Yaffe 1997

Home Search Collections Journals About Contact us My IOPscience

X-ray detectors for digital radiography

This content has been downloaded from IOPscience. Please scroll down to see the full text.

1997 Phys. Med. Biol. 42 1
(http://iopscience.iop.org/0031-9155/42/1/001)
View the table of contents for this issue, or go to the journal homepage for more
Download details:
IP Address: 93.180.53.211
This content was downloaded on 05/02/2014 at 06:51
Please note that terms and conditions apply.

Phys. Med. Biol. 42 (1997) 1–39. Printed in the UK PII: S0031-9155(97)36090-4
REVIEW
X-ray detectors for digital radiography
M J Yaffe and J A Rowlands

Imaging Research Program, Sunnybrook Health Science Centre, The University of Toronto,
2075 Bayview Avenue, Toronto, Ontario, Canada M4N 3M5
Received 29 March 1996, in final form 16 August 1996
Abstract. Digital radiography offers the potential of improved image quality as well as
providing opportunities for advances in medical image management, computer-aided diagnosis
and teleradiology. Image quality is intimately linked to the precise and accurate acquisition
of information from the x-ray beam transmitted by the patient, i.e. to the performance of the
x-ray detector. Detectors for digital radiography must meet the needs of the specific radiological
procedure where they will be used. Key parameters are spatial resolution, uniformity of response,
contrast sensitivity, dynamic range, acquisition speed and frame rate. The underlying physical
considerations defining the performance of x-ray detectors for radiography will be reviewed.
Some of the more promising existing and experimental detector technologies which may be
suitable for digital radiography will be considered. Devices that can be employed in full-
area detectors and also those more appropriate for scanning x-ray systems will be discussed.
These include various approaches based on phosphor x-ray converters, where light quanta are
produced as an intermediate stage, as well as direct x-ray-to-charge conversion materials such
as zinc cadmium telluride, amorphous selenium and crystalline silicon.
1. Introduction
The benefits of acquisition of medical radiological images in digital form quickly became
obvious following the introduction of computed tomography (CT) by Hounsfield (1973).
These benefits include greater precision of recording the information, increased flexibility
of display characteristics and ease of transmitting images from one location to another over
communications networks.
Computed tomography is a rather sophisticated application of digital radiography, and
more recently, digital approaches to simpler, more mainstream imaging techniques such as
angiography and conventional projection radiography as well as to ultrasound and nuclear
medicine imaging have been developed. Part of the reason for this chronology was that CT
was immediately accepted because of the obvious benefits of true transverse tomography and
the ability of CT to display subtle differences in tissue attenuation. These outweighed the
desire for high spatial resolution which could not be achieved with the coarse detectors and
limited computer capacity available at the time, but which could be obtained with standard
radiographic projection imaging.
The development of improved detector technologies, as well as much more powerful
computers, high-resolution digital displays and laser output devices was necessary before
digital radiography could progress further. Initially, it was thought that digital radiography
would have to match the very demanding limiting spatial resolution performance of film-
based imaging. However, film imaging is often limited by a lack of exposure latitude due to
the film’s characteristic curve and by noise associated with film granularity and inefficient
use of the incident radiation. Further experience has suggested that a high value of limiting
0031-9155/97/010001+39$19.50
c 1997 IOP Publishing Ltd 1
2 M J Yaffe and J A Rowlands
resolution is not as important as the ability to provide excellent image contrast over a
wide latitude of x-ray exposures for all spatial frequencies up to a more modest limiting
resolution (Yaffe 1994). A digital radiographic system can provide such performance, as well
as allowing the implementation of computer image processing techniques, digital archiving
and transmission of images and extraction of medically useful quantitative information from
the images.
Historically, there has been a strong interest in developing digital imaging systems for
chest radiography because of the inherent weaknesses of film-screen systems in providing
adequate latitude and simultaneously good contrast over the lung and mediastinal regions
and the desire to implement features such as image processing, teleradiology and digital
archiving and retrieval systems (PACS). Tesic et al (1983) described a single-line-scanning
digital system for chest radiography which used an array of 1024 discrete photodiodes
coupled to a gadolinium oxysulphide phosphor. This required a scan time of 4.5 s and
provided a limiting spatial resolution of 1 cycle/mm. Goodman et al (1988) and Fraser
et al (1989) reviewed the strengths and weaknesses of various approaches to digital chest
radiography available at that time. They identified the potential for digital chest radiography
while indicating improvements that would be necessary for the technique to become accepted
by radiologists.
Digital systems for subtraction angiography and for some types of projection radiography
are now in widespread clinical use and specialized systems for demanding applications
such as mammography are currently under development. The availability of such digital
systems will potentially permit the introduction of computer-aided diagnosis (Chan et al
1987, Giger et al 1990). There have been several previous reviews of digital imaging
detector technology, notably by Rougeot (1993).
2. Digital images
Virtually all x-ray images are based on transmission of quanta through the body, with
contrast occurring due to variations in thickness and composition of the internal anatomy.
The x-ray transmission pattern in the plane of the imaging system can be considered as a
continuous variation of x-ray fluence with position. A hypothetical pattern is shown in one
dimension in figure 1(a). An analogue imaging detector attempts to reproduce this pattern
faithfully, for example as variations of optical density on a developed film emulsion. In
principle, these variations are spatially continuous and, provided that enough x-ray quanta
are used, they are also continuous on the intensity scale.
A schematic diagram of a generic digital radiography system is given in figure 2.
Here, the analogue image receptor is replaced by a detector that converts energy in the
transmitted x-ray beam into an electronic signal which is then digitized and recorded in
computer memory. The image can then be processed, displayed, transmitted or archived
using standard computer and digital communication methods.
In a digital imaging system, at some stage, the x-ray transmission pattern is sampled both
in the spatial and intensity dimensions, as illustrated in figure 1(b). In the spatial dimension,
samples are obtained as averages of the intensity over picture elements or pixels. These
are generally square areas, which are spaced at equal intervals throughout the plane of the
image. In the intensity dimension, the signal is binned into one of a finite number of levels.
This is normally a power of 2 and the value, n, of this power is designated as the number of
bits to which the image is digitized. Intensity values of the digital image can, therefore, only
take on discrete values, and information regarding intermediate intensities and variations on
a subpixel scale is lost in the digitization.
X-ray detectors for digital radiography 3
Figure 1. Concepts of digital imaging. (a) Profile of an analogue image varies continuously,
both spatially and in signal intensity. (b) In a digital image, sampling takes place at discrete
intervals in position and intensity.
Figure 2. Schematic diagram of a digital radiography system.
To avoid degradation of image quality in the digitization process, it is important that

