ionizing. In this manner the exciton drifts until it is
trapped. Trapping of the exciton at an activator site is a prelude to the recombination of the
electronhole pair (ehp) and the release of the characteristic radiation of the activator (e.g.
Eu2+). If the electron and hole were separated after creation as in a semiconductor, they
would have to diffuse separately to the same activation centre before they could recombine in
order for light to be emitted. This is a highly unlikely event. Thus conditions favourable for
R126 Topical Review

Figure 1. The storage of energy and the subsequent generation of light of energy EF in
photostimulable phosphors. (a) Band structure representation of a bound electronhole pair or
exciton. (b) The energy levels at which an electronhole can be trapped at a PSL complex.

exciton formation and subsequent stability are desirable and probably essential for an efficient
phosphor.
The efficiency of a phosphor for x-ray irradiation is conventionally defined as the ratio of
the emitted light energy to the energy of the x-rays absorbed. Typically the best phosphors
have an efficiency 1020% (Blasse and Grabmaier 1994). For our purposes, it is more
useful to define a different quantity, the number of light photons released per unit of energy
absorbed in the phosphor or more commonly its reciprocalthe average energy W, which
must be absorbed to release a single light photon. Thus stated, the correspondence between
phosphors and semiconductors for x-ray detection becomes apparent. In either case, the first
step in conversion of absorbed energy is to create ehps by the internal photoelectric effect
where an electron is excited from the valence band to the conduction band leaving behind a
hole. The minimum energy required to create an ehp is the band gap energy Eg. However,
as shown by Shockley, this is an absolute minimum. When further considerations such as
the randomness of the energy left to each electron (energy less than Eg carried by an electron
or hole is useless to create further ionization) and the requirements of conservation of both
energy and momentum are included, the average value should satisfy W > Eg. In fact
W 3Eg . (1)
This behaviour has been shown to apply to practically all semiconductors,
photoconductors andat sufficiently high fieldsinsulators (Klein 1968). It therefore
represents the fundamental limiting energy requirement for ehp production in the solid state.
However, in conventional phosphors there is one further stage before light can be emitted, the
ehps must be allowed to recombine with the emission of radiation. This process can be made
to approach 100% efficiency by incorporating appropriate activators, which form luminescent
recombination centres. Thus equation (1) also applies to activated phosphors and can be used
to establish an approximate value for the lower limit of W on any phosphor once Eg is known.

2.1.3. Mechanisms of trapping and photostimulation. In photostimulated phosphors there

are several more stages in the conversion of incident energy to light than in a conventional
phosphor. Each stage causes further inefficiency and therefore an increase in W. In table 1
the physical properties of several important conventional and photostimulable phosphors
including W and wavelength of the light emitted by the activator have been listed and
represent an update of the classic table given by Arnold (1979). A notable feature is that the
Topical Review R127

Table 1. Physical properties of conventional and photostimulable phosphors.

G Light Spectrum for

EK Eg Density W (photons/ Decay time emission stimulation
Phosphor Z (keV) (eV) (g cm3) (eV) 50 keV) (s) peak (nm) (nm)

Gd2O2S:Tb3+ 64 50.2 7.34 20 2500 3 550 na

BaFCl:Eu2+ 56 37.4 (8) 4.56 25 2000 0.7 390 500600
BaFBr:Eu2+ 56 37.4 8.3 5.1 360 140 0.7 390 500650
BaFBr0.85I0.15:Eu2+ 56/53 37.4/33.2 (8) (5.1) 360 140 0.7 390 550700
BaFI:Eu2+ 56/53 37.4/33.2 (8) (5.6) 0.6 405 550700
CsI:Tl+ 55/53 36/33.2 6.2 4.52 20 2500 0.98 550 na
CsBr:Eu2+ 55 36 7.3 4.45 250 200 0.7 440 685
RbBr:Tl+ 37/35 15.2/13.4 3.35 0.35 433 735
Conventional phosphors, i.e. those used in screenfilm combinations, are shown in normal type and the
photostimulable phosphors in bold. Entries that are unknown are left blank or if an entry is inappropriate na
is used. In some cases the value has been estimated by comparison with similar known materials, in this case the
value is enclosed in parentheses. Z is the atomic number of the radiologically important, i.e. heavy element or
elements, EK is K-absorption edge energy of the heavy elements, Eg is the energy gap of the host crystal in electron
volts (eV), density is quoted for a single crystal, i.e. 100% packing factor, G = conversion gain (emitted light photons
per 50 keV of absorbed x-ray energy). For the conventional phosphors this is the direct light emitted and for the
photostimulable phosphors is the PSL light obtained when the trapped energy is fully stimulated and shown with an
asterisk. Decay time constants are usually the same whether the light is given promptly on stimulation by x-rays or
after stimulation of trapped charge by a laser since they are limited by characteristics of the same luminescence centre
and not the trapping, and release mechanisms. Data were compiled from Arnold (1979), Blasse and Grabmaier (1994),
Holl et al (1988), Leblans et al (2001), Schweizer (2001) and Sonoda et al (1983).

conversion gain for the photostimulable phosphors is an order of magnitude smaller than that
for conventional phosphors.
In a storage phosphor, excitons can be trapped without the emission of light. It is believed
that if photostimulation is to occur later, the trapping must occur on sites spatially correlated
with the activator. This is called the PSL complex shown in figure 1(b). The energy levels in
the crystal are critical to effective storage phosphor operation. The energy difference between
the electron traps and the conduction band edge must be small enough to allow stimulation with
laser light, yet sufficiently large to prevent significant random thermal release of the charge
carriers from the traps. In BaFBr:Eu2+ the image storage is due to: (i) electron trapping at
positive ion (Br or F) vacancies, forming an F-centre, or (ii) hole trapping at an unidentified site
(Blasse and Grabmaier 1994). An activated photostimulated luminescent site or PSL centre is
therefore thought to be an arrangement of three spatially correlated components: an electron
trap, a hole trap and the luminescent activation centre. The PSL emission spectrum has been
correlated with an internal transition within the activator, Eu2+. The stimulation spectrum has
been correlated with the absorption spectrum of the F-centre showing that the first step in
the stimulation process is excitation of the trapped electron. However, it is believed that a
great inefficiency arises because 80% of the electrons are trapped at F sites and 20% at
Br sites but only the latter contribute to PSL (von Seggern et al 1988, Thoms et al 1991). It
was first thought that the hole was trapped on the activator site itself (Eu2+) thereby increasing
the valency to Eu3+ (Takahashi et al 1984, 1985). However as there is no change in the
electron spin resonance spectrum following x-ray irradiation, this trapping mechanism cannot
be operative (Schweizer 2001). Thus the nature of the hole trap is in doubt and the details of
the entire process are only known approximately.
In the needle phosphor RbBr:Tl+ the luminescent centre is the Tl+ ion; the electron
is trapped at a Br vacancy; and the hole is assumed to be trapped at a Tl+ ion (Blasse
R128 Topical Review

and Grabmaier 1994). Optical stimulation excites the electron, which recombines with the
hole on the Tl+ yielding the PSL light.
There are many other possible photostimulable phosphors but little is published relevant
to their application in CR. Significant studies have been made into large classes of materials
(Shimada et al 1991). It is probable that there are more to be discovered and improvements
to be made to already known phosphors. It is important to continue to better understand
the operation of the storage phosphors so as to be in a position to improve their imaging
properties.

2.2. X-ray properties of photostimulable phosphors

The number of x-rays used limits the image quality since x-ray image noise arises from the
random interactions of x-rays with the detector. The square of signal-to-noise ratio (SNR) at
the input of a detector SNR2in, i.e. the ultimate SNR which a perfect detector could achieve,
is equal to the number of x-rays incident on the detector NI. At the output of the detector
SNR2out = Nd (the number of x-rays detected) for a detector that counts x-rays. The ratio of
detected to incident x-rays is called the quantum efficiency AQ. Thus for a photon counting
detector
SNR2out Nd
= = AQ . (2)
SNR2in NI
From equation (2) it is evident that AQ is the single most important determinant of the ultimate
image quality possible from an x-ray detector. In figure 2, AQ is plotted against incident
x-ray energy E for some common photostimulable phosphors. In figure 2(a) the AQ for
BaFBr0.85I0.15:Eu2+ is plotted for phosphor loadings of 40 and 70 mg cm2 representing the
typical values for a high resolution and standard IPs, respectively, used in CR (Kato 2002). In
figure 2(b) the attenuation of BaFI:Eu2+ is shown for the same phosphor mass loadings and in
figure 2(c) a more direct comparison of BaFBr0.85I0.15:Eu2+ and BaFI:Eu2+ layers of the same
physical thickness of 200 m. It is seen that the K-edge is very pronounced and the total
absorption is never very large but can be significant in the energies just above the K-edge of Ba
and/or I. The very large variation of AQ with energy, shown in the relatively thin layer of the
powder phosphors, gives rise to an undesirably large absorption of scattered radiation relative
to the primary (unscattered) radiation in these screens. This is a disadvantage (Tucker et al
1993 Shaw et al 1994 Yip et al 1996) compared to for example, Gd2O2S screens with a higher
K-edge (Bogucki et al 1995, Floyd et al 1991). There is controversy in the literature as
to whether the increase in the scatter to primary ratio in CR is due to the position of the
K-edge or is correlated to the relatively poor absorption of CR screens (McLean and Gray
1996). Thus, in general, the SNR is increased in CR by reduction of scatter (Miettunen and
Korhola 1991).
In figure 2(d) AQ is shown for the needle-structured phosphors. In this case the absorption
is (with the exception of RbBr) larger than in the powder phosphors and the increase in
attenuation seen at the K-edge is less pronounced.

3. Screens and CR imaging plates (IPs)

The design and physics of IPs are very similar in concept to conventional phosphor screens
used with film. Thus a brief review of the design and operation of conventional screens
follows.
Topical Review R129

(a) (b)
1.0 BaFBr0.85I0.15 1.0 BaFI

0.8 0.8

0.6 0.6
AQ

AQ
0.4 0.4
2
70 mg/cm
2 70 mg/cm

0.2 0.2
2 2
40 mg/cm 40 mg /cm
0.0 0.0
0 20 40 60 80 100 120 140 0 20 40 60 80 100 120 140

Energy (keV) Energy (keV)

(c) 200 m thick phosphor layer (d) 400 m thick phosphor layer
1.0 1.0
with 50% packing fraction
(100 m of 100% phosphor)
0.8 0.8

0.6 0.6
AQ

AQ
0.4 0.4
RbBr CsBr CsI

0.2
BaFI 0.2
BaFBr0.85I0.15

0.0 0.0
0 20 40 60 80 100 120 140 0 20 40 60 80 100 120 140

Energy (keV) Energy (keV)

Figure 2. X-ray attenuation curves for phosphors. (a) BaFBr0.85 I0.15:Eu2+ for 40 and 70 mg cm2.
(b)BaFI:Eu2+ for 40 and 70 mg cm2. (c) Comparison of BaFBr:Eu2+ with BaFI:Eu2+ at the same
phosphor layer thickness. (d) Comparison of the needle-structured phosphor RbBr, CsBr and CsI
for the same 400 m thick layer (graphs courtesy of G DeCrescenzo).

3.1. Conventional phosphor screens

The phosphor grains are highly scattering particles due to the high refractive index of phosphors
compared to the plastic binder and air pockets within the screen. The scattering is sufficient
in that the flow of photons can be considered diffusive and the layer turbid. This scattering
limits the spreading of light from its point of origin to a lateral distance comparable to the
phosphor layer thickness. Consequently, if high resolution is desired, a thin screen must
be used also reducing AQ. While travelling within the phosphor, light will be spread by
scatteringa random walk processthe amount of lateral diffusion being proportional to the
path length required to escape the phosphor. The fraction of light escaping from the screen
depends on: the bulk absorption of the screen, generally negligible for most screens unless a
dye is intentionally added; and the boundary condition on the screen, primarily the nature of
the backing, whether it is reflective or absorptive. The boundary condition is used to adjust
the light escape efficiency as a function of depth of x-ray absorption. The blurring for x-rays
in a given screen also depends on the depth of x-ray absorption. Thus the optical boundary
conditions (and bulk absorption due to dye) can be optimized to appropriately weight the
contribution from x-rays absorbed at different depths. Analytical models of light transport in
phosphor screens have been developed to permit the calculation of screen resolution from their
R130 Topical Review

Figure 3. Manufacturing methods and structure of powder phosphor screens. (a) Doctor blade
method for depositing a thick uniform layer of phosphor slurry on the moving web of backing
material from a supply roll. (b) Overall structure of screen.

basic properties (Swank 1973). The models use parameters to represent the scattering length
(usually the phosphor grain size but this can be increased by using a binder with a similar
refractive index to the phosphor grains (Kato 1994)), the absorption length and the boundary
conditions.
The phosphor grain size is a critical design parameter for several reasons. First, there are
factors related to the intrinsic luminescence properties. If the grains are too small, then the
defects created on the outside of the grain, e.g. non-photostimulable electron or hole traps,
may dominate the bulk properties of the material. Secondly, there are factors related to the
diffusion of light through the screen. If the grains are too large, then non-uniformity of light
output or structural noise will be more pronounced. Thirdly, a larger scattering length will
reduce the resolution of the screen. Thus screens designed for a low-resolution task, e.g. chest,
can use larger phosphor grains than a high-resolution, e.g. mammographic, task. Screens are
usually designed to be 1020 phosphor grains thick.

3.2. Manufacture of IPs

CR IPs are powder phosphor screens made with photostimulable phosphor. Phosphor grains
are combined with a polymer binder and deposited on a substrate. The ratio of binder volume
to phosphor volume in the mixture controls the fractional volume of the phosphor layer finally
occupied by air pockets or voids. A solvent is used to liquefy the phosphor/binder mixture
into slurry in preparation for deposition onto the flexible but strong backing layer of the screen.
The doctor blade controls the phosphor layer thickness deposited on the backing as shown in
figure 3(a). The backing material is transported as a web below the doctor blade at a carefully
controlled spacing. The doctor blade deposits the phosphor slurry onto the backing, and its
edge establishes the thickness of the phosphor layer as shown in figure 3(b). The final screen
thickness will be less than the height of the blade due to shrinkage as the solvent is driven off
during the subsequent drying process. Typically the binder is nitrocellulose, polyester, acrylic
or polyurethane and the backing material is also a polymer, e.g. polyethylene terephthalate
200400 m thick (Kato 1994). The use of a black or white backing permits adjustments
of the reflectivity and absorptivity at the phosphor interface. In BaFBr0.85I0.15:Eu2+ IPs the
typical phosphor grain size is 4 or 5 m (Kengyelics et al 1998a, Matsuda et al 1993) with
a trend to yet smaller sizes as the ability to make smaller grains with good PSL properties
improves.
A recently introduced capability in the design of IPs is the possibility of having different
optical boundary conditions for the laser (absorptive) and PSL light (reflective) by use of
Topical Review R131

analyser

Figure 4. Pulse height spectra in phosphor screens. Top row: measurement method. Bottom row:
example pulse height spectra obtained under the conditions shown.

anti-halation backing layers (Gingold and Schaetzing 2001). The exposed phosphor surface
and the back of the screen have additional protective layers. The purpose of both these layers
is to protect the optical surface of the phosphor layer because during stacking and transport
within the reader, the bottom surface rubs against the top of other IPs.