the pixel size and the bit depth are appropriate for the requirements of the imaging task and
are consistent with the intrinsic spatial resolution and precision of the image as determined
by such fundamental limiting factors as the focal spot unsharpness, anatomical motion and
the level of quantum noise.
3. Detector properties
Important detector properties are: field coverage, geometrical characteristics, quantum

efficiency, sensitivity, spatial resolution, noise characteristics, dynamic range, uniformity,
acquisition speed, frame rate and cost. In most, if not all, cases different detector
technologies necessitate compromises among these factors.
3.1. Field coverage

The imaging system must be able to record the transmitted x-ray signal over the projected
area of the anatomy under investigation. One can estimate the requirements of digital
radiology detectors from the image receptors used for conventional imaging. For example,
chest radiography requires an imaging field of 35 cm × 43 cm, while mammography can
be accommodated by a receptor of dimensions 18 cm × 24 cm or 24 cm × 30 cm. Image
intensifiers used for fluoroscopy and film photofluorography provide circular fields with
diameters ranging from 15 cm to 40 cm. In addition, because of x-ray beam divergence, the
image always undergoes some degree of radiographic magnification. Often, this is only on
the order of 10%; however, for examinations where magnification is intentionally applied,
this can be a factor of 2 or more and, therefore, the clinical use must be carefully considered
when specifying detector size requirements.
3.2. Geometrical characteristics

Some of the factors to be considered here are the ‘dead regions’ that may exist within and
around the edges of the detector. In an electronic detector used for digital radiography,
these might be required for routing of wire leads or placement of auxiliary detector
components such as buffers, clocks, etc. Dead regions can also result when a large-area
detector is produced by abutting smaller detector units (tiling). For detectors composed of
discrete sensing elements, we can define the fill-factor as the fraction of the area of each
detector element that is sensitive to the incident x-rays. In some applications (for example
mammography) it is important that the detector has negligible inactive area on one or more
edges to avoid excluding tissue from the image. This may preclude the use of detectors
with bulky housings, such as vacuum image intensifiers, from those applications. In any
case, dead area within the detector results in inefficient use of the radiation transmitted by
the patient unless prepatient collimation can be used to mask the radiation that would fall
on these dead areas. Usually, because of alignment complexity and focal spot penumbra,
this is not practical.
Another geometrical factor which must be considered is distortion. A high-quality
imaging system will present a faithful spatial mapping of the input x-ray pattern to the
image output. The image may be scaled spatially; however, the scaling factor should be
constant over the image field. Distortion will cause this mapping to become nonlinear. It
may become spatially or angularly dependent. This may be the case when lens, fibre or
electron optics are used in the imaging system and give rise to ‘pincushion’ or ‘barrel’
distortion.
Finally, it should be noted that digital detectors can be of two general types, captive
sensors or replaceable cassettes. In the former, the receptor and its readout are integrated
into the x-ray machine. While this requires a specially designed machine with higher capital
cost, it also eliminates the need for loading, unloading and carrying of cassettes to a separate
reader and the labour costs involved. As well, the use of a single or a limited number of
receptors simplifies the task of correction for non-uniformities of the receptors (see below).
A reusable cassette system may be advantageous where a high degree of portability or
flexibility is required, such as in intensive care situations or operating theatres, and has the
advantage of being compatible with existing radiographic units.
3.3. Quantum efficiency