3.3. Gain-fluctuation noise in photostimulable phosphor screens

Ideally in a phosphor screen, the same amount of light would be given off from the absorption
of every x-ray. However, in practice there are variations in the light emitted per x-ray,
which gives rise to gain-fluctuation noise. Such effects can be visualized by considering a
pulse height spectrum PHS (Ginzburg and Dick 1993). This is a histogram of the number of
events in which a specific number of light photons are given off after the absorption of a single
monoenergetic x-ray (figure 4). Such a PHS may be measured experimentally for conventional
phosphor screens (Drangova and Rowlands 1986) as shown in figure 4(a) or may be modelled
(Fahrig et al 1995). The physics of the mechanisms giving rise to changes in the PHS is usually
clear (Trauernicht and Van Metter 1988) and illustrative examples are given in figure 4(b)-(d).
Furthermore, the effect of gain-fluctuation noise on image quality can be calculated by use
of the Swank factor, which is derived directly from the PHS (Swank 1973a). The underlying
idea is that the gain-fluctuations sensed by an energy-integrating detector will decrease SNRout
compared to a counting detector. The generalization of the quantum efficiency concept is
recognized in the terminology as detective quantum efficiency, which is used to understand
the noise properties of an integrating detector. Thus the following relationship can be derived:
SNR2out
= DQE. (3)
SNR2in
This concept has been described fully by Swank (1973a) and can be further generalized
to include effects that occur as a function of spatial frequency f. If we look first at the large
area effect, that is for f = 0, Swank showed that the added gain-fluctuation noise could be
expressed as a multiplicative factor now called the Swank factor, AS:
DQE(0) = AQ AS . (4)
R132 Topical Review

Thus the correction term AS converts the DQE of a device used as a counting detector
to that for an integrating detector. The Swank factor can, to a useful approximation, be
subdivided into further multiplicative components with the common feature that they are all
ideally unity but in practice are usually less than unity by only a small factor. We will now
examine these components of AS.
When an x-ray spectrum is used, the screen reacts to x-rays with a different energy by
giving off a different amount of light. The correction term for a broad x-ray spectrum is called
AXED (x-ray energy distribution). Its value depends both on the spectrum and the absorption of
the screen and is modest with AXED 0.9. There are ongoing discussions to redefine DQE(0)
for radiographic and fluoroscopic detectors so as to exclude this effect. The current literature
still includes it.
Further variations of light output from the screen from x-ray to x-ray arise from two
effects shown in figure 4(d): (i) K-fluorescence occurs when an x-ray of energy E above
the K-edge of a material in the screen may give off a K-fluorescent x-ray of energy EK and
the K-photon may escape (resulting in a PHS output with a total energy per x-ray of EEK)
or be reabsorbed (the PHS output has a total energy per x-ray of E ). The effect on gain-
fluctuation noise arising from K-escape and reabsorption is called AAED (absorbed energy
distribution). Values for commonly used radiographic phosphors have been tabulated (Swank
1973a, Rowlands and Yorkston 2000) and calculated by Monte Carlo methods (Chan and Doi
1984, Boone et al 1999). The AAED including K-reabsorption is unity for E below the K-edges,
i.e. there is no effect on noise, but AAED drops considerably for E > K-edge. The smaller
the total absorption in the screen, the greater K-escape above the K-edge and so the smaller
is AAED. Due to the relatively low AQ in powder phosphor screens AAED 0.7 is reached at
energies just above the K-edge. It then slowly increases back to unity with increase in energy.
Averaged over the whole x-ray spectrum AAED is 0.750.85 for powder phosphor screens
and in needle phosphor screens, with a smaller K-escape fraction, AAED 0.90.95. (ii) The
effects due to the interactions of the depth of absorption of the x-ray and the optical boundary
conditions (e.g. for a screen with an absorptive backing the amount of light emitted for an
x-ray depends on whether it is absorbed near the output face of the screen {full light emitted}
or the back of the screen near the absorber {no light emitted}). This gives rise to the correction
factor AOPD (optical pulse distribution). For screens with a reflective backing the PHS is quite
narrow showing that the amount of light collected from each x-ray is independent of the depth
of x-ray absorption. Thus AOPD 1. However for a black backing or a bulk dye the light
collected is highly dependent on the depth of x-ray absorption resulting in an exponential PHS
and AOPD 0.5. For a practical system it is also necessary to consider contributions to DQE(0)
from structural noise Astruct.
This understanding of conventional screens can be applied to CR IPs. There are two
additional important effects: (i) the fraction of the trapped charge latent image released by
the simulating light, i.e. the discharge fraction F, depends nonlinearly, on the stimulating
light intensity (Lubinsky et al 1987). Therefore AOPD depends on F. For example, in IPs with
absorptive (black) backing AOPD 0.5 at low F, i.e. F < 0.1, but AOPD 1 as F 1 where all
depths of the IP are completely discharged. (ii) The number of photoelectrons emitted from
the photocathode of the PMT per x-ray absorbed in the IP is relatively small so that secondary
quantum noise (specifically called luminescence noise in the context of CR) represented by
the parameter ASQ also plays a role in the total DQE(0).
Other sources of noise, which will be ignored here because they can be generally reduced to
insignificant values, include stimulating laser noise, multiplicative gain noise in the electronic
amplifier and quantization noise. Especially note that amplifier dark noise, which can be
a dominant factor degrading DQE in DR systems (e.g. flat panel detectors, Rowlands and
Topical Review R133

Figure 5. Screen opticsthe path of laser light entering and scattering within the phosphor layer
and the same for the resulting PSL emission depending on the optical boundary conditions defined
by the backing layer. The small graphs represent the depth distribution of laser light IL and escape
probability IPSL for stimulated PSL generated at a given depth d in the IP. First for transport of laser
light into the IP for (a) reflective, (b) absorptive backing layers and for transport of the PSL out
of the IP, (c) reflective and (d) absorptive backing layers. Full lines represent the situation where
there is no bulk absorption of light in the phosphor layer and the dashed lines are for the situation
with a significant amount of bulk light absorption.

Yorkston (2000)), has no role in CR due to the essentially zero dark current of the PMT used
as the first stage amplifier.
In summary, the product of all the individual factors modifies the detective quantum
efficiency and thus
DQE(0) = AQ AXED AAED AOPD (F )AStruct ASQ . (5)

3.4. Resolution of photostimulable phosphor screens

Using a flying spot scanner, resolution is not dependent on the scattering of the PSL but rather
on the scattering of the stimulating laser light. The light transport properties are shown for
both PSL and laser light in figure 5. The boundary conditions for the penetration of the
stimulating laser light and escape of PSL will be examined separately. In figures 5(a) and (b)
the spreading of the laser light is illustrated from its point of incidence on the front surface
of the screen to the back. The spreading is less if the backing absorbs rather than reflects the
laser light. X-rays absorbed closer to the surface of the IP will have better resolution than
those absorbed at greater depths and the resolution is worse for an IP of greater thickness.
In order to understand the efficiency with which the latent image is read out as a function of
depth d in the screen, it is necessary to know the amount of stimulating light reaching each
R134 Topical Review

Figure 6. Comparison of the geometry and orientation of screens used in different radiographic
imaging systems. The backing layer defines the optical boundary condition and in most cases is
highly reflecting but can, in some cases be absorbing. In (a) orientation of mammographic screen
with x-rays incident on light emitting surface and (b) illustrating similar geometry and screen
thickness for computed radiography are shown. Note the x-ray attenuation curve showing the
relatively larger absorption of x-rays at the incidence surface. In (c) the arrangement for flat panel
imagers used in DR is shown, where the absorption is at the less favourable back surface of the
screen, far from the readout active matrix. (d) Shows that for screenfilm with dual screens one of
the screens is in the front screen orientation and the other in the back screen.

layer. The quantity IL is defined as the total intensity of the laser light at d independent lateral
spread of the laser and is shown in figure 5. If, as is often the case, the bulk absorption of
light in the IP can be neglected, then for reflective backing IL is essentially independent of
d. (There is, extraordinarily, a smaller intensity of laser light close to the incident face than
deeper within the phosphor layer. This arises because the laser light is trapped in the IP by
multiple scattering and it can more readily escape close to the surface.) In marked contrast,
for an absorptive backing IL drops linearly from the front surface to almost zero at the back.
Bulk absorption, which may be intentionally created by the incorporation of dye to absorb the
laser light preferentially. The effect of the dye absorber on IL is shown by the dotted lines on
the graphs in figure 5. The efficiency of escape of PSL, IPSL, has the same general nature as IL,
i.e. constant for a reflective backing and linearly dependent on the distance from the backing
for an absorptive backing (figures 5(c) and (d)). The PSL is more blurred from x-rays at the
back than the front of the IP. (The PSL blurring has no effect on resolution if the flying spot
readout method is used.) Note that the paths of most optical quanta will be the shortest and
hence blurring least if the read out is performed at the x-ray entrance side of the phosphor as is
normal in CR (figure 6(b)) and single screenfilm as used in mammography (figure 6(a))
but not in flat panel DR imagers (figure 6(c)) or conventional dual screen screenfilm
(figure 6(d)). This provides a much-needed advantage to CR over flat panel imagers. It
also is the explanation of the surprising, but little remarked fact that CR is far more universal
than other imaging systems. A single type of CR plate (standard resolution) is usable for all
imaging tasks except mammography, which is to be compared to three types of screenfilm in
general use for radiography (high resolution, general purpose and high speed).
Topical Review R135

Figure 7. Overall concept of CR readout systems. (a) Cassette-based requiring carrying cassette
back and forth between the x-ray and readout systems. (An important component in practical
application not shown in the diagram is a stacker needed to buffer the demand on the system.)
(b) Integrated readout systems requiring no operator intervention in the exposure readout cycle.

4. CR readersflying spot scanner

Present day CR systems are of two general types: (i) cassette-based systems as shown in
figure 7(a) where the IP is enclosed in a light-tight cassette for the x-ray exposure, and
subsequently moved by hand to the readout system; (ii) integrated readout systems, shown
R136 Topical Review

in figure 7(b) where the IPs are captive within the readout system, re-circulated and reused
without handling. Both types use a flying spot readout system, i.e. a laser spot is scanned with
a mirror over the exposed IP in a point-by-point raster pattern. The flying spot scanner is not
the only possible approach but is common in commercially available medical CR systems and
will therefore be described in detail.

4.1. Gaussian optics of laser beam

For mirror scanning, highly collimated beams produced by lasers are essential. A circular
beam with a Gaussian intensity in cross section is desirable and this can be accomplished
using heliumneon gas lasers or solid-state laser diodes (Kengyelics et al 1998b). Gas lasers
naturally have a circular Gaussian form if operated in their fundamental transverse mode
(TEM00). Astigmatic lenses are needed with solid-state lasers due to their highly elliptical
beam shape. With either kind of laser, once a circular Gaussian beam has been obtained, its
cross-sectional diameter can be modified using a laser beam expander. However, the beam
cannot be collimated to a parallel beam of the small size required at the IP as such a beam will
have a large inherent divergence caused by diffraction. A larger diameter, less divergent beam
must be used and focused to a small spot at the image plane. The diameter of the collimated
Gaussian beam Wl where it enters the focusing lens of focal length f is related to the focused
Gaussian spot size Wf by the relationship (Siegman 1986)
Wl Wf = f /. (6)
From equation (6) it can be seen that the smaller the focal spot required, the larger the
beam diameter before entering the lens. For CR, f = 50 cm for a field-of-view of 35 cm.
Applying equation (6) for the desired Wf = 100 m (diameter measured at the 1/e2 intensity
points (Siebert 1997)), then Wl 1 mm. This is a useful beam size, often encountered in
laboratory lasers, as a 1 mm beam does not diverge significantly over a distance of a few
metres.

4.2. Laser types

The advantages of solid-state laser diodes compared to continuous gas lasers are: (i) the output
intensity of a solid-state laser can be controlled electrically; Note that gas lasers, e.g. HeNe
with wavelength = 633 nm, need an additional external device such as an electro-optical or
electro-acoustical modulator; (ii) The solid-state diodes used in CR ( = 680 nm) are more
compact, energy efficient, and have a longer operational lifetime than gas lasers. They do not,
however, have as good a match to the optical stimulation requirements of BaFBr:Eu2+. This
necessitated using both a greater laser power and a redesign of the phosphor (replacing Br by
Br0.85I0.15, Matsuda et al (1993)).

4.3. Readout rate limits

In designing a readout system one must know the required readout rate. The plate throughput of
medical scanners is 30110 plates per hour (Seibert1997), which is adequate for the workload
of a typical clinic. The engineering limit on the transport and stacking of IPs is a component
of the total readout time. There is also a fundamental limit related to the characteristics of
the photostimulable phosphor, i.e. the intrinsic decay time of the luminescence centre (Eu2+
in the case of BaFX). PSL from a previously stimulated region will continue to glow. This
decays with a time constant, which is characteristic of the activator and the host lattice. For
BaFBr0.85I0.15:Eu2+, the time constant is 0.7 s (table 1). If scanning were performed too
Topical Review R137

(a)
(b)
1.0 1.0
PSL output from complete plate

0.8 0.8

PSL output from pixel

0.6 0.6

0.4 0.4

0.2 0.2

0.0 0.0
0 1 2 3 4 0 10 20 30 40 50
2
Laser Irradiance (J/m ) Laser power (mW)

Figure 8. The nonlinear, stimulation energy dependence of PSL (a) measured by uniform
irradiation of complete imaging plate (Kato 1994) and (b) measured for a single pixel in scanning
mode (Arakawa et al 2000). In both graphs the full line is an exponential fit to the discharge curve,
which can be seen not to be quite exact. It can, under certain circumstances, be represented by a
power law.

quickly, the PSL signal from one pixel would not be completely decayed before the PSL from
the next was initiated. Consequently, it would bleed into the next pixel and cause spatial
blurring. To avoid this, several time constants should elapse between the readout of one pixel
and the next, i.e. usually 5 time constants (1/e5 < 1% lag) resulting in 4 s per pixel. Thus
with 2000 2000 pixels the shortest readout time would be 16 s. Fuji AC-3 readers take
30 s to readout an array of this size (Fetterly and Hangiandreou 2000). Another practical
limitation on readout time is the laser power available. The PSL output with stimulating laser
irradiance is shown in figure 8 for BaFBr0.85I0.15:Eu2+ (a) for the complete plate irradiated
uniformly and (b) for an individual pixel. Note that since the PSL output saturates, i.e. the IP
is within 10% of complete discharge, with laser input of 2 J m2, there is no need to use
more light. The energy deposited by a 30 mW laser spread uniformly over the whole area
of a typical IP (0.33 m 0.33 m 0.1 m2) is 0.3 J m2 s1. Thus it will take 2/0.3 7 s
to discharge the IP by 90%, i.e. F = 90%. In practice, therefore, the characteristic phosphor
decay time provides the limit on readout time in flying spot scanners.