The initial image acquisition operation is identical in all x-ray detectors. In order to produce
a signal, the x-ray quanta must interact with the detector material. The probability of
interaction or quantum efficiency for quanta of energy E = hν is given by
η = 1 − e−µ(E)T (1)
where µ is the linear attenuation coefficient of the detector material and T is the
active thickness of the detector. Because virtually all x-ray sources for radiography are
polyenergetic, and, therefore, emit x-rays over a spectrum of energies, the quantum efficiency
must either be specified at each energy or must be expressed as an ‘effective’ value over
the spectrum of x-rays incident on the detector. This spectrum will be influenced by the
filtering effect of the patient which is to ‘harden’ the beam, i.e. to make it more energetic
and, hence, more penetrating.
The quantum efficiency can be increased by making the detector thicker or by using
materials which have higher values of µ because of increased atomic number or density. The
quantum efficiency versus x-ray energy for various thicknesses of some detector materials is
plotted in figures 3 and 4. The quantum efficiency will in general be highest at low energies,
gradually decreasing with increasing energy. If the material has an atomic absorption edge
in the energy region of interest, then quantum efficiency increases dramatically above this
energy, causing a local minimum in η for energies immediately below the absorption edge.
At diagnostic x-ray energies, the main interaction process is the photoelectric effect
because of the relatively high atomic number of most detector materials. Interaction of an
x-ray quantum with the detector generates a high-speed photoelectron. In its subsequent loss
of kinetic energy in the detector, excitation and ionization occur, producing the secondary
signal (optical quanta or electronic charge).
3.4. Spatial resolution

Spatial resolution in radiography is determined both by the detector characteristics and by
factors unrelated to the receptor. The second category includes unsharpness arising from
geometrical factors. Examples are: ‘penumbra’ due to the effective size of the x-ray source
and the magnification between the anatomical structure of interest and the plane of the image
receptor or relative motion between the x-ray source, patient and image receptor during the
exposure. Detector-related factors arise from its effective aperture size, spatial sampling
interval between measurements and any lateral signal spreading effects within the detector
or readout.
Detectors for digital radiography are often composed of discrete elements, generally of
constant size and spacing. The dimension of the active portion of each detector element
defines an aperture. The aperture determines the spatial frequency response of the detector.
For example, if the aperture is square with dimension, d, then the modulation transfer
function (MTF) of the detector will be of the form sinc f , where f is the spatial frequency
along the x or y directions, and the MTF will have its first zero at the frequency f = d −1 ,
expressed in the plane of the detector (figure 5). A detector with d = 50 µm will have an
Figure 3. Quantum interaction efficiency, η, of various thicknesses of selected phosphors. Note

that, except for CsI, the phosphor particles are combined with a binder, causing the packing
density to be reduced (typically to 50%), so that the screen thickness must be increased to
provide the attenuation values shown.
MTF with its first zero at f = 20 cycles/mm. Because of magnification, this frequency will
be higher in a plane within the patient.
Also of considerable importance is the sampling interval, p, of the detector, i.e. the pitch
in the detector plane between sensitive elements or measurements. The sampling theorem
states that only spatial frequencies in the pattern below (2p)−1 (the Nyquist frequency)
can be faithfully imaged. If the pattern contains higher frequencies, then a phenomenon
known as aliasing occurs wherein the frequency spectrum of the image pattern beyond
the Nyquist frequency is mirrored or folded about that frequency in accordion fashion and
added to the spectrum of lower frequencies, increasing the apparent spectral content of the
image at these lower frequencies (Bendat and Piersol 1986). In a detector composed of
discrete elements, the smallest sampling interval in a single image acquisition is p = d,
so that the Nyquist frequency is (2d)−1 while the aperture response falls to 0 at twice
that frequency (higher if the dimension of the sensitive region of the detector element is
smaller than d, for example because the fill-factor of the detector element is less than
1.0).
Figure 4. Quantum interaction efficiency, η, of selected direct conversion detector materials.
Aliasing can be avoided by ‘band limiting’ the image, i.e. attenuating the higher
frequencies such that there is no appreciable image content beyond the Nyquist frequency.
The blurring associated with the focal spot may serve this purpose. Note that this does
not prevent the aliasing of high spatial frequency noise. Alternative methods which reduce
aliasing effects of both signal and noise require the sampling frequency of the imaging
system to be increased. One method for achieving this, known as dithering, involves
multiple acquisitions with a physical motion of the detector by a fraction of the pixel pitch
between successive acquisitions. The subimages are then combined to form the final image.
This effectively reduces p, thereby providing a higher Nyquist frequency. Some detectors
are not pixellated at the x-ray absorption stage, but rather d and p are defined in their
readout mechanism. This is the case for the photostimulable phosphor detector system
described below, where the phosphor plate is continuous, but the laser readout samples
the plate at discrete locations. This can provide some flexibility in independently setting
the sampling interval (scanning raster) and effective aperture size (laser spot size) to avoid
aliasing. Issues of sampling in digital radiographic systems have been reviewed by Dobbins
(1995).
In the overall design of an imaging system, it is important that other physical sources of
unsharpness be considered when the aperture size and sampling interval are chosen. If, for
example, the MTF is limited by unsharpness due to the focal spot, it would be of little value
to attempt to improve the system by designing the receptor with smaller detector elements.
Figure 5. Effect of a 50 µm rectangular detector aperture on MTF of the image receptor. The
Nyquist frequency, fN , above which aliasing occurs is indicated.
3.5. Noise
All images generated by quanta are statistical in nature, i.e. although the image pattern can
be predicted by the attenuation properties of the patient, it will fluctuate randomly about
the mean predicted value. The fluctuation of the x-ray intensity follows Poisson statistics,
so that the variance, σ 2 , about the mean number of x-ray quanta, N0 , falling on a detector
element of a given area, is equal to N0 . Interaction with the detector can be represented as a
binomial process with probability of success, η, and it has been shown (Barrett and Swindell
1981) that the distribution of interacting quanta is still Poisson with standard deviation
σ = (N0 η)1/2 . (2)
If the detection stage is followed by a process that provides a mean gain ḡ, then the
‘signal’ becomes
q = N0 ηḡ (3)
while the variance in the signal is
σq2 = N0 η(ḡ 2 + σg2 ). (4)
In general, the distribution of q is not Poisson even if g is Poisson distributed. Similarly,
the effect of additional stages of gain (or loss) can be expressed by propagating this
expression further (Rabbani et al 1987, Cunningham et al 1994, Yaffe and Nishikawa
1994). It is also possible that other independent sources of noise will be contributed at
different stages of the imaging system. Their effect on the variance at that stage will be
additive and the fluctuation will be subject to the gain of subsequent stages of the imaging
system.
A complete analysis of signal and noise propagation in a detector system must take into
account the spatial frequency dependence of both signal and noise. Signal transfer can be
characterized in terms of the modulation transfer function, MT F (f ), where f is the spatial
frequency, while noise is described by the noise power or Wiener spectrum W (f ). Methods
for calculating the Wiener spectral properties of a detector must correct for nonlinearities
in the detector and must properly take into account the spatial correlation of signal and
statistical fluctuation (Rabbani et al 1987, Cunningham et al 1994).
A useful quantity for characterizing the overall signal and noise performance of imaging
detectors is their spatial frequency-dependent detective quantum efficiency, DQE(f ). This
describes the efficiency in transferring the signal-to-noise ratio (squared) contained in the
incident x-ray pattern to the detector output. Ideally, DQE(f ) = η for all f , but additional
noise sources will reduce this value and often cause the DQE to decrease with increasing
spatial frequency. DQE(f ) can be treated as a sort of quantum efficiency, in that when
it is multiplied by the number of quanta incident on the detector, one obtains SNR2out (f ),
also known as the number of noise equivalent quanta, N EQ(f ), used to form the image.
Typically DQE for a screen-film detector has a value on the order of 0.2 at a spatial frequency
of 0 cycles/mm and this may fall to 0.05 at a few cycles/mm (Bunch et al 1987).
As discussed in section 4, it is important that the number of secondary quanta or electrons
at each stage of the detector is somewhat greater than N0 η, to avoid having the detector
noise being dominated by a ‘secondary quantum sink’.
Table 1. Properties of phosphors and photoconductors used as x-ray detectors for digital
radiography, including atomic number, Z, and K absorption energy, EK , of the principal
absorbing elements. Sensitivity is expressed as the energy, w, which must be absorbed to
release a quantum of light in a phosphor or an electron–hole pair in a photoconductor. The
fluorescent yield, ωK , is the probability that when a K-shell photoelectric interaction occurs,
there will be a fluorescent (characteristic) x-ray rather than an Auger electron given off.
Material Z EK (keV) w (eV) ωK (approx.)a
CdTe 48/52 26.7/31.8 4.4 0.85–0.88

High-purity Si 14 1.8 3.6 < 0.05
Amorphous selenium 34 12.7 50 (at 10 V µm−1 )b 0.6
CsI(Tl) 55/53 36.0/33.2 19 0.87
Gd2 O2 S 64 50.2 13 0.92
BaFBr (as photostim.
phosphor) 56/35 37.4/13.5 50–100c 0.86
a From Evans (1955).
b 7 eV (theoretical value at infinite field).
c Estimated by multiplying the bandgap of 8.3 eV by 3 (Klein 1968) and then 2 for the 50%
efficiency of trap filling during x-ray exposure. The higher value reflects a possible additional
loss of up to a factor of 2 is due to retrapping during readout.
3.6. Sensitivity
The final output from virtually all x-ray detectors is an electrical signal, so that sensitivity can
be defined in terms of the charge produced by the detector (before any external amplification)
per incident x-ray quantum of a specified energy. The sensitivity of any imaging system
depends on η and on the primary conversion efficiency (the efficiency of converting the
energy of the interacting x-ray to a more easily measurable form such as optical quanta
or electric charge). Conversion efficiency can be expressed in terms of the energy, w,
necessary to release a light photon in a phosphor, an electron–hole pair in a photoconductor
(or semiconductor) or an electron–ion pair in a gaseous detector. Values of w for some
typical detector materials are given in table 1. The limiting factor is related to the intrinsic
Figure 6. Energy level diagrams for crystals used in (a) direct conversion x-ray detectors,
(b) conventional phosphors, (c) photostimulable phosphor imaging.
band structure of the solid from which the detector is made. In figure 6(a) the basic band
structure of a crystalline material is shown. Normally the valence band is fully populated
with electrons and the conduction band is empty. The energy gap governs the scale of energy
necessary to release an electron–hole pair, i.e. to promote an electron from the valence band
to the conduction band. However, though this energy is the minimum permitted by the
principle of conservation of energy, this can be accomplished only for photons of energy
exactly equal to the energy gap. For charged particles releasing energy (e.g. through the
slowing down of high-energy electrons created by an initial x-ray interaction), requirements
of conserving both energy and crystal momentum as well as the presence of competing
energy loss processes require, on average, at least three times as much energy as the bandgap
to release an electron–hole pair (Klein 1968). In figure 6(b) the situation for a phosphor is
shown. In this case, the first requirement is to obtain an electron–hole pair. Subsequently,
the electron returns to the valence band via a luminescence centre created by an activator
added to the host material. This requires that the energy EF of the fluorescence light must
be less than the bandgap energy EG and therefore there are further inevitable inefficiencies
in a phosphor compared to a photoconductor of the same EG .
3.7. Dynamic range