4.4. Beam scanning

The diode laser beam is sent through several subsystems before reaching the CR plate as
shown in figure 9(a). The laser beam is divided (not necessarily equally) into two with a beam
splitter such as a partially silvered mirror. The main beam passes to the scanning system;
the side beam is sent to a photodiode used to monitor, and with feedback, stabilizes the laser
output intensity. The laser focusing lens is generally of the F/ design. It has three further
functions: (i) to make the focal plane flat so that focus is uniform across the IP, (ii) to convert
the constant angular motion of the scan mirror into a constant linear speed at the image plane
so that the pixel spacing is constant and (iii) to move the entrance aperture of the lens, i.e.
locus defining region where any ray within the angle of acceptance will be imaged by the lens,
significantly out from the body of the lens so that the rapidly scanning mirror has space to
R138 Topical Review

Figure 9. Flying spot CR readout scanner. (a) Scanner components also showing the graph of
direct output of PMT response to a step in intensity on the IP, (b) processing stages, (c) response
curves of processing stages and (d) typical form of output signal as it passes through the processing
stages.

operate. Either a rotating polygonal mirror driven by a synchronous motor or an oscillating

flat mirror driven by a galvanometer can perform the scanning function. The advantage of the
rotating polygon mirror is that the transition from one facet to the next performs the flyback,
i.e. retrace which occurs at the end of one line as the beam rapidly returns to the start of the next
line. This maintains a high laser duty cycle, i.e. the fraction of the time the laser is actually
reading out the IP. In contrast, a galvanometer has to be driven in an oscillatory manner by a
sawtooth signal and after flyback takes a fixed time to return to stable operation. Its duty cycle
is therefore lower than a polygon and at high line rates, the duty cycle decreases even further.
Thus the polygon is used for faster scanning speeds but it has two disadvantages which lead
to periodic errors that appear as banding artefacts in the subscan direction. (i) The reflective
Topical Review R139

properties of its facets may each differ slightly, which requires a further correction to the laser
output (Matsuda et al 1993). (ii) Unintentional scanning of the beam perpendicular to the scan
direction results in an effect called cross-scan error, caused by slight angular shifts between
facets. This can be corrected passively using a pair of cylindrical optics (Matsuda et al 1993).
CR, as in any destructive readout, is very sensitive to cross-scan error. If the beam moves
one way, there will be reduced signal as it rereads already partially discharged regions of the
IP and in the other way, it will steal signal from the next line. Control of cross-scan error to
<1 m is required (Matsuda et al 1993). Any uncorrected periodic error leads to banding
with a period equal to the number of facets on the polygon (Kengyelics et al 1998b).
The scan mirror repetitively scans a line and retraces the laser beam thus defining the scan
direction. During retrace the laser beam is turned off and restarted just before it is expected
to reach the active area of the IP. The laser beam positioning with respect to the previously
read out pixels is accomplished with a line start detector, i.e. a photodiode near to the starting
point of the scanning. The linear motion of the plate in the subscan direction combined with
the laser scan creates a raster pattern that is read out progressively.

4.5. Collection of PSLlight guide

The light guide is located as close as feasible to the IP to efficiently collect the blue PSL
as shown in figure 10. Using the physical process of total internal reflection, the light guide
transfers the PSL to the photomultiplier (PMT). To be efficient, careful design of the light guide
is needed. Any light reaching the guide entrance must be accepted at an angle appropriate
for total internal reflection within the guide. Made of a transparent plastic such as acrylic,
the conditions for total internal reflection only occur if the guide is bent gently, i.e. on a large
radius. It cannot perform demagnification, as light would escape from the sides. The light
guide shape changes from a lineto match the laser scan line at the IP, to an annulusto
match the circular shape of the input window of the PMT (Tanaka et al 1984). A more compact
system using laser cuts to the acrylic has also been shown (Matsuda et al 1993). Care has to be
exercised to prevent laser light scattering or reflecting from the light guide onto the IP where
it would discharge another part of the image creating flare (figure 10(c)), i.e. a low frequency
drop in the modulation transfer function. A related effect can occur in the top layer of the IP
where unwanted reflections can produce halation (figure 10(d)). Thus the top layer thickness
has been reduced to 3 m (Ogawa et al 1995). A conductive layer is also incorporated in
the IP structure to prevent build up of static electricity.

4.6. Signal processing

The processing of the stimulated PSL signal from the IP is shown in figure 9. In figure 9(b)
are the processing modules, in (c) their function and in (d) examples of the waveform
at corresponding points in the processing cycle. The first processing step is logarithmic
amplification, which reduces the dynamic range prior to digitization and prepares the data
for better visualization on a monitor. The second processing step is to temporally filter the
signal. This has several important functions: to correlate the signal prior to digitization so as
to optimally match the analogue signal to the digitization sample rate fd. This also prevents
aliasing of noise and reduces or eliminates fixed pattern noise of an x-ray grid oriented with
the grid lines perpendicular to the scan direction. An ideal filter would be a low pass filter
with a sharp cutoff at fc = 12 fd to satisfy the sampling theorem.
The third processing step is digitization (Kato 1994). The digitization rate is relatively
slow, e.g. 4 s per pixel, therefore an a/d (analogue to digital converter) operating at a
R140 Topical Review

Figure 10. Details of light collection system (a) in normal operation, (b) details of Fuji ST-V
structure, (c) generation of flare by reflection of laser light from the face of the light guide back
onto the IP where PSL can be stimulated at a point remote from the original incident point of laser
and (d) generation of halation by reflection of light within the 3 m protective layer. This layer
has been reduced to the point where halation is insignificant.

frequency fd = 250 000 samples per second is required. How can we establish an adequate
digitization dynamic range? The upper limit is the highest exposure to which the IP can be
exposed or the raw x-ray beam (air kerma 1000 Gy). The lower limit is related to the
noise at the lowest exposure level encountered. For a/d quantization noise to be insignificant
(Dolazza and Poulo 1984) the least significant bit has to be equal to or less than the noise at the
lowest exposure level. The lowest exposure required may be assumed to be 1/10 of the mean.
The mean exposure level to which IPs are exposed is 3 Gy, which, for a 200 m square
pixel (assuming a quantum efficiency of 0.5), is 1200 x-rays per pixel (2 1016 x-rays
Gy1 m2 0.5 3 106 Gy (2 104 m)2). One tenth of this is 120 x-rays with a noise
of (120)1/2 11 x-rays (0.03 Gy). This is a raw dynamic range of 1000/0.03 = 33 000
15 bits.
In the earliest commercial CR readers an 8-bit a/d was used. Two methods were
required for the large raw dynamic range to be adequately digitized. The first used a pre-
read cycle to scan the laser across the image at a greatly reduced power. This allowed a
determination of the actual image content and the PMT gain was adjusted to match that
content (Seibert 1997, Kengyelics et al 1998b). The second method added a logarithmic
Topical Review R141

>100,000

10000

2000 Flat panel 1800

1300
1000
1000
130 Line scan CR
Number of quanta

100 65
133 33
67
Flying spot CR
10 22

Photostimulable centres
5.5
Electron-hole pairs

Electrons released
Collected photons
Emitted photons

Output electrons
Absorbed x-ray

0.1

0.01
1 2 3 4 5 6 7
Stage

Figure 11. Quantum accounting diagram for several different radiographic imaging systems
namely the flying spot CR system, line scanner CR system and flat panel DR system.

converter to process the signal from the PMT before it entered the a/d as shown in figure 9.
This further reduced the effective dynamic range of the signal (Dolazza and Poulo 1984)
and the number of bits required. Currently, analogue logarithmic processing or square root
processing (Dolazza and Poulo 1984) is used to reduce the dynamic range of the signal before
digitization by a 12-bit a/d. Alternatively if logarithmic processing is performed digitally, i.e.
after the a/d, using look-up tables, then a 16-bit a/d will be necessary. The prescan approach
has been eliminated due to the ready availability of high bit depth converters and the ability of
handling large amounts of digital data, which was far from trivial in the early eighties when
CR systems were introduced. Finally a shading correction is applied to allow for the varying
light collection efficiency of the light guide as a function of the laser position along the line.
This is a one-dimensional correction as every line is the same.

4.7. Isolation of PSL from laser light

To permit the PSL signal to be isolated from the laser light, the phosphors chosen for CR can be
stimulated with laser light with a different wavelength from the PSL light. Is there a problem
in isolating these two sources of light, both of which will enter the light guide and reach the
PMT? Laser power used in CR is 30 mW (Matsuda et al 1993, Seibert 1997, Leblans et al
2001) or 2 1017 red light photons/s (assuming 2 eV per red light photon). At this power
and the 4 s dwell time per pixel derived earlier, there are 8 1011 red light photons incident
on each pixel. The stimulated light output from the IP for a single fully absorbed 50 keV x-ray
is 70 blue PSL photons (figure 11). Thus the ability to detect 120 x-rays, i.e. lowest exposure
level defined above at which signal needs to be measured1/10 mean exposure, will require
the detection of 8000 blue photons in the presence of 8 1011 red photons. Therefore the
R142 Topical Review

Figure 12. Propagation of noise in CR systems illustrating the contribution of noise from secondary
quantum statistics. The effect on image quality by changing g is investigated. In the upper row a
representation of the image of a single x-ray photon is illustrated. With g = 1000 each x-ray has
enough secondary quanta that its image is smooth and representative of the characteristic blurring
of the systems. As g decreases, each x-ray has fewer light photons representing it resulting in
a rougher appearance. The corresponding Wiener noise power spectra for the same g values are
shown in the lower part of the figure.

ratio of stimulating red to PSL blue photons at the surface of the IP at the worst case of the
minimum x-ray exposure is 108 or eight orders of magnitude! The first approach used to
separate these light photons depends on the PMT. A typical transparent bialkali photocathode
has a quantum efficiency of 25% in the blue and 0.1% in the red. The second approach is
the filter, which still needs to selectively remove five orders of magnitude brighter light in the
red while efficiently passing light in the blue if interference with the output of the PMT from
the laser light is to be avoided.

4.8. Quantum accounting diagram

The maximum signal to noise ratio (SNR) of any imaging system occurs where the x-rays are
absorbed. If the SNR of the imaging system is essentially determined here, the system is said
to be x-ray quantum noise limited in its performance. Inevitably the SNR is reduced as the
signal passes through the system. It is possible to make high image quality x-ray detectors
because of the large intrinsic conversion gain as the x-ray energy, e.g. 50 keV, is converted to
many secondary particles, e.g. 3 eV photons. It is important that the detector maintains a large
number of quanta representing each x-ray if secondary quantum noise is to be minimized. A
quantum accounting diagram aids in locating the secondary quantum sink. Figure 11 illustrates
the propagation of quanta through the CR conversion stages. Each 50 keV x-ray interacting in
the detector produces 2000 ehp. However, only a small fraction (1/15) of these ehps are
trapped in a manner that permits them to be laser photostimulable and produce PSL. Of these,
only half are actually stimulated, i.e. discharge fraction F = 0.5, and PSL photons emitted.
For an IP with a reflective backing we assume all of these PSL photons escape from the IP.
However only 1/3 can be collected by the light guide and brought to the face of the PMT
where 1/4 release photoelectrons. Taking into account all these factors an absorbed x-ray
is represented on average by 5.5 e, i.e. the system gain g = 5.5 electrons per 50 keV x-ray
(2000 1/15 12 1/3 1/4). Fortunately, the PMT can efficiently and adequately
amplify 5.5 electrons such that no significant further noise is added. The noise inherent in
the few electrons representing a single x-ray is the weakest noise link in the CR system and
is the secondary quantum sink. For comparison, in the quantum accounting diagram for flat
panel DR (figure 11) g equals 1000. In figure 12 the effect on image quality by changing
Topical Review R143

g is investigated. With g = 1000 each x-ray has enough secondary quanta that its image
is smooth and representative of the characteristic blurring of the systems. As g decreases,
each x-ray has fewer photons resulting in a rougher appearance. The resulting noise power
spectrum is shown in the lower part of figure 12. As g decreases the noise of the secondary
quanta NPSSQ approaches, and in the case where g = 1, equals the noise from the x-rays
NPSX. This has a deleterious effect on both DQE(0) and as f increases, DQE( f ). Secondary
quantum noise is evident in optically coupled x-ray image devices (Yaffe and Rowlands 1997),
especially where optical demagnification is significant. It has been investigated in the context
of phosphor screens coupled to CCDs using tapered fibre optics (Maidment and Yaffe 1994).
The situation is also analogous to the use of film in screenfilm (Nishikawa and Yaffe 1990).
Film has an optical DQE(0) 1% (Dainty and Shaw 1974) leading to g 20, which is
comparable to g for CR. What options are available to increase g in CR?
4.9. Possible sources of improvement in conversion gain in flying spot CR systems
In figure 11, the sources of loss in g are highlighted. BaFX:Eu2+ has been developed to produce
the maximum possible PSL output and practically may have reached a limit. However, it does
not appear to violate any conservation laws to expect that a significant fraction of losses seen
within the phosphor (failure to trap ehps, non-photostimulable sites) could be eliminated.
Using previously established data leads to a theoretically possible increase in g by 15.
By careful analysis it may be possible to elucidate the exact mechanism of operation of
the phosphor, reduce the losses and improve the gain.
The next stage is the light collection system. The theoretical limit would collect all the
emitted light resulting in an increase in g 3. Means for accomplishing this could include
reading from the side of the IP opposite to that of the laser and increasing the light guide
efficiency by placing it in contact with the IP (Arakawa et al 1999, 2000).
The PMT has an optical quantum efficiency of only 25% since it relies on the external
photoelectric effect, i.e. an electron is given enough energy to eject it from a solid and into the
vacuum. In theory, solid-state devices can approach an optical quantum efficiency of 100%
because of the internal photoelectric effect where an electron is transferred from the valence
band to the empty conduction band within the solid. This potential four times increase in
optical quantum efficiency would help to reduce the secondary quantum noise characteristic
of CR. However, if a large area photodiode replaced the PMT, then both the electronic noise
from the capacitance of the diode (14 cm2 50 pF cm2) and the fluctuations in the dark
current (>14 5 nA cm2) would each yield an electronic noise Ne > 5000 e rms. For this
to be equal to the x-ray noise from NX x-rays, we have the relationship Ne = g (NX)0.5. Our
example is then 5000 = 5.5 (NX)0.5, i.e. NX 106 x-rays/pixel, which corresponds to 4 mGy
incident air kerma. Thus a flying spot CR system using a silicon photodiode would not be
quantum noise limited below 4 mGy (three orders of magnitude larger than the typical incident
air kerma of 3 Gy used in CR) incident on the IP! Thus, PMTs continue to be used because:
(i) the gain achieved by impact ionization using a cascaded series of 10 dynodes, each with
an adjustable gain up to 4, yields an overall gain of a million. This gain is essentially
noiseless. Unlike an electronic amplifier, there are no parasitic capacitances and the inherent
gain fluctuations are insignificant. (ii) The dark current Id is negligible even for a very large
active area (14 cm2) required in mapping the input of the light guide (35 cm 4 mm).
Id < 1000 e cm2 s1 or 0.06 electrons from the 14 cm2 area of the photocathode during the
pixel integration time of 4 s. Thus the electronic noise and dark current fluctuations, from a
practical-sized PMT, are insignificant compared to the 60 e (11 x-rays g = 5.5) noise that
corresponds to the lowest signal encountered in CR. Thus for a flying spot scanner, g cannot
be expected to be increased significantly in the near future.
R144 Topical Review
5
10 1000

(a) (b)

4
10
PMT output (AU)

CR Reader Units
3
10

500

2
S=200
10
L=4

ST-V
1
10 HR-V
S=200
L=2
0
10 0
0.1 1 10 100 1000 10000 0.1 1 10 100 1000
air Kerma (Gy) air Kerma (Gy)

Figure 13. Linear response of IP to x-ray exposure. (a) CR reader output measured at PMT in
arbitrary units (AU) plotted against irradiation to IP expressed in air kerma. This curve shows
the linear inputoutput relationship over more than four orders of magnitude (Bootstrapped from
three individual plots from Kato 1994). (b) CR reader output in CR Reader unit plotted against
the logarithm of x-ray kerma, i.e. exposure, showing a straight line fit which demonstrates a linear
relationship between the PSL output and air kerma (Kengyelics et al 1998b) for four combinations
of reader and IP parameters (latitude parameter L and IP type ST or HR).