The dynamic range can be defined as:
Xmax
DR = (5)
Xnoise
where Xmax is the x-ray fluence providing the maximum signal that the detector can
accommodate and Xnoise is the fluence that provides a signal equivalent to the quadrature
sum of the detector noise and the x-ray quantum noise.
While this definition describes the performance of the detector on an individual pixel
basis, it is less useful for predicting the useful range of detector operation for a particular
imaging task. This is because at the bottom of this range the signal-to-noise ratio (SNR) is
only 1 and this is seldom acceptable. Also, it is rare to base a medical diagnosis on a single
image pixel and therefore, for most objects, the SNR is based on the signal from multiple
pixels. For a large object, the noise on a pixel by pixel basis can be large, but if there
is integration over the object, the effective SNR will improve approximately as the square
root of the area (with some correction for correlation effects caused by unsharpness of the
imaging system). We have, therefore, offered a second definition of ‘effective dynamic
range’ which we have found useful
k2 Xmax
DReff = . (6)
k1 Xnoise
Here, the constant k1 is the factor by which the minimum signal must exceed the noise for
reliable detection. Rose (1948) has argued that k1 should be on the order of 4 or 5 depending
on the imaging task. The constant k2 , which is dependent on the imaging task and the system
MTF, reflects the improvement in SNR due to integration over multiple pixels. Effectively,
this causes the dynamic range of the imaging system to increase even though the maximum
signal level and the single pixel noise level have not changed. Maidment et al (1993) and
Neitzel (1994) have analysed this problem for the case of digital mammography.
In practice, the required dynamic range for an imaging task can be decomposed into
two components. The first describes the ratio between the x-ray attenuation of the most
radiolucent and most radio-opaque paths through the patient to be included on the same
image. The second is the precision of x-ray signal to be measured in the part of the image
representing the most radio-opaque anatomy. If, for example, there was a factor of 50 in
attenuation across the image field and it was desired to have 1% precision in measuring the
signal in the most attenuating region, then the dynamic range requirement would be 5000.
The dynamic range requirements for certain applications may exceed the capabilities of
available detectors. It is often possible to reduce the requirement by employing prepatient
bolusing filters to increase attenuation in lucent areas of the image and so reduce the range
of intensities that must be accommodated.
The dynamic range requirements differ between imaging tasks, but some general
principles for establishing the requirements of each modality can be put forward. First
it is important to recognize that the x-rays are attenuated exponentially, thus an extra tenth-
value layer thickness of tissue will attenuate the beam by an additional factor of 10, while the
same tenth-value thickness missing will increase the x-ray fluence by a factor of 10. Thus
when a mean exposure value Xmean for the system is established by irradiating a uniform
phantom, we are interested in multiplicative factors above and below this mean value, i.e.
Xmean is a geometric rather than arithmetic mean. Thus, for example in fluoroscopy, it is
generally established that a range of 100:1 is useful but it is also essential to understand
this range should be between Xmean /10 and 10Xmean rather than distributed in equal linear
increments, i.e. between Xmean /50 and 2Xmean .
In defining the range of operation for a detector, one must consider both the need for
adequate x-ray fluence to achieve the desired quantum counting statistics at the low end of
the range as well as detector phenomena such as saturation or ‘blooming’ that can occur
with large signals.
3.8. Uniformity
It is important that the radiographic imaging system provide uniformity, i.e. the sensitivity
be constant over the entire area of the image. Otherwise patterns that might disrupt the
effective interpretation of the image may result. These patterns are sometimes referred to
as ‘fixed pattern noise’. In an analogue imaging system, great pains must be taken in the
design and manufacture of detectors to ensure that they provide uniform response.
In a digital system, the task is much easier, because, at least over a considerable range,
differences in response from element to element can be corrected. This is accomplished
by imaging an object of uniform x-ray transmission, recording the detector response and
using this as a ‘correction mask’. If the detector has linear response to x-rays, then the
correction involves two masks—one with and one without radiation—to provide slope and
intercept values for the correction of each element. If the detector response is nonlinear,
then measurements must be made over a range of intensities and a nonlinear function fit
to the response of each element to obtain the correction coefficients. In some detectors,
non-uniformities might exist only over rows and columns of the detector rather than over
individual elements. This greatly reduces the number of coefficients that must be stored.
4. Phosphor-based detector systems
Most x-ray imaging detectors employ a phosphor in the initial stage (figure 7(a)) to absorb
the x-rays and produce light which is then coupled to an optical sensor (photodetector). The
use of phosphor materials with a relatively high atomic number causes the photoelectric
effect to be the dominant type of x-ray interaction. The photoelectron produced in these
interactions is given a substantial fraction of the energy of the x-ray. This energy is much
larger than the bandgap of the crystal (figure 6(b)) and, therefore, in being stopped, a single
interacting x-ray has the potential to cause the excitation of many electrons in the phosphor
and thereby the production of many light quanta. We describe this ‘quantum amplification’
as the conversion gain, g1 . For example, in a Gd2 O2 S phosphor, the energy carried by a
60 keV x-ray quantum is equivalent to that of 25 000 green light quanta ( Eg = 2.4 eV).
Because of competing energy loss processes and the need to conserve momentum, the
conversion efficiency is only about 15%, so that, on average, it requires approximately
13 eV per light quantum created in this phosphor (table 1). The conversion gain is then
approximately 4500 light quanta per interacting x-ray quantum.
The energy loss process is stochastic and, therefore, g has a probability distribution,
with standard deviation, σg , about its mean value as illustrated in figure 8(a). Swank (1973)
described this effect, and the ‘Swank factor’, As , characterizes this additional noise source.
The Swank factor is calculated in terms of the moments of the distribution of g as
M12
As = (7)
M0 M2
where Mi indicates the ith moment of the distribution.
The actual number of quanta produced by an interacting x-ray will also depend both
on its incident energy and the mechanism of interaction with the phosphor crystal. The
most likely type of interaction, the photoelectric effect, will result in both an energetic
photoelectron and either a second (Auger) electron or a fluorescent x-ray quantum. The
energy of fluorescence depends on the shell in which the photoelectric interaction took place.
The threshold K-shell energy for these interactions is shown for some common radiographic
phosphors in table 1. Also in the table is the K-fluorescence yield; the probability of
emission of x-ray fluorescence, given that a K-shell photoelectric interaction has occurred.
For example, K-shell interactions for the Gd in Gd2 O2 S have a threshold of 50.2 keV and
produce the most intense fluorescence (92% of K-shell interactions yield these quanta) just
below 43 keV. The fluorescent quanta are either reabsorbed in the phosphor or escape. In
either case, if they are not absorbed locally, the apparent energy deposited in the phosphor
from the x-ray quantum is reduced, giving rise to a second peak in the distribution with a
lower value of g. The effect of fluorescence loss is to broaden the overall distribution of g
(figure 8(b)), thus decreasing As and causing an increase in σg .
Figure 7. Three types of detector structures: (a) settled phosphor, (b) columnar CsI phosphor
and (c) direct x-ray convertor with charge collection in an electric field. A hypothetical line-
spread function of each system is shown.
There are both advantages and disadvantages in imaging with an x-ray spectrum that
exceeds the K edge of the phosphor. Clearly, the value of η increases, but the ‘Swank
noise’ does also. In addition, deposition of energy from the fluorescence at some distance
from the point of initial x-ray interaction causes the point spread function of the detector to
increase, resulting in decreased spatial resolution.
After their formation, the light quanta must successfully escape the phosphor and be
effectively coupled to the next stage for conversion to an electronic signal and readout. It
is desirable to ensure that the created light quanta escape the phosphor efficiently and as
near as possible to their point of formation.
Figure 9 illustrates the effect of phosphor thickness and the depth of x-ray interaction
on spatial resolution of a phosphor detector. The probability of x-ray interaction is
exponential so that the number of interacting quanta and the amount of light created will
be proportionally greater near the x-ray entrance surface.
While travelling within the phosphor, the light will spread—the amount of diffusion
being proportional to the path length required to escape the phosphor. The paths of most
optical quanta will be shortest if the photodetector is placed on the x-ray entrance side of
the phosphor. It is usually more practical, however, to record the photons which exit on
the opposite face of the phosphor screen, i.e. those which have had a greater opportunity to
spread. In addition, if a phosphor layer is made thicker to improve quantum efficiency, the
spreading becomes more severe. This imposes a fundamental compromise between spatial
resolution and η. Methods to collect the emission from the entrance side of the phosphor
Figure 8. Effect of fluorescence loss on the distribution of g1 for monoenergetic x-rays. The
value of g corresponding to conversion of x-rays at the K absorption edge energy is shown
on the abscissa. (a) For x-rays of energy below the edge, there is a single distribution of the
number of light quanta about the mean value, g¯1 while for energies above the absorption edge
(b) there is a bimodal distribution, where the upper peak corresponds to total absorption of the
incident x-ray, while the lower peak corresponds to conversion of the energy of the incident
x-ray minus the energy of the fluorescent x-ray that has escaped the phosphor.
Figure 9. Phosphor thickness, depth of x-ray interaction and line spread function: (a) thin
screen, (b) increased line spread function for a thicker screen. In either case, spatial resolution
would be improved if it were possible to measure the signal from the face of the detector on
which the x-rays are incident.
or to channel the optical photons out of the phosphor without spreading would significantly
improve phosphor performance.
Phosphor screens are typically produced by combining 5–10 µm diameter phosphor
particles with a transparent plastic binder (figure 7(a)). The phosphor grains are highly
scattering particles due to the high refractive index of phosphors compared with that of
the binder. The scattering is sufficiently intense to cause the layer to be turbid: i.e. the
Figure 10. Number of quanta or charges at different stages in an imaging system: Full curve,
x-ray limited; broken curve, inadequate conversion and/or coupling yields a secondary quantum
sink.
propagation of photons can be considered to be diffusive. This results in a limit to the