4.10. Readout linearity

Figure 13(a) shows the native characteristic curve, i.e. a plot of PMT signal output before
logarithmic compression, of a CR reader. The author constructed this composite graph by
bootstrapping data from several individual curves (Kato 1994) to show the overall dynamic
range of the system. The exposure sensitivity of the CR plate is shown to be linearly
proportional to exposure from 0.11000 Gy (a dynamic range of 104). This curve must be
linear (or linearizeable) for MTF, NPS and thus DQE to be defined. However, the linearity
of the characteristic curve, although necessary, is not a sufficient condition. The creation of
PSL depends nonlinearly on the stimulation light intensity as can be seen by referring back to
figure 8. If the system is to satisfy the requirements for linear analysis, three further conditions
must be satisfied.
The first condition for effective linearity is that the laser light intensity must be kept
constant during the readout scanning process to maintain the discharge fraction F of the IP
at a constant level. This is because both the spatial resolution of CR systems (figure 14)
and the x-ray to charge conversion gain g, and hence the secondary quantum noise, depends
on F. The second condition (satisfied automatically for photostimulable phosphors in current
use) is that the discharge process must be very inefficient. This ensures that F is independent
of the latent image. If the discharge process were more efficient, then an effect similar to
adjacency in film could occur leading to a complete breakdown of linear analysis. This
is an edge enhancement that arises close to a high-contrast sharp edge during development
(Dainty and Shaw 1974). In both film development and IP readout such effects arise from the
latent image using up the developer or laser light, respectively. The third condition is that any
nonlinear processing applied to the signal from the PMT, e.g. the logarithmic compression, be
reversed.
Topical Review R145

Figure 14. Phosphor blunting is illustrated by (a) a plot of MTF plotted for three different laser
power intensities to the IP and corresponding relative discharge of plate obtained from figure 8
(Arakawa et al 2000) and (b) schematic illustration of origin of phosphor blunting.

Thus although the readout process is intrinsically nonlinear, the linear parameters defining
spatial resolution (modulation transfer function (MTF)), spatial noise (Wiener noise power
spectrum (NPS)) and the spatial frequency dependent detective quantum efficiency DQE( f )
can nevertheless be defined for CR systems in current use (with a possible exception to be
discussed later). Note that the values of MTF, NPS and DQE will be different for the same
reader and IP if factors affecting F are changed. Furthermore, the optimization procedure by
which the scan parameters are obtained is a nonlinear process.

4.11. Plate reconditioning

A residual latent charge image remains on the CR plate after readout. Erasure of the image
using a high intensity light source must be performed before the plate is reused. This is
accomplished using a high-pressure sodium or fluorescent lamp (Siebert 1997). The erasure
time depends on the brightness of the lamp and the level of erasure required. The level of
erasure achieved also depends on the prior x-ray exposure and F. The erasure rate is enhanced
if the initial erasure is performed with a light spectrum including ultraviolet followed by a
spectrum with the UV filtered out (Matsuda et al 1993). This is probably due to the creation of
trapped charge by UV irradiation. Erasure times are 1020 s for the Fuji AC-3 and 50 s for the
Lumisys ACR-2000 readers (Fetterly and Hangiandreou 2000). In practice, unless exposed to
an excessive x-ray irradiation (such as during a quality control procedure), the previous latent
image is effectively removed during a single erase cycle. It is possible to accumulate an image
on the CR plate due to natural radioactivity and cosmic radiation. Therefore, before clinical
use the IPs should pass through a further erasure cycle if they have been in storage for more
than a day (Seibert 1997).

5. Configuration and operation of CR systems

After the image has been acquired digitally, it has to be processed and manipulated before it can
be displayed effectively and a diagnosis made. Processing for CR images is specific for each
different application (Vuylsteke et al 1997). The system configuration, how the components
parameters are chosen and the CR system performance will now be described.
R146 Topical Review

Table 2. Imaging plate parameters and derivation of AS.

Phosphor layer Phosphor System DQE(0)

mass loading Phosphor layer packing gain E<1
IP type (mg cm2) thickness (m) factor (e/x-ray) AQ (Gy) AS

ST-Va 70d 230c 0.60 16.6b 0.43 0.30 0.70

HR-Va 40d 140e 0.56 4.0b 0.31 0.14 0.45
Two categories of IP are shown, ST (standard resolution) and HR (high resolution) all read out in an AC3 reader
under the x-ray conditions of 75 kVp (W) with 1.5 mm added Cu filtration. Data in a given row were obtained from
the reference given in the first column unless otherwise noted. AS is obtained by dividing DQE(0) by AQ.
a Kengyelics et al (1998).
b Estimated from NPS in the original paper by Hillen et al (1987) method.
c Ogawa et al (1995).
d Kato (2002).
e Measured by author.

5.1. System parameters

The most significant system choice is not in the readout system but in what it reads outthe
IP or screen. The amount of blurring and the effectiveness of the IP in absorbing radiation
are defined by the thickness of the phosphor layer and its optical parameters (scattering length
and absorption length for light photons and the boundary conditions). Fuji IPs are available in
standard resolution (ST) and high resolution (HR) followed by a generation number expressed
as a Roman numeral. The standard resolution plate is intended for all general radiographic
applications while the HR plate is intended specifically for mammography. A reflective layer
is used in Fuji ST-V plates (Ogawa et al 1995), which reflects only blue (PSL) lightit is
absorptive for the red laser light (Kato 2002) as shown in figure 10(b). The phosphor layer
including binder is 230 m thick. In contrast HR-V has a black backing, i.e. absorbs both red
and blue light, and has bulk absorptive dye specifically for the red laser light (Kato 2002). The
measured x-ray absorption of Fuji IPs under well-defined x-ray spectral conditions have been
given for ST-V and HR-V (Kengyelics et al 1998b). This could be used for an independent
determination of phosphor loadings from knowledge of attenuation coefficients and modelling
of x-ray spectra (Tucker et al 1990). A summary of the phosphor mass loadings and other
parameters for the IPs mentioned above is given in table 2. When the phosphor layer thickness
is known, an estimate of the phosphor packing factor can be made. The literature suggests
that, in general, maximum packing factors of 60% can be achieved (Vosburgh et al 1977) but
precise values depend significantly on the morphology of the phosphor, the distribution of
grain sizes and other factors and requirements such as flexibility of the IP (Kengyelics et al
1998a).
Measurements of the physical thickness of the phosphor layer are also important for the
understanding of the MTF. The laser spot size, from the point of view of maximum resolution,
should be as small as possible (Kato 1994). An 100 m 1/e2 diameter at the focal plane is
in common use.

5.2. Operational parameters

The pixel separation is solely a function of the reader. Most cassette-based readers are capable
of scanning at various different pixel sizes with 200 m pixels (Nyquist frequency fNY =
2.5 lp/mm) or 100 m pixels ( fNY = 5 lp/mm) for general purposes with ST. In earlier
mammographic readers using HR IPs 100 m pixels ( fNY = 5 lp/mm) were used but in more
Topical Review R147

recent readers 50 m pixels ( fNY = 10 lp/mm) have been used. In earlier readers the scanning
pitch was changed depending on the IP type. In recent readers the readout may occur at a finer
pitch than required and digital rebinning of the data used. In practice, the pixel size is chosen
so as not to be the limiting factor on the MTF.
The discharge fraction F depends on the laser energy incident per unit area of the IP. As
a system parameter, F depends on three factors: the plate size, the time to scan the complete
plate and the laser power (corrected by the duty cycle of the scanner). F may also depend
on other factors such as: contrast and density settings of the reader, and whether an ST or
HR plate is being read. It could also depend on the IP generation and other factors controlled
by the CR reader based on information from the bar code specific to the individual IP. This
is an area that merits further investigation. F can, in principle, be established by rereading
[IR2 = F(1 F)k] a uniformly irradiated IP that has been read out once [IR1 = Fk] with the
erasure cycle bypassed. Then, F = 1 IR2/IR1. In practice, F 50% (Leblans et al 2001)
represents a trade off between decreased secondary quantum noise and increased blurring as
F is increased. However, systematic measurements of F and correlation with image quality
for different designs of IPs and readers are lacking.

5.3. System performance

The CR system, though not strictly linear, satisfies most conditions for linear analysis. Thus
meaningful values for MTF, NPS and DQE can be measured in the conventional manner with
the proviso that the measurements cannot be identified with the IP alone but are a function of
the type and mode of reader and F must be constant.

5.3.1. System response. The characteristic curve of a CR reader at the output of the PMT is
shown in figure 13(a) plotted on a loglog graph where it shows a slope of unity demonstrating
linear response over many orders of magnitude. However, the logarithmic response of the
amplifier modifies the characteristic of the complete CR system. The resulting curve is shown
in figure 13(b). The logarithmic response has to be removed for linear analysis. The fact that
a straight line can be fitted through the data shows that the logarithmic amplifier is accurate
and that any signal offsets have been removed precisely before logarithmic processing.

5.3.2. Modulation transfer function. The MTF is the single most critical determinant of the
spatial frequency dependence of the DQE. Conventionally, measurements of detector MTF
correct for the size of the x-ray source (focal spot unsharpness) and the finite slit size. It
should be measured at a point in the CR system where the MTF has had no artificial edge
enhancement or other image processing. This is the native MTF.
MTF measurement on digital systems is usually based on the method of over-sampling
using a slanted edge or slit (Fujita et al 1992, Beutel et al 2000). This approach eliminates
aliasing arising from the limited sampling inherent in practical digital systems (Dobbins 1995)
and is called the presampling MTF. There are specific problems related to measurement of
MTF in CR systems due to their dependence on F (Lubinsky et al 1986) shown in figure 14.
MTF is reduced as F increases, which we call phosphor blunting. In figure 14(b) the
mechanism is explained in terms of the same Gaussian intensity profile used to discharge
an IP to different F. The increase in full width at half maximum along with the distortion
of the shape of the discharge is shown. The saturation and widening of the curves arises
from the nonlinear nature of photostimulation shown in figure 8. Another potentially opposite
effect, called beam sharpening improves the MTF as F increases and is shown in figure 15. In
(a) the raster is defined along with the position of the laser beam at the instant of interest,
R148 Topical Review

Figure 15. Outline of the principle of destructive reading of CR plates and the origin of laser beam
sharpening. Note that, for clarity, blurring arising from scattering in the screen has been neglected
in this diagram (black represents 100% and white 0%).

(b) shows the form of the x-ray latent image of a letter A stamped from an otherwise opaque
lead sheet, the three rows show, (c) the exhaustion image or map of the discharge fraction of
the image that represents the remaining potential for image readout, (d) the exhaustion image
multiplied by the original latent image, i.e. the total remaining latent image at this instant in the
readout procedure and (e) the image read out to the computer. The readout discharge fraction
(dependent on the laser energy per pixel) in the range 0100% are shown in the columns. In
(f ) where the laser beam size is doubled laser beam sharpening can occur if F is large. Beam
sharpening arises from the interaction of only the leading edges of the readout beam with the
remaining latent image resulting in better MTF at high F than at low. Operating the reader in
manual mode permits the control of F (Kengyelics et al 1998b).
The components of MTF for CR consist of (Ogawa et al 1995): (i) x-ray effects such as
scattering and fluorescence reabsorption MTFx-ray, (ii) blurring within the IP MTFIP, (iii) the
spatial response of the reader, including the laser spot size MTFlaser and, if the system
scans at a fine pitch then synthesizing a coarser pitch by averaging the pixel pitch and line
pitch creates MTFAverage( f ), and (iv) the response of the electronic temporal filter used for
antialiasing (figure 9) at the temporal frequency , Rtemporal() = MTFelec( f ) where = fv
and v is the scan speed (pixels per second pixel size) introduced to eliminate aliasing and
reduce noise. Combining these components and assuming that linear system analysis applies,
then

MTFR = MTFx-ray MTFIP MTFlaser MTFaverage MTFelec . (7)

Topical Review R149

Figure 16. Modulation transfer functions for Fuji ST-V and HR-V IPs measured in the same readout
system (Fuji AC3) for: (a) scan and subscan directions (Kengyelics et al 1998b) and (b) MTF( f )
in subscan direction is replotted on Swanks universal curve, i.e. against f T where T is the phosphor
layer thickness from table 2 (ST-V is 230 m and HR-V is 140 m) showing overlap of graphs.
Note a correction to the data was applied prior to scaling, the MTF for a 100 m laser spot size
was divided into the measured MTFs for both IPs.