lateral spreading of the light to the order of the thickness of the layer. Other optical effects
can also be used to control the imaging properties of the screen—for example a reflective
backing helps to increase the amount of light escaping the opposite side of the screen, but
at the cost of increased lateral spread, and hence reduced resolution. Typically, without the
backing, fewer than one-half of the created light quanta escape the phosphor on the side
facing the photodetector and are potentially available to be recorded. Light-absorbing dye
can also be added to the screen to enhance the resolution, but this results in a loss of signal.
These optical techniques affect sensitivity, spatial resolution and (through their influence on
the Swank factor (Drangova and Rowlands 1986)) the noise properties of the detector.
It should also be noted that the packing factor of phosphor particles in the screen may
be of the order of 50% by volume. When calculations of η are made, the reduction of
effective attenuation coefficient due to the binder must be considered.
Figure 10 illustrates the propagation of signal through the various energy conversion
stages of an imaging system. In the diagram, N0 quanta are incident on a specified area
of the detector surface (stage 0). A fraction of these, given by the quantum detection
efficiency, η, interact with the detector (stage I). In a perfect imaging system η would be
equal to 1.0. The mean number, N1 , of quanta interacting represents the ‘primary quantum
sink’ of the detector. The fluctuation about N1 is σN1 = (N1 )1/2 . This defines the SNR of
the imaging system which increases as the square root of the number of quanta interacting
with the detector.
Regardless of the value of η, the maximum SNR of the imaging system will occur at
this point and if the SNR of the imaging system is essentially determined there, the system
is said to be x-ray quantum limited in its performance. However, the SNR will, in general,
become reduced in passage of the signal through the imaging system because of losses and
additional sources of fluctuation.
To avoid losses that can occur at subsequent stages, it is important that the detector
Figure 11. Effect of optical coupling efficiency on DQE(f ) of a phosphor–fibre optic–CCD