Figure 16(a) shows the measured MTF for Fuji ST-V and HR-V IP in both scan and
subscan directions (Kengyelics et al 1998b). For each IP type, MTF is similar in both and,
in particular, the MTF is significant at the Nyquist frequency in the scan direction. This is
unexpected, as the purpose of the temporal filter is to reduce the MTF at the Nyquist frequency
to zero in the scan direction. In the authors opinion, this occurs because the anti-aliasing
filter is after the logarithmic amplifier (Ogawa et al 1995). The temporal filter operates in
the logarithmic domain, not the linear domain, leading to nonlinearity. Further evidence for
nonlinearity is that the MTF of a Fuji 5000 system measured using several different methods
yields radically different results (Fetterly et al 2002). In this case an MTF cannot be defined
for the filter, and thus a system MTF does not exist. The filter response can be seen clearly in
the noise power spectrum (see section 5.3.3). Fortunately these arguments do not apply to the
MTF in the subscan direction, which therefore is a true MTF of the system.
In the subscan direction the behaviour of the MTF in terms of the components in
equation (7) is as follows. The laser spot size provides an almost negligible contribution
to MTF, the temporal filter is not relevant and the MTFs shown are taken without averaging.
The only significant MTF factors are the x-ray interactions and laser scattering in the IP.
It is enlightening to apply Swanks (1973) scaling analysis to compare the MTFs of IPs
of differing thicknesses. From table 2 the ST-V IP thickness is almost twice that for HR-V.
After correcting for the laser spot size and pixel sizes, the MTFs can be rescaled onto the
same graph by plotting MTF as a function of the dimensionless quantity f T. In figure 16, the
curves for ST-V and HR-V superimpose precisely on Swanks universal curve, indicating that
for practical purposes the optical designs of these IPs are identical at the values of F set by the
reader. In conclusion, the MTF of IPs is similar to the expectation for any scattering screen.
Caution is advised since some readers do not operate linearly in the scan direction and in this
case, a meaningful MTF can only be defined in the subscan direction.

5.3.3. Noise power spectra. The NNPS (NPS normalized to signal) for the same IPs as
above are plotted in figure 17(a) in the scan and subscan directions. The presence of the
R150 Topical Review

0
f /fS 1
1E-3
(a) 75 kVp (W) 1.5 mm Cu 0.5 Gy (b)
Butterworth 6 th order filter f =3.35 lp/mm
Normalized Noise Power Spectrum (mm )
2

HR-V AC3
1E-4 1

Normalized fitler response

1E-5 ST-V AC3

ST-V AC3
HR-V AC3
1E-6

scan f /fS=0.5
1E-7
subscan

0.1
0 1 2 3 4 5 1 10

frequency (lp/mm) frequency (lp/mm)

Figure 17. (a) Normalized noise power spectra for both Fuji ST-V and HR-V obtained with an
AC3 reader (Kengyelics et al 1998b). The data are plotted in both the scan and subscan directions.
The scan direction shows a marked drop at high spatial frequencies that is directly attributable
to the temporal anti-aliasing filter shown conceptually in figure 9. The form of the filter can be
estimated by dividing the NNPS (scan) by NNPS (subscan) and the result is shown in (b) where it
can be seen that the filter is the same for both ST-V and HR-V as might have been expected since
the common element is the AC3 scanner where the filter is present.

temporal filter is clearly seen in the scan direction. Dividing the NPS in the scan direction
by that in the subscan direction and taking the square root (figure 17(b)) produces the filter
response. The filter for both IPs measured on the same reader under the same scan conditions
is the same. It can be fitted as a sixth-order Butterworth filter with a 3 dB point at 3.35 lp/mm
and a drop, by a factor of ten, at f/fc = 0.5 which is the Nyquist frequency (5 lp/mm). This
filter is therefore acting as a perfect anti-aliasing filter in the NPS measurements. It seems
inconsistent to plot the filter function here since the system, based on our analysis of MTF, is
nonlinear. However, an explanation is now feasible. During measurement of NPS, the image
is uniform with small variations due to noise. Thus the system will mimic a linear response, as
will any characteristic over a small enough excursion. However, the MTF situation is exactly
opposite. The range of values is very large, i.e. 0 when far from the position of the image
of the slit to a very significant value at the centre of the slit. Under these circumstances the
nonlinearities will be worst. Therefore the function shown in figure 17(b) is the true small
signal response of the system corresponding to the actual electronic circuit and measurements
of NNPS are valid in both the scan and subscan directions.

5.3.4. Detective quantum efficiency. How do we judge what is a good DQE? First, there is
an important quantity called DQE(0) the detective quantum efficiency at (or close to) f = 0.
Next there is the change, always a drop, of DQE as the spatial frequency increases. This is
the shape of the DQE or DQE( f )/DQE(0). The best DQE starts at unity and stays constant
at unity to the largest frequency of interest. However, for a sampled system, the DQE is not
defined above the Nyquist frequency fNY(half the sampling frequency). In practice DQE(0) and
DQE( f )/DQE(0) are intertwined and cannot be selected independently. DQE is calculated
from the MTF, NNPS and exposure to the system in terms of the mean number of x-ray
Topical Review R151

(a) (b) ST-II 901 75 kVp 0.5 mm Cu

0.25
ST-II 901 75 kVp 0.5 mm Cu
Average scan and subscan Average scan and subscan
0.3 Gy
0.20 0.1
f=0.5 lp/mm

DQE (f )
3 Gy
DQE

0.15

0.10 0.01
30 Gy

0.05
300 Gy f=2 lp/mm

0.00 1E-3
0 1 2 0.1 1 10 100

Spatial frequency (lp/mm) Exposure (Gy)

Figure 18. Illustration of the role of structure noise in the detective quantum efficiency of CR.
(a) DQE( f ) plotted as a function of exposure for an example IP/reader system (Fuji ST-II/901)
(Dobbins et al 1995). Note that the highest value of DQE is for the lowest exposure level due
to the decrease in x-ray noise as a function of exposure while structural noise remains constant.
(b) DQE( f ) plotted against exposure at two different spatial frequencies as shown in the diagram.
The air kerma at which DQE drops by a factor of 2 is illustrated.

photons incident per unit area  using the expression:

MTF2
DQE(f ) = . (8)
 NNPS
Structural noise is a fact of life for both the CR and screenfilm systems. In DR systems
it is normal to correct for structural noise or fixed pattern noise. However, CR systems
based on flying spot scanners lack the precise geometrical alignment for such corrections
to be successful. Structural noise NNPSStruct manifests itself at high exposures since it is
independent of exposure while NNPSX decreases linearly with exposure. Figure 18 shows
DQE( f ) for Fuji ST-II/901 IP/reader combination plotted against f for several exposure
values (Dobbins et al 1995). In this system quantum noise is dominant up to an exposure
of 1020 Gy (depending on spatial frequency) as seen in figure 18(b). This is adequate
immunity to structural noise for most medical uses of CR and has been achieved by incremental
improvements (Dobbins et al 1995, Ogawa et al 1995). Our discussion on DQE will now focus
on experimental values obtained at low exposures where the effect of structural noise can be
ignored. There is no evidence in a loss of DQE as exposure decreaseseven at 0.3 Gy
exposure (corresponding to approximately 1/10 of the mean exposure used in CR) and
therefore there cannot be any significant contribution to the overall NPS from digitization
noise indicating that the number of a/d bits is adequate.
In figure 19 an analysis of the effect of secondary quantum noise on DQE( f ) is shown.
In figure 19(a) the normalized measured noise power spectrum (NNPSmeas) is plotted as a
function of spatial frequency. From the flattening of NNPSmeas at high spatial frequency, the
spatial frequency independent secondary quantum noise NNPSSQ can be identified. Using
NNPSmeas, the DQE is calculated. Kengyelics et al (1998) have shown that this DQE is for
g = 14. Simply rescaling the values of NNPSSQ is equivalent to changing g. This has been
performed for g values of interest and the resulting values of DQE are plotted in figure 19(b).
Changing g has an effect on DQE especially at high f. When g is decreased, DQE( f ) drops
R152 Topical Review

1E-3
(a) (b) Measured DQE
0.30
Recalculated DQE
NPST=NPSX+NPSSQ
0.25

1E-4 ST-V AC3 subscan 0.03 mR

ST-V AC3 subscan 0.03 mR
NPS (mm )

0.20 75 kVp 1.5 mm Cu

2

75 kVp 1.5 mm Cu
g/AS+1

DQE
0.15
g=1000
1E-5
0.10
100
14 50
10
NPSSQ 0.05 4

1E-6 0.00
0 1 2 3 4 5 0 1 2 3 4 5

Spatial frequency (lp/mm) Spatial frequency (lp/mm)

Figure 19. Extraction of secondary quantum noise and calculation of optical coupling gain g.
(a) Plots of NNPS( f ) in the subscan direction (where the effect of the temporal filter can be
ignored) and at low exposure (where structural noise can be ignored) for Fuji ST-V IP and AC3
reader (Kengyelics et al 1998b). The fractional drop between the NNPS at f = 0 (due to both x-ray
noise NNPSX and secondary quantum noise NNPSSQ) and f very large (where NNPSSQ is the only
noise) is approximately equal to g and more precisely g/AS + 1. In (b) DQE( f ) for the same IP
and system is shown. A large part of the drop in DQE with increasing f is due to the increasing
importance of NNPSSC. This can be emphasized by estimating the value of DQE( f ) at other g
simply by changing the amount of frequency-independent NNPSSC.

Table 3. Contributions to the Swank factor in Fuji imaging plates readout in the AC3 reader.
Calculation of AOPD from measured AS from table 2 and estimated AXED, AAED and ASQ as
explained in text.

IP type AS AXED AAED ASQ AOPD

ST-V 0.70 0.90 0.85 0.95 0.91
HR-V 0.45 0.86 0.64

precipitously. If g is increased then DQE( f ) increases, but only to a point. For g = 1000,
secondary quantum noise is negligible and any further increase in g makes no significant
improvement in DQE( f ). DQE still decreases rapidly as f is increased. This is due to the
Lubberts (1968) effect where the MTF for the ith layer MTFi is a function of the depth of
absorption of the x-ray. If the same MTFi was valid
for all depths, then the DQE would be
independent of f, i.e. DQE proportional to MTFi2 NPSi with NPS proportional to MTFi2.
However, in the case of depth-dependent MTFi, the total MTF is a weighted sum of MTFi.
Similarly, the NPS is a weighted sum of NPSi each of which is proportional to MTFi2 at the
corresponding depths. Since the Schwartz inequality states that the square of sums is always
larger (equal only for the case of one layer) than the sum of squares, DQE( f )/DQE(0) is
always less than unity and decreases as f increases. This accounts for the drop in DQE seen
even when secondary quantum noise is negligible, i.e. g = 1000.
The value of DQE(0) can be expressed as components, primarily the quantum efficiency,
but also affected by several other correction terms (equation (5)). An analysis of these
terms has been performed for ST-V and HR-V IPs and the results are shown in table 3.
Topical Review R153

Using data obtained at low x-ray exposure has eliminated structural noise. ASQ, which arises
from the secondary quantum noise, has been calculated from the formula ASQ = (NPStotal
NPSSQ)/NPStotal. From table 3 the estimated values of AOPD (0.6 for HR-V with black
backing; 0.9 for ST-V with reflective backing for PSL) are consistent with the known design
of the modern IPs and when is assumed that the reader has been set at a readout depth F 0.5
or more.
Systematic measurements of the spatial frequency dependent parameters: modulation
transfer function MTF, Wiener noise power spectrum NPS and detective quantum efficiency
DQE( f ), have been performed for several generations of Fuji CR systems (Hillen et al 1987,
Dobbins et al 1995, Kengyelics et al 1998b, Samei and Flynn 2002) and IPs (Kengyelics et al
1998a). A gradual improvement in CR IP and reader technology is evident and due
to incremental improvements in several independent factors (Ogawa et al 1995). Other
readers and IPs have also been investigated: Kodak (Bradford et al 1999), Lumisys (Fetterly
and Hangiandreou et al 2000), inter-comparisons of all the former systems with Agfa
(Samei and Flynn 2002) and Fuji IPs with Konica IP on the same Konica reader (Nakano et al
2002). General rules for the evaluation of systems have been proposed (Samei et al 2001).
Super high-resolution systems have been evaluated (Flynn and Samei 1999, Kengyelics et al
1998a). The kVp dependence of DQE has also been investigated (Fetterly and Hangiandreou
2001).
The difference between the DQE( f ) of ST-V and HR-V IPs is summarized in figure 20(a).
The DQE(0) of ST-V is approximately twice that of HR-V which can be expected due to the
almost doubling of phosphor loadings (table 2). The DQE( f ) at high frequency is higher for
HR, at least if the reader is optimized. This is similar to conventional screenfilm combinations
shown in figure 20(b).
The DQE( f ) of CR systems is lower than a perfect system could accomplish but can be
completely understood from the limitation of the scattering phosphors and the CR readers.

5.4. First principles calculations of imaging properties

It would be advantageous in the design and optimization of CR systems to have a complete
analytical model describing the readout of a CR plate. Such a model has been developed
(Lubinsky et al 1986, Korn et al 1986, Lubinsky et al 1987). In their model, the latent image
is read, point-by-point, in the IP as the laser beam traces out the raster for all depths in the
phosphor layer. The blurring of the laser readout light is modelled using analytical approaches
developed by Kubella and Munk (Kortum 1969) and Swank (1973). The model has so far
had only limited application (to a prototype IP design). It describes an inherently nonlinear
process containing aspects of nonlinear discharge of the latent image and destructive scanning
in two dimensions. To analyse needle-structured phosphors the modelling will have to be
extended. This work has commenced in the context of XRIIs (Hillen et al 1991) and flat-panel
detectors (Rowlands and Yorkston 2000). Considerable further work is needed before making
a practical, first principles, calculation of MTF, NPS and DQE for arbitrary designs of CR
systems. Once complete, this work should create a firm foundation for the optimization of the
CR system design.

5.5. How close are powder phosphor CR systems to fundamental limits?

Current CR systems are not optimal for image quality, but are practical and flexible. The best
photostimulable phosphor, BaFX:Eu2+ has an atomic number and density less than the best
conventional phosphors (table 1) and has a gain lower than theoretically possible. However,
R154 Topical Review

(a) (b)
0.30 0.30

Hi-Plus/XRP
0.25 0.25
ST-V AC3 g =16.6
75 kVp 1.5 mm Cu
0.20 0.20
DQE

DQE
0.15 0.15

HR-V 7000 g =8
0.10 75 kVp 0.5 mm Cu 0.10 Par/XRP

0.05 HR-V AC3 g =4 0.05

75 kVp 1.5 mm Cu

0.00 0.00
0 1 2 3 4 5 0 1 2 3 4 5
frequency (lp/mm) frequency (lp/mm)

Figure 20. (a) DQE comparisons between Fuji ST-V and HR-V with AC3 reader (Kengyelics et al
1998b) and HR-V with 7000 reader (Dobbins et al 1995). The concept of using HR-V for general
radiography is that a loss of DQE(0) due to reduced absorption of IP permits reduced thickness of
phosphor layer, which permits better MTF that will allow DQE to be maintained above DQE for
ST-V at high enough spatial frequencies. However, this is true only for one of the two IP/reader
combinations shown here. A further requirement is that g must be kept large as well. The values of
g are shown on the plotted lines (obtained from Kengyelics et al (1998) and, for HR-V 7000 reader,
estimated by the author using the procedure described in the text). (b) Classic DQE as a function of
the thickness diagram (Sandrik and Wagner 1982) for conventional screenfilm systems showing
crossing of DQE curves due to thickness, absorption and MTF tradeoff. Note that Hi-Plus is
approximately twice the CaWO4 phosphor loading of Par.

unless a new higher density photostimulable phosphor and a means to increase the conversion
gain are discovered, the powder phosphor approach combined with the flying spot reader is
close to a physical limit. Thus a radical new approach is required.

6. Concepts for improved CR

What new approaches are available which might improve on conventional powder phosphors
CR used with a flying spot scanner? Several possibilities will be examined and estimates made
of possible improvements.

6.1. Dual-sided readout

The concept of dual-sided readout is shown in figure 21 (Arakawa et al 1999, 2000). The
substrate of the double-sided IP (DS IP) is made transparent. The advantages are twofold:
(i) the IP construction is simpler, there is no need to make a special anti-halation layer,
(ii) the admixture of the top and bottom light outputs is digital and so can be optimized at
each frequency. The potential improvement in DQE is not yet known as dual-sided readout
systems have not been compared directly with optimized single-sided readout systems. The
estimate for DQE improvement is of the order of a factor of 2060% depending on the
application.

6.2. Needle-structured phosphors

Figure 22(a) shows conceptually the image quality improvements using needle-structured
phosphors. Laser light is guided from the surface to the point of x-ray interaction and
Topical Review R155

Figure 21. Two-sided readout system with two light collection systems and a single laser
(Arakawa et al 1999, 2000). (a) Concept, (b) IP structure, (c) laser light penetration plotted
as a function of the depth of phosphor and (d) PSL escape efficiency to top and bottom PMTs.

Figure 22. Properties of needle-structured phosphors imaging plates. (a) Optical properties of
the IP showing forward-directed PSL light emission and distribution of laser light reflection and
scattering and (b) read-out method to ensure the collection of the majority of the forward-directed
PSL light while eliminating the direct reflection of the laser.

PSL is guided back to the surface by a structure similar to a fibre optic plate. Blue light
emitting CsI:Na (Stevels and Schrama-dePaw 1974) was developed in a needle-structured
form for use in XRIIs (Vosburgh et al 1977). More recently green emitting structured
CsI:Tl has been developed (Nagarkar et al 1998) for digital radiographic systems using
silicon detectors (Yaffe and Rowlands 1997), e.g. indirect conversion flat-panel imagers
(Rowlands and Yorkston 2000). The layers of CsI are made by evaporation in a vacuum
under carefully controlled conditions such that the material crystallizes into long needle-
shaped crystals, perpendicular to the substrate. The size of the crystals is 110 m in
diameter. Subsequently, heat is used to crack the crystals along their long boundaries to
separate them into columns. Ideally the columns would be the full thickness of the layer
and 50 m in diameter, i.e. considerably larger than the individual crystals but less than
a typical pixel size, separated by cracks several wavelengths thick, e.g. 1 m, to prevent
transmission of the evanescent ray between columns. Alternatively, special substrates can
create crystal growth in a preferential manner so that spontaneously cracking occurs during
the evaporation process (Jing et al 1992). The control of these processes is very difficult.
R156 Topical Review

The precise preparation approaches are trade secrets. Probably various methods are used so
different imaging properties may result.
For CR there are two image quality advantages of a structured IP compared to a powder
IP made using the same phosphor. First, the effective density of a structured phosphor is
almost doubled, as the screen (except for minor losses due to cracking) is 100% phosphor
compared to the 60% packing factor in a powder phosphor (table 2). Secondly, the blurring in
a structured phosphor of a given thickness is empirically the same as for a powder phosphor
screen of half that thickness (Rowlands and Yorkston 2000). Thus a needle phosphor screen
can have an increase in mass loading by a factor of four compared to a powder screen made
from the same phosphor and yet maintain similar spatial resolution.
A chest system has been developed using a 300 m layer of RbBr:Tl+ (Nakazawa et al
1990). It is encapsulated to protect the hygroscopic phosphor from fading due to humidity
damage. A favourable phosphor microstructure was achieved using special substrates prepared
by flame deposition of polymers.
The readout of IPs made with structured phosphors requires a different approach than for
powder phosphors. It must take into account the highly specular reflections from the phosphor
surface and the forward directionality of the emitted PSL light as shown in figure 22(a).
The readout approach in figure 22(b) was developed to give the laser light a large angle of
incidence. This permits the acceptance of the forward-directed PSL and avoids acceptance of
the reflected laser beam (Shimada et al 1991, Isoda et al 2001, Schaetzing et al 2002).
Recent developments in needle phosphors have used a 500 m layer of CsBr:Eu2+
(Leblans et al 2001). These show much better DQE( f ) than is possible with BaFX:Eu2+and
approach that possible with flat-panel DR systems.

6.3. Line scanning readers

In a line scanning reader, a bar is moved across the plate (or the plate may be moved under
the scanner bar) to facilitate the readout of a complete line or lines (figure 23). An array of
light emitting diodes (or lasers) and lenses create a finely focused line of stimulating (red)
light on the surface of the IP. The resulting PSL is focused onto a silicon photodiode array.
Photodiodes can be used in this application because each photodiode can be very small (e.g.
0.1 mm in the scan direction by 1 mm in the subscan direction) yielding a very low electronic
noise that does not overwhelm the PSL signal.
Line scanning readers have several advantages over flying spot scanners. Reading out a
complete line potentially speeds up the readout process so that a complete line can be read out
in the time it would take a flying spot scanner to read a pixel. This could, in principle, result
in a three order of magnitude increase in readout rate, but to achieve this would also require
a corresponding increase in laser power. Line scanning readers also offer fundamental image
quality advantages; improvement in the system gain g, shown in figure 11 is the primary one.
Two effects contribute to the improvement in g: (i) the collection of photons can be improved
by having the image sensor closer to the irradiated area of the plate; (ii) the conversion
efficiency of light to charge in the photodiode, is potentially 4 times greater than a PMT.
Together these two effects could increase g by an order of magnitude which would have a
very significant impact on DQE as seen in figure 19(b). Another image quality advantage is
that the readout process is similar to that of flat-panel imagers and thus resolution depends on
the PSL spreading, not the laser beam spreading (only for the subscan direction of single line
readers). Therefore nonlinearities that cause phosphor blunting (figure 14) do not occur in the
scan direction for either type of reader but do remain in the subscan direction for single line
scanners (Isoda et al 2001).
Topical Review R157

Figure 23. Schematic of the line scanner that permits readout from an entire row (or rows) of the
image at a time. (a) Detailed cross section of a reflection mode scan head in the cross scan direction
and (b) transmission mode. (c) Overall concept of the scanner where the plate is moved while a
line (slit scan) or bar (slot scan) of stimulating red light is projected onto the IP and the emitted
PSL is collected by the lens array and focused on the linear photodiode array or multiline CCD.
Note that the MTF is controlled by the laser spread in the subscan direction and by the spread of
the PSL in the scan direction. The multiline CCD has the capability of scanning the latent charge
image across the CCD plane in the subscan direction (such devices are known as time domain
integration (TDI) CCDs) simultaneously and in synchronization with the mechanical scanning of
the IP.

6.4. Transparent photostimulable phosphor screens

Figure 24 shows the improvements possible using transparent phosphor screens. It is necessary
to carefully define a transparent CR IP. It is essentially transparent to the laser and PSL light
but to function it must have some interactions with the stimulating light and be capable of
absorbing a fraction of it. Perhaps unexpectedly, a transparent phosphor can be used with the
flying spot CR readout method (though not single or multiple line scanners) to create a higher
image resolution than with conventional scattering IPs or even needle-structured phosphors.
The reason is shown in figure 24. The stimulating laser beam is passed directly through the
transparent IP, interacting only with PSL centres and thus resolution is, in principle, defined by
the unscattered laser beam diameter. However, there are major problems with this approach:
(i) laser light which scatters from unintentional but inevitable defects on the surfaces of the
IP, e.g. scratches and dust particles, reflects within the transparent layer causing flarea low
frequency drop in MTF. (ii) Since the laser passes straight through, without being delayed
by scattering, a more powerful laser is required. No estimates of the increase in laser power
appear in the literature but it must be several orders of magnitude. (iii) Unless the laser beam is
incident normally on the IP, the light reflected from the exit plane of the phosphor will traverse
through the phosphor and so reach other points within the IP and cause blurring. Normal
incidence can be achieved using a telecentric scan lens so that the laser beam always emerges
parallel to the principal axis of the lens. (iv) It is much more difficult to collect the PSL light
from a transparent than a scattering IPs, as much of the light can be trapped within the IP by
total internal reflection. A solution proposed by DeBoer and Luckey (1988) uses a powder
phosphor in a binder whose refractive index matches that of the phosphor for red light making
the IP transparent for the stimulating laser beam. The materials are chosen to have different
refractive indices for the blue PSL. Thus the layer acts as a scattering phosphor for the PSL,
which can be collected in the usual way. The transparent phosphor approach is appealing but
R158 Topical Review

Figure 24. Transparent plate CR system. (a) Showing principle of creation of high resolution
image by readout of a transparent plate. (b) Showing origin of geometrical blurring due to the
oblique incidence of x-rays at the edge of the field of view of the image combined with the high
resolution expectation for a transparent phosphor and the increased thickness permitted by the
approach. The virtual pixel is the region readout by each position of the laser beam at which
digitization is performed, and it can be seen that the path of the x-ray is different from the pixel.
(c) The readout method shown uses a telecentric scan lens which, as well as acting as a scan lens
with F/ characteristics also has the property of light emerging from the lens, is parallel to the
principal axis of the lens and so is always incident normally on the plate/lens system. In addition
this system has the feature of readout in the same direction that the x-rays were incident, thus
ensuring elimination of geometric blurring independently of the thickness of the IP, i.e. the x-ray
absorption path is parallel to the division line between virtual pixels.

not ideal. Considering the extremely high sensitivity to unwanted scattering events and the
high laser power required, it is suggested that this system cannot easily be made practical.

6.5. Dual energy

By obtaining two images at different x-ray energies and by appropriate manipulation, removal
of the complex bony detail may be accomplished to aid in visualization of soft tissue
abnormalities. This is possible because the attenuation of x-rays by matter is due to two
independent processes, photoelectric absorption and Compton scattering, each of which have
different energy dependences. At each energy, the corresponding attenuation coefficients have
a constant ratio for any element in the periodic table. Therefore two images obtained at
different energies can be re-expressed as a ratio of the quantity of one element (or an arbitrary
mixture of elements) to another, e.g. bone to soft tissue. In early applications, two separate
exposures were made at different kVps and captured using an x-ray image intensifier XRII. In
general, the farther apart the energies of the two imaging beams, the better the separation of
the images and the lower the noise.
In applying dual energy for general radiology, an XRII is impractical so screenfilm or
CR has been used. Two cassettes cannot be moved mechanically between exposures quickly
enough to avoid patient motion artefacts. Thus the two exposure approach used with XRIIs
is not practical. The solution proposed was the single cassette/single exposure approach in
which two (Ishigaki et al 1988, Kido et al 1993, Takeo et al 1996) or more (Ergun et al 1990)
IPs are exposed simultaneously. This eliminates patient movement and the dual energy effect is
obtained by the preferential absorption of softer radiation in the front IP and harder in the back
IP. The spectra of the two images can be further separated by interposing a beam-hardening
filter between the front and back IPs, but with the disadvantage of loss of quantum efficiency
Topical Review R159

(Stewart and Huang 1990) and/or by using a stack of several IPs each subject to a slightly
different effective energy (Ergun et al 1990). Despite considerable effort, simultaneous high
quantum efficiency and good spectral separation in these images could not be obtained. At the
same exposure level much noisier bone/tissue separated images are obtained using the single
rather than the dual exposure method.
Alvarez (1996) solved the problem of a dual plate CR system with no moving parts yet
achieving a high degree of spectral separation between the two images. He used the erasability
of CR plates to establish a highly differentiated x-ray spectrum to be absorbed in each IP. First
he made a high kVp exposure to create a latent image in both the front and back IPs. This
image was erased from the front plate by flooding it with light. Once erased, the second low
kVp image was made. Due to the combination of the softness of the beam and an intermediate
beam-hardening filter, the second image was absorbed essentially only by the front IP. The
image separation in mean spectral energy is much higher than with the single kVp approach.
This leads to a significant increase in SNR almost equivalent to an instant readout system
such as a flat-panel imager or XRII but at much lower cost.

6.6. Overcoming the thickness-resolution optimization problemmultiple plate readout

The first clinical CR images seen by the author were part of an RSNA scientific exhibit in the
early eighties. A stack of seven thin CR plates exposed simultaneously in the same cassette
provided the raw images. These were subsequently digitized and recombined into a single
digital image. This approach yields an MTF( f ) of a very thin plate, has the quantum absorption
of a plate seven times as thick and an excellent DQE( f ) at all relevant spatial frequencies. In
practice, this multiple plate approach was not adopted outside Japan. Presumably there was
difficulty in registration (Shaw et al 1997) and lengthy time requirements.
This concept of plate stacking has been rediscovered (Barnes 1993). Two ordinary IPs
were loaded into the same cassette (Workman and Cowen 1995) and the images were matched
by cross-correlation techniques and recombined. The improvement in x-ray absorption is
<100% since the second IP can only absorb x-rays not already absorbed by the first. Thus
it is impressive that Shaw et al (1997) were able to show a 60% increase in DQE, i.e.
an improvement of SNR of 30%. This approach has considerable merit as a means of
overcoming the thickness blurring tradeoff in IPs. However, it needs a new concept to make
the recombination of a considerable number of IPs a practical reality.

7. Clinical application of complete CR systems

The first issue in establishing a new imaging system is how it compares to previous systems
(Cowen et al 1993). For x-ray systems, the important measurements are image quality and
patient dose, the former being highly dependent on the latter. Computed radiography has been
compared to the digitized film from a filmscreen combination (Yoshimura et al 1993). Image
quality is very difficult to compare across imaging systems because of their inherently different
properties (Fuhrman et al 1988). An interesting technical concept is to expose the CR IP in
the same cassette with a conventional screenfilm system, thus obtaining a perfectly matched
CR and screenfilm image in the same exposure and without additional radiation (Chotas
et al 1991, MacMahon et al 1991, Sanada et al 1991, Wilson and West 1993). Recent studies
have used more conventional approaches (Cook et al 1994) and comparisons of CR against
DR systems (Rong et al 2001). There is little agreement as to whether the image quality of
CR is better, worse or the same as screenfilm.
R160 Topical Review

In a digital system the exposure can be varied over a large range yet still deliver an
optimally bright image with a correct mean intensity and tone scale. In practice, there is a very
narrow margin where the image is subjectively equivalent to screenfilm. This is the basis for
the statement that twice the exposure is necessary for standard CR plates (Huda et al 1997,
Seibert et al 1996) to achieve the image noise perception characteristic of commonly used 400
speed screenfilm, i.e. 2.5 Gy mean exposure to obtain an optical density of 1 above base
plus fog, since standard CR plates are considered to be 200 speed (5 Gy mean exposure).
This is a very subjective criterion, based on systems with differing MTFs and noise behaviour
but it is significant as it represent a clinical consensus.
New systems mean new artefacts. The form of these artefacts and the diseases they
may camouflage or imitate are well documented (Oestmann et al 1991, Solomon et al 1991,
Cesar et al 2001).
A primary criterion used by radiologists in comparing image quality in films is the
appearance of x-ray noise in the image or quantum mottle. Any visual signs of mottle make an
image unacceptable. The blurring in CR, by reducing the x-ray bandwidth, effectively reduces
the appearance of noise. The control of latitude (Chotas and Ravin 1992) is important and the
digital nature of CR makes quantification of images possible (Floyd et al 1990). The specific
clinical applications of CR will now be reviewed.

7.1. Portable
Photostimulable phosphor systems are widely used for emergency and bedside radiography
where the variable readout sensitivity allows for compensation for under- and overexposure
problems. Portable chest radiography presents particular problems (Niklason et al 1993)
especially in intensive care situations (Jennings et al 1992). In bedside applications accurate
lining up a grid is impractical and thus scatter (Floyd et al 1992) is difficult to control.

7.2. General radiography

For chest radiography, comparative studies of the threshold perception of CR compared to
screenfilm (Dobbins et al 1992, Launders and Cowen 1995), indirect conversion flat-panel
imagers (Rong et al 2001), and selenium-based systems (Bernhardt et al 1999) have been
performed. Radiation dose requirements for adults (Marshall et al 1994) and paediatrics
(Kogutt et al 1988, Huda et al 1996) have been investigated. Oda et al (1996) determined the
optimal beam quality and Lo et al (1994) showed the advantages of correcting for scatter using
a beam stop technique. A one-shot dual energy CR procedure (Ishigaki et al 1988) has been
investigated for the detection of pulmonary nodules (Kido et al 1993) and has been compared
with asymmetrical screenfilm (Kelcz et al 1994). Further studies with variation of dose have
been performed (Kimme-Smith et al 1995, 1996).
For both adults (Murphey et al 1992) and children (Kottamasu et al 1997)
musculoskeletal, extremities (Wilson et al 1991), hand (Swee et al 1997), cervical spine
(Wilson et al 1994) and scoliosis (Jonsson et al 1995) have been identified as useful
applications of CR. The combination of equalization with CR has also been investigated
(de Rooy et al 1993, Dobbins et al 1993).
The application of CR to radiographic magnification is obvious, since magnification
requires increasing the distance between the detector and patient, and hence the IP from the
x-ray source. Using screenfilm with a fixed radiographic speed requires additional radiation
to the patient to obtain an adequately exposed image. CR works well at reduced dose so that
the exposure can be based on the radiation requirement of the imaging task and not simply the
Topical Review R161

technology (Nakano et al 1986, 1987). A similar argument can be used for the application of
CR to phase contrast imaging where the need for very large patientimager distances combined
with as small a magnification as possible (to avoid focal spot blurring) gives an additional
freedom for optimization not available with screenfilm (Ohara et al 2002).

7.3. Mammography

Early CR mammography used a 100 m pixel and conventional HR IP (Higashida et al

1992, Cowen et al 1997b) with a mixed reception. Using a contrastdetail phantom specially
designed for mammography (Jarlman et al 1991, Cowen et al 1992), study of the visibility
of microcalcifications (Cowen et al 1997a) and linear analysis (Workman et al 1994)
demonstrated comparable performance to screenfilm despite a poorer limiting resolution.
These results were confirmed using a clinical ROC study (Brettle et al 1994).
However, the introduction of flat-panel DR systems with improved dose efficiency and
MTF (Cowen et al 1997b) have raised expectations. CR readers with 50 m pixel size and
a two-sided readout mammographic system (Jouan 1999, Arakawa et al 2000) have shown
improved results (Seibert et al 2002) and clinical trials of these systems have begun.

8. Concluding remarks

Computed radiography, based on the use of photostimulable phosphor IPs, is a very practical
approach for digital radiography. It has penetrated practically every niche in clinical
radiography. Combined with a cassette, it can be used in conventional x-ray rooms. IPs
can be conveniently produced in large areas. The readout time is comparable to the standard
screenfilm and thus must be regarded as acceptable. It is not instantaneous so there is no
immediate image evaluation on acquisition as in DR. Unlike screenfilm, the CR readout
process requires no consumable items. As PACS (picture archiving and communication
systems) become more widespread, there will be little if any need for film for diagnosis
or archiving. CR IPs have a linear response over a wide range of x-ray intensities, and
are erased simply by exposure to a uniform stimulating light source. Systems based on
the use of flying spot scanners and the BaFBr0.85I0.15:Eu2+ and BaFI:Eu2+ powder phosphor
IPs have been highly developed and perfected to the limits of powder phosphor technology.
Unfortunately physical limits prevent this mode of CR reaching ideal behaviour and the CR
systems have a relatively poor resolution and DQE compared to DR systems such as flat-panel
imagers.
Methods to improve CR performance are possible and need to be investigated. Study
of needle phosphor IPs, dual-sided readout and linear scanners and possibly the use of
transparent phosphors may allow these new methods to be competitive. The theoretical
basis of CR does not yet permit the accurate prediction of image properties. CR seems poised
to make an explosive leap forward. This should be taken as an opportunity and a challenge for
medical physicists to contribute to the further development and understanding of this important
modality.

Acknowledgment

The National Cancer Institute of Canada (NCIC) financially supports this work through a
Terry Fox Program Project Grant Imaging for Cancer.
R162 Topical Review

References

Alvarez R E 1996 Active energy selective image detector for dual-energy computed radiography Med. Phys. 23
173948
Arakawa S, Itoh W, Kohda K and Suzuki T 1999 Novel computed radiography system with improved image quality
by detection of emissions from both sides of an imaging plate Proc. SPIE 3659 572-81
Arakawa S, Yasuda H, Kohda K and Suzuki T 2000 Improvement of image quality in CR mammography by detection
of emissions from dual sides of an imaging plate Proc. SPIE 3977 590600
Arnold B A 1979 Physical characteristics of screenfilm combinations The Physics of Medical Imaging: Recording
System Measurements and Techniques (AAPM Medical Physics Monograph no 3) ed A G Haus (New York:
AIP) pp 3071
Barnes G T 1993 Digital x-ray image capture with image intensifier and storage phosphor plates: imaging principles,
performance and limitations Digital Imaging (AAPM Monograph no 22) ed W R Hendee and J H Trueblood
(Madison: Medical Physics) pp 2348
Bernhardt T M, Otto D, Reichel G, Ludwig K, Seifert S, Kropf S and Rapp-Bernhardt U 1999 Detection of simulated
interstitial lung disease and catheters with selenium, storage phosphor and film-based radiography Radiology
213 44554
Beutel J, Kundel H L and VanMetter R L 2000 Handbook of Medical Imaging Vol 1. Physics and Psychophysics
(Bellingham: SPIE)
Blasse G and Grabmaier B C 1994 Luminescent Materials (Berlin: Springer)
Bogucki T M, Trauernicht D P and Kocher T E 1995 Characteristics of a storage phosphor system for medical imaging
Technical and Scientific Monograph No 6 (Eastman Kodak Health Sciences Division)
Boone J M, Seibert J A, Sabol J M and Tecotzky M A 1999 Monte Carlo study of x-ray fluorescence in x-ray detectors
Med. Phys. 26 90516
Boone J M 2000 X-ray production, interaction, and detection in diagnostic imaging Medical Imaging Vol 1. Physics
and Psychophysics ed J Beutel, H L Kundel and R L Van Metter (Bellingham: SPIE) pp 178
Bradford C D, Peppler W W and Dobbins J T III 1999 Performance characteristics of a Kodak computed radiography
system Med. Phys. 26 2737
Brettle D S, Ward S C, Parkin G J, Cowen A R and Sumison H J 1994 A clinical comparison between conventional
and digital mammography utilizing computed radiography Br. J. Radiol. 67 4648
Cesar L J, Schueler B A, Zink F E, Daly T R, Taubel J P and Jorgenson L L 2001 Artefacts found in computed
tomography Br. J. Radiol. 74 195202
Chan H-P and Doi K 1984 Studies of x-ray energy absorption and quantum noise properties of x-ray screens by use
of Monte Carlo simulations Med. Phys. 11 3746
Chotas H G and Ravin C E 1992 Digital radiography with photostimulable storage phosphors: control of detector
latitude Invest. Radiol. 27 8228
Cook L T, Insana M F, McFadden M A, Hall T J and Cox G G 1994 Comparison of the low-contrast detectability of
a screenfilm system and third generation computed radiography Med. Phys. 21 6915
Cowen A R, Brettle D S, Coleman N J and Parkin G J 1992 A preliminary investigation of the imaging performance
of photostimulable phosphor computed radiography using a new design of mammographic quality control test
object Br. J. Radiol. 65 52835
Cowen A R, Launders J H, Jadav M and Brettle D S 1997a Visibility of microcalcifications in computed and
screenfilm mammography Phys. Med. Biol. 42 153348
Cowen A R, Parkin G J S and Hawkridge P 1997b A direct digital mammography image acquisition Eur. J. Radiol. 7
91830
Cowen A R, Workman A and Price J S 1993 Physical aspects of photostimulable phosphor computed radiography Br.
J. Radiol. 66 33245
Cunningham I A 2000 Applied linear-systems theory Handbook of Medical Imaging Vol 1. Physics and Psychophysics
ed J Beutel, H L Kundel and R L VanMetter (Bellingham: SPIE) pp 79160
Dainty J C and Shaw R 1974 Image Science: Principles, Analysis and Evaluation of Photographic-type Imaging
Processes (London: Academic)
DeBoer C D and Luckey G W 1988 Screens for storing x-ray images and methods for their use US Patent
no 4 733 090
de Rooy T P, Oestmann J W, Schultze-Kool L J, Vrooman H and Buchmann F 1993 Advanced multiple beam
equalization radiography (AMBER) combined with computed radiography preliminary evaluation Acta Radiol.
34 4459
Dobbins J T III 1995 Effects of undersampling on the proper interpretation of modulation transfer function, noise
power spectra and noise equivalent quanta of digital imaging systems Med. Phys. 17181
Topical Review R163

Dobbins J T III, Ergun D L, Rutz L, Hinshaw D A, Blume H and Clark D C 1995 DQE(f ) of four generations of
computed radiography acquisition devices Med. Phys. 22 158193
Dobbins J T III, Rice J J, Beam C A and Ravin C E 1992 Threshold perception performance with computed and
screenfilm radiography: implications for chest radiography Radiology 183 17987
Dobbins J T III, Rice J J, Goodman P C, Patz E F Jr and Ravin C E 1993 Variable compensation chest radiography
performed with a computed radiography system: design considerations and initial clinical experience Radiology
187 5563
Dolazza E and Poulo L 1984 Optimal quantization of noisy signals Proc. SPIE 454 40317
Drangova M and Rowlands J A 1986 Optical factors affecting the detective quantum efficiency of radiographic screens
Med. Phys. 13 1507
Ergun D, Mistretta C A, Brown D E, Bystrianyk R T, Kwong Sze W, Kelcz F and Naidich D P 1990 Single-exposure
dual-energy computed radiography: Improved detection and processing Radiology 174 2439
Fahrig R, Rowlands J A and Yaffe M J 1995 X-ray imaging using amorphous selenium: detective quantum efficiency
of photoconductive image receptors for digital mammography Med. Phys. 22 15360
Fetterly K A and Hangiandreou N J 2000 Image quality evaluation of a desktop computed radiography system Med.
Phys. 27 266979
Fetterly K A and Hangiandreou N J 2001 Effects of x-ray spectra on the DQE of a computed radiography system
Med. Phys. 28 2419
Fetterly K A, Hangiandreou N J and Schueler B A 2002 Measurements of the presampled two-dimensional modulation
transfer function of digital imaging systems Med. Phys. 29 91321
Floyd C E, Baker J A, Lo J Y and Ravin C E 1992 Measurement of scatter fractions in clinical bedside radiology
Radiology 183 85761
Floyd C E, Chotas H G, Dobbins J T III and Ravin C E 1990 Quantitative radiographic imaging using a photostimulable
phosphor system Med. Phys. 17 4549
Floyd C E, Lo J Y, Chotas H G and Ravin C E 1991 Quantitative scatter measurement in digital radiography using a
photostimulable phosphor imaging system Med. Phys. 18 40813
Flynn M J and Samei E 1999 Experimental comparison of noise and resolution for 2k and 4k storage phosphor
radiography systems Med. Phys. 26 161223
Fuhrman C R, Gur D, Good B, Rockette H, Cooperstein L A and Feist J H 1988 Storage phosphor radiographs versus
conventional film: Interpreters perceptions of diagnostic quality Am. J. Roentgenol 150 10114
Fujita H, Tsai D-Y, Itoh T, Doi K, Morishita J, Ueda K and Ohtsuka A 1992 A simple method for determining the
modulation transfer function in digital radiography IEEE Trans. Med. Imaging 11 349
Fujita H, Ueda K, Morishita J, Fujikawa T, Ohtsuka A and Sai T 1989 Basic imaging properties of a computed
radiographic system with photostimulable phosphors Med. Phys. 16 529
Gingold E and Schaetzing R 2001 Recent advances in computed radiography technology 2001 AAPM Annual Meeting
(private communication)
Ginzburg A and Dick C E 1993 Image information transfer properties of x-ray intensifying screens in the energy
range from 17 to 320 keV Med. Phys. 20 101321
Higashida Y, Moribe N, Morita K, Katsuda N, Hatemura M, Takada T, Takahashi M and Yamashita J 1992 Detection
of subtle microcalcifications: Comparison of computed radiography and screenfilm mammography Radiology
183 4836
Hillen W, Eckenbach W, Quadfleig P and Zaengel T 1991 Signal-to-noise performance in cesium iodide x-ray
fluorescent screens Proc. SPIE 1443 12031
Hillen W, Schiebel U and Zaengel T 1987 Imaging performance of a digital storage phosphor system Med. Phys. 14
74551
Holl I, Lorenz E and Mageras G 1988 A measurement of the light yield of common inorganic scintillators IEEE
Trans. Nucl. Sci. 35 1059
Huda W, Rill L N and Bruner A P 1997 Relative speeds of Kodak computed radiography phosphors and screenfilm
systems Med. Phys. 24 16218
Huda W, Slone R M, Belden C J, Williams J L, Cumming W A and Palmer C K 1996 Mottle on computed radiographs
of the chest in pediatric patients Radiology 199 24952
Ishigaki T, Sakuma S and Ikeda M 1988 One-shot dual-energy subtraction chest imaging with computed radiography:
clinical evaluation of film images Radiology 168 6772
Isoda Y, Nishihata S, Arakawa S, Takahashi K, Miyagawa I and Khoda K 2001 Radiation image readout method and
apparatus US Patent no 6 326 636
Jarlman O, Samuelson L and Braw M 1991 Digital luminescence mammography Acta Radiol. 32 1103
Jennings P, Padley S P and Hansell D M 1992 Portable chest radiography in intensive care: a comparison of computed
and conventional radiography Br. J. Radiol. 65 8526
R164 Topical Review

Jing T, Cho G, Drewery J, Fujieda I, Kaplan S N, Mireshghi A, Perez-Mendez V and Wildermuth D 1992 Enhanced
columnar structure in CsI layer by substrate patterning IEEE Trans. Nucl. Sci. 39 11958
Jonsson A, Jonsson K, Eklund K, Holje G and Pettersson H 1995 Computed radiography in scoliosis: diagnostic
information and radiation dose Acta Radiol. 36 42933
Jouan B 1999 Digital mammography performed with computed radiography technology Eur. J. Radiol. 31 1824
Kato H 1994 Photostimulable phosphor radiography design considerations Specification, Acceptance Testing and
Quality Control of Diagnostic X-ray Imaging Systems (AAPM Monograph no 20) ed J A Seibert, G T Barnes
and R G Gould (Woodbury: AIP) pp 73170
Kato H 2002 Private communication
Kengyelics S M, Davies A G and Cowen A R 1998a A comparison of the physical imaging properties of Fuji ST-V,
ST-VA, and ST-VN computed radiography image plates Med. Phys. 25 21639
Kengyelics S M, Launders J H and Cowen A R 1998b Physical imaging performance of a compact computed
radiography acquisition device Med. Phys. 25 35460
Kelcz F, Zink F E, Peppler W W, Kruger D G, Ergun D L and Mistretta C A 1994 Conventional chest radiography
vs dual-energy computed radiography in the detection and characterization of pulmonary nodules Am. J.
Roentgenol. 162 2718
Kido S, Ikezoe J, Naito H, Tamura S, Kozuka T, Ito W, Shimura K and Kato H 1993 Single-exposure dual-energy
chest images with computed radiography Evaluation with simulated pulmonary nodules Invest. Radiol. 28 4827
Kimme-Smith C, Aberle D R, Sayre J W, Hart E M, Greaves S M, Brown K, Young D A, Deseran M D, Johnson T
and Johnson S L 1995 Effects of reduced exposure on computed radiography: comparison of nodule detection
accuracy with conventional and asymmetric screenfilm radiographs of a chest phantom Am. J. Roentgenol. 165
26973
Kimme-Smith C, Hart E M, Goldin J G, Johnson T D, Terwilliger R and Aberle D R 1996 Detection of simulated
lung nodules with computed radiography: effects of nodule size, local optical density, global object thickness,
and exposure Acad. Radiol. 3 73541
Klein C 1968 Bandgap dependence and related features of radiation ionisation energies in semiconductors J. Appl.
Phys. 39 2029
Kogutt M S, Jones J P and Perkins D D 1988 Low-dose digital computed radiography in pediatric chest imaging Am.
J. Roentgenol. 151 7759
Korn D M, Lubinsky A R and Owen J F 1986 Storage phosphor system for computed radiography: destructive
scanning Proc. SPIE 626 10819
Kortum G 1969 Reflectance Spectroscopy (New York: Springer)
Kotera N, Eguchi S, Miyahara J, Matsumoto S and Kato H 1980 Method and apparatus for recording and reproducing
a radiation image US Patent no 4 236 078
Kottamasu S R, Kuhns L R and Stringer D A 1997 Pediatric musculoskeletal computed radiography Pediatr. Radiol.
27 56375
Launders J H and Cowen A R 1995 A comparison of the threshold detail detectability of a screenfilm combination
and computed radiology under conditions relevant to high-kVp chest radiography Phys. Med. Biol. 40 13938
Leblans P, Struye L and Willems P 2001 New needle-crystalline CR detector Proc. SPIE 4320 5967
Lo J Y, Floyd C E Jr, Baker J A and Ravin C E 1994 Scatter compensation in digital chest radiography using the
posterior beam stop technique Med. Phys. 21 43543
Lubberts G 1968 Random noise produced by x-ray fluorescent screens J. Opt. Soc. Am. 58 147583
Lubinsky A R, Whiting B R and Owen J F 1986 Storage phosphor system for computed radiography: Screen optics
Proc. SPIE 626 12032
Lubinsky A R, Whiting B R and Owen J F 1987 Storage phosphor system for computed radiography: Optical effects
and detective quantum efficiency (DQE) Proc. SPIE 767 16777
Luckey G W 1975 Apparatus and method for producing images corresponding to patterns of high-energy radiation
US Patent no 3 859 527
MacMahon H, Sanada S, Doi K, Giger M, Xu X W, Yin F F, Montner S M and Carlin M 1991 Direct comparison of
conventional and computed radiography with a dual-image recording technique Radiographics 11 25968
Maidment A D A and Yaffe M J 1994 Analysis of the spatial-frequency-dependent DQE of optically coupled digital
mammography detectors Med. Phys. 21 7219
Marshall N W, Faulkner K, Busch H P, Marsh D M and Pfenning H 1994 An investigation into the radiation dose
associated with different imaging systems for chest radiology Br. J. Radiol. 67 3539
Matsuda T, Arakawa S, Koda K, Torii S and Nakajima N 1993 New technical developments in the FCR9000 Fuji
Computed Radiography Technical Review No 2 (Fuji Photo Film)
McKeever S W S 1985 Thermoluminescence of Solids (Cambridge: Cambridge University Press)
McLean D and Gray J E 1996 Scatter-to-primary ratio and absorption efficiency in screenfilm and computed
radiography systems Eur. J. Radiol. 21 2126
Topical Review R165

Miettunen R H and Korhola O A 1991 The effect of scatter reduction on the signal-to-noise ratio in computed
radiography Eur. J. Radiol. 12 16770
Murphey M D, Quale J L, Martin N L, Bramble J M, Cook L T and Dwyer S J 1992 Computed radiography in
musculoskeletal imaging: state of the art Am. J. Roentgenol. 158 1927
Nagarkar V V, Gupta T K, Miller S R, Klugerman Y, Squillante M R and Entine G 1998 Structured CsI(Tl) scintillators
for x-ray imaging applications IEEE Trans. Nucl. Sci. 45 4926
Nakano Y, Gido T, Honda S, Maezawa A, Wakamatsu H and Yanagita T 2002 Improved computed radiography
image quality from a BaFI:Eu photostimulable phosphor plate Med. Phys. 29 5927
Nakano Y, Hiraoka T, Togashi K, Nishimura K, Itoh K, Fujisawa I, Sagoh T, Minami S, Itoh H and Torizuka K 1987
Direct radiographic magnification with computed radiography Am. J. Roentgenol. 148 56973
Nakano Y, Togashi K, Nishimura K, Itoh K, Fujisawa I, Asato R, Adachi H, Itoh H and Torizuka K 1986 Stomach
and duodenum: radiographic magnification using computed radiography Radiology 160 3837
Nakazawa M, Morikawa O, Nitta M, Tsuchino H and Shimada F 1990 Effect of protective layer on resolution
properties of photostimulable phosphor detector for digital radiographic systems Proc. SPIE 1231 35063
Niklason L T, Chan H P, Cascade P N, Chang C L, Chee P W and Mathews J F 1993 Portable chest imaging:
comparison of storage phosphor digital, asymmetric screenfilm, and conventional screenfilm systems
Radiology 186 38793
Nishikawa R and Yaffe M J 1990 Model of the spatial-frequency-dependent detective quantum efficiency of phosphor
screens Med. Phys. 17 894904
Oda N, Nakata H, Murakami S, Terada K, Nakamura K and Yoshida A 1996 Optimal beam quality for chest computed
radiography Invest. Radiol. 31 12631
Ogawa E, Arakawa S, Ishida M and Kato H 1995 Quantitative analysis of imaging performance for computed
radiography systems Proc. SPIE 2432 42131
Ohara H, Honda C, Ishusaka A and Shimada F 2002 Image quality in digital phase contrast imaging using a tungsten
anode x-ray tube with a small focal spot size Proc. SPIE 4682 71323
Oestmann J W, Prokop M, Schaefer C M and Galanski M 1991 Hardware and software artifacts in storage phosphor
radiography Radiographics 11 795805
Rong X J, Shaw C C, Liu X, Lemacks M R and Thompson S K 2001 Comparison of an amorphous silicon/cesium
iodide flat-panel digital chest radiography system with screen/film and computed radiography systemsa
contrast-detail phantom study Med. Phys. 28 232835
Rowlands J A and Yorkston J 2000 Flat panel detectors for digital radiology Medical Imaging Vol 1. Physics and
Psychophysics ed J Beutel, H L Kundel and R L Van Metter (Bellingham: SPIE) pp 223328
Samei E, Seibert J A, Willis C E, Flynn M J, Mah E and Junck K L 2001 Performance evaluation of computed
radiography systems Med. Phys. 28 36171 (Summary of AAPM Task group)
Samei E and Flynn M J 2002 An experimental comparison of detector performance for computed radiography systems
Med. Phys. 29 44759
Sanada S, Doi K, Xu X-W, Yin F-F, Giger M L and MacMahon H 1991 Comparison of imaging properties of a
computed radiography system and screenfilm systems Med. Phys. 18 41420
Sandrik J M and Wagner R F 1982 Absolute measures of physical image quality: measurement and application to
radiographic magnification Med. Phys. 9 5409
Schaetzing R, Fasbender R and Kersten P 2002 New high speed scanning technique for computed radiography Proc.
SPIE 4682 51120
Schweizer S 2001 Physics and current understanding of x-ray storage phosphors Phys. Stat. Sol. 187 33593
Seibert J A 1997 Computed radiography: technology and quality assurance The Expanding Role of Medical Physics
in Diagnostic Imaging for AAPM ed G D Frey and P Sprawls (Madison: Advanced Medical Publishing)
pp 3783
Seibert J A, Boone J M and Cooper V N 2002 Determination of imaging performance of a photostimulable phosphor
system for digital mammography Proc. SPIE 4682 44756
Seibert J A, Shelton D K and Moore E H 1996 Computed radiography x-ray exposure trends Acad. Radiol. 4 3138
Shaw C C, Wang T P, Breitenstein D S and Gur D 1997 Improvement of signal-to-noise and contrast-to-noise ratios
in dual-screen computed radiography Med. Phys. 24 12931302
Shaw C C, Wang T and Gur D 1994 Effectiveness of antiscatter grids in digital radiography: a phantom study Invest.
Radiol. 29 63642
Shimada F, Kano A, Amitani K and Nakamaru N 1991 Method for converting radiographic image radiation energy
storage panel having stimulable phosphor-containing layer and alkali halide phosphor US Patent no 5 028 509
Siegman A E 1986 Lasers (Mill Valley: University Science Books) pp 66397
Solomon S L, Jost R G, Glazer H S, Sagel S S, Anderson D J and Molina P L 1991 Artifacts in computed radiography
Am. J. Roentgenol. 157 1815
R166 Topical Review

Sonoda M, Takano M, Miyahara J and Kato H 1983 Computed radiography utilizing scanning laser stimulated
luminescence Radiology 148 8338
Stevels A L N and Schrama-de Paw A D M 1974 Vapor-deposited CsI:Na layers: II. Screens for application in x-ray
imaging devices Philips Res. Rep. 29 35362
Stewart B K and Huang H K 1990 Single exposure dual-energy computed radiography Med. Phys. 17 86675
Swank R K 1973a Calculation of modulation transfer function of x-ray fluorescent screens Appl. Opt. 12 186570
Swank R K 1973b Absorption and noise in x-ray phosphors J. Appl. Phys. 44 4199203
Swee R G, Gray J E, Beabout J W, McLeod R A, Cooper K L, Bond J R and Wenger D E 1997 Screenfilm versus
computed radiography imaging of the hand: a direct comparison Am. J. Roentgenol. 168 53942
Takahashi K, Kohda K, Miyahara J, Kanemitsu Y, Amitani K and Shionoya S 1984 Mechanism of photostimulated
luminescence in BaFX:Eu 2+ (X = Cl, Br) Phosphors J. Luminescence 3132 2668
Takahashi K, Miyahara J and Shibahara Y 1985 Photostimulated luminescence (PSL) and color centers in BaFX:Eu2+
(X = Cl, Br, I) phosphors J. Electrochem. Soc. 132 14924
Takeo H, Shimura K, Fuseda Y, Kakajima K and Hishinuma K 1996 A new FCR image processing function: energy
subtraction Fuji Computed Radiography Technical Review No 4 (Fuji Photo Film)
Tanaka K, Kato H and Matsumoto S 1984 Radiation image readout device US Patent no 4 485 302
Thoms M, von Seggern H and Winnacker A 1991 Spatial correlation and photostimulability of defect centres in the
x-ray storage phosphor BaFBr:Eu2+ Phys. Rev. B 44 92407
Trauernicht D P and Van Metter R 1988 The measurement of conversion noise in x-ray intensifying screens Proc.
SPIE 914 1006
Tucker D M, Barnes G T and Chakraborty D P 1990 A semi-empirical model for generating tungsten target x-ray
spectra Med. Phys. 18 2118
Tucker D M, Souto M and Barnes G T 1993 Scatter in computed radiography Radiology 188 2714
von Seggern H, Voigt T, Knupfer W and Lange G 1988 Physical model of photostimulated luminescence of x-ray
irradiated BaFBr:Eu2+ J. Appl. Phys. 64 140512
Vosburgh K G, Swank R K and Houston J M 1977 X-ray image intensifiers Adv. Electron. Electron Phys. 43 20544
Vuylsteke P, Dewaele P and Schoeters E 1997 Optimizing computed radiography imaging performance The Expanding
Role of Medical Physics in Diagnostic Imaging ed G D Frey and P Sprawls (Madison: Advanced Medical
Publishing) pp 10751
Wilson A J, Mann F A, Murphy V A, Monsees B S and Linn M R 1991 Photostimulable phosphor digital radiography
of the extremities: Diagnostic accuracy compared with conventional radiography Am. J. Roentgenol. 157 5338
Wilson A J, Mann F A, West O C, McEnery K W and Murphy V A 1994 Evaluation of injured cervical spine:
comparison of conventional and storage phosphor radiography with a hybrid cassette Radiology 193 41922
Wilson A J and West O C 1993 Single-exposure conventional and computed radiography: The hybrid cassette revisited
Invest. Radiol. 28 40912
Workman A and Cowen A R 1995 Improved image quality utilizing dual plate computed radiography Br. J. Radiol.
68 1828
Workman A, Cowen A R and Brettle D S 1994 Physical evaluation of computed radiography as a mammographic
X-ray imaging system Br. J. Radiol. 67 98896
Yaffe M J and Rowlands J A 1997 X-ray detectors for digital radiology Phys. Med. Biol. 42 139
Yip K L, Whiting B R, Kocher T E, Trauernicht D P and Van Metter R L 1996 Understanding the relative sensitivity
of radiographic screens to scattered radiation Med. Phys. 23 172737
Yoshimura H, Xu X W, Doi K, MacMahon H, Hoffmann K R, Giger M L and Montner S M 1993 Development of a
high quality film duplication system using a laser digitizer: comparison with computed radiography Med. Phys.
20 518