detector. CD is the number of electrons produced in the CCD per x-ray interacting in the
phosphor. (From Maidmont and Yaffe 1994.)
provide adequate quantum gain, g1 directly following the initial x-ray interaction. Stages II
and III illustrate the processes of creation of many light photons from a single interacting
x-ray (often referred to as conversion gain) and the escape of quanta from the phosphor
with mean probability g2 . Here, light absorption, scattering and reflection processes are
important.
Further losses occur in the coupling of the light to the photodetector which converts
light to electronic charge (stage IV) and in the spectral sensitivity and optical quantum
efficiency of the photodetector (stage V). If the conversion gain of the phosphor is not
sufficiently high to overcome these losses and the number of light quanta or electronic
charges at a subsequent stage falls below that at the primary quantum sink, then a ‘secondary
quantum sink’ is formed. In this case the statistical fluctuation of the light or charge at this
point becomes an additional important noise source. Even when an actual secondary sink
does not exist, a low value of light or charge will cause increased noise. This becomes
especially important when a spatial-frequency-dependent analysis of SNR is carried out and,
as discussed earlier, its effect is to cause reduction of the detective quantum efficiency with
increasing spatial frequency. Figure 11 illustrates the effect of optical coupling efficiency
of light from a phosphor to a photodetector on DQE(f ) for an optically coupled system
(Maidment and Yaffe 1994).
As illustrated in figure 12, there are several approaches for coupling a phosphor to
a photodetector. Possibly the simplest involves the use of a lens and/or mirror system
(figure 12(a)) to collect light emitted from the surface of the phosphor material and couple
it to either a conventional video camera (see figure 18) or to a CCD camera. The operation
of these cameras is discussed briefly later in this article.
Because the size of available photodetectors such as CCDs is limited from manufacturing
considerations to a maximum dimension of only 2–5 cm, it is often necessary to demagnify
Figure 12. Methods to couple a phosphor to a photodetector: (a) lens, (b) fibre optic, (c) direct
coupling to a photocathode whose emission is electrostatically collected.
the image from the phosphor to allow coverage of the required field size in the patient
(Karellas et al 1992). The efficiency of lens coupling is determined largely by the solid
angle subtended by the collecting optics. For a single lens system, the coupling efficiency
is given by (Miller 1971, Maidment and Yaffe 1996)
τ
ξ= (8)
4F 2 (m + 1)2
where τ is the optical transmission factor for the lens, F is the ‘f -number’ of the lens (ratio
of the focal length to its limiting aperture diameter) and m is the demagnification factor
from the phosphor to the photodetector. For a lens with F = 1.2, τ = 0.8 and m = 10, ξ
will be 0.1%. Because of this low efficiency, the SNR of systems employing lens coupling
is often limited by a secondary quantum sink, especially in such applications as radiotherapy
portal imaging where the demagnification factor is large (Munro et al 1990). On the other
hand, lenses are successfully used for coupling in situations where m is small and g1 is
large (Roehrig et al 1994).
It is also possible to use fibre optics to effect the coupling. These can be in the form of
fibre optic bundles (figure 12(b)), where optical fibres of constant diameter are fused to form
a light guide. The fibres form an orderly array so that there is a one-to-one correspondence
Figure 13. Comparison of the efficiency of lens and fibre optic coupling between a phosphor
and CCD. The f number of the lens is 1.2. (Modified from Hejazi and Trauernicht 1996.)
between the elements of the optical image at the exit of the phosphor and at the entrance
to the photodetector. To accomplish the required demagnification, the fibre optic bundle
can be tapered by drawing it under heat. While facilitating the construction of a detector to
cover the required anatomy in the patient, demagnification by tapering also reduces coupling
efficiency by limiting the acceptance angle at the fibre optic input. A simplified expression
for the coupling efficiency of a fibre optic taper is
NA2
ξ = ατ (θ) (9)
m2
where α is the fraction of the entrance surface that comprises the core glass of the optical
fibres, τ (θ ) is the transmission factor for the core glass, N A is the numerical aperture of
the untapered fibre and m is the demagnification factor due to tapering. For example, a
taper with 10 times demagnification (m = 10), with α = 0.8, τ = 0.9 and N A = 1.0, has
an efficiency of 0.7%, about seven times higher than a lens with F = 1.2 with the same
demagnification factor and about 2.5 times higher than a lens with F = 0.7. It should be
noted that for both lenses and fibre optics, the transmission efficiency is dependent on the
angle of incidence, θ of the light and, therefore, a complete analysis involves an integral of
the angular distribution of emission of the phosphor over θ . A comparison of the efficiency
of lens versus fibre optic coupling is shown in figure 13 (Hejazi and Trauernicht 1996).
Systems of both designs are used in cameras with a small field of view for digital
mammography, both for guiding needle biopsy and for localization of suspicious lesions.
In such applications, much lower demagnification, typically two times, is used, resulting in
acceptable coupling efficiency. By abutting several camera systems to form a larger matrix,
a full-field digital breast imaging system can be constructed (Feig and Yaffe 1995), while
still maintaining a low value of m and thus an adequate value of g1 to avoid a secondary
quantum sink.
Fibre optic bundles are subject to geometric distortion which must be minimized. To
maintain high resolution, the crosstalk of signal between fibres must be controlled and this is
accomplished, in part, by the use of extramural absorber (EMA), i.e. an optically attenuating
Figure 14. Phosphor–fibre optic–CCD detector assembly for slot-scanned digital mammography.
Detector input dimensions are 24 cm × 3.2 mm.
material incorporated between individual fibres in the bundle to absorb light that escapes
from the fibres or that directly enters the fibre cladding material on the entrance surface of
the bundle.
5. Scanned-beam acquisition
One way of overcoming the size and cost limitations of available high-resolution
photodetectors in producing a large imaging field is to create an image receptor that is
essentially one dimensional and acquires the second dimension of the image by scanning
the x-ray beam and detector across the patient. In principle this could be done by employing
a single line detector and a very highly collimated slit beam of x-rays. This is extremely
inefficient because of the poor utilization of the output of the x-ray tube. Most of the
x-rays would be removed by the collimator and a full scan would impose an enormous heat
loading on the tube. It is possible to improve the efficiency of such systems tremendously
by employing a multiline or ‘slot’ detector. Here, the x-ray beam would extend across
the full image field in one dimension but would be narrow (e.g. 3–15 mm) in the other.
In our group, we have used such a design to construct a digital mammography system
(Nishikawa et al 1987, Yaffe 1993). This is composed of a strip of phosphor material
with dimensions 3.2 mm × 240 mm. This is coupled to three fibre optic tapers which
are abutted with mitre joints at their input surfaces as shown in figure 14. Their taper
ratio of 1.58:1 provides demagnification with acceptable light collection efficiency for this
application while providing a space between the tapers at the output to accommodate the
outer non-active regions of three CCD arrays, which are bonded directly on the tapers.
Acquisition takes place in time delay integration (TDI) mode in which the x-ray beam
is activated continuously during the image scan and charge collected in pixels of the
Figure 15. Kinestatic gaseous ionization imaging detector employing the TDI principle.
CCDs is shifted down CCD columns at a rate equal to but in the opposite direction as
the motion of the x-ray beam and detector assembly across the breast. The collected
charge packets remain essentially stationary with respect to a given projection path of
the x-rays through the breast and the charge is integrated in the CCD column to form
the resultant signal. When the charge packet has reached the final element of the CCD,
it is read out on a transfer register and digitized. The CCD array can be cooled using
thermoelectric devices to reduce noise and increase the dynamic range of the image receptor
as necessary.
Another (non-phosphor) TDI approach, called ‘kinestatic imaging’ has been
demonstrated by DiBianca and Barker (1985) and Wagenaar and Terwilliger (1995) where
the imaging detector is a gaseous ionization chamber in which the charge signal is collected
by a large number of linear electrode strips which run parallel to the direction of propagation
of the x-ray beam (figure 15). This provides one-dimensional spatial localization. By
carefully controlling the drift velocity of the ions toward the collectors in an electric
field and synchronizing this with the motion of the detector assembly in the opposite
direction, a TDI acquisition can be accomplished. The value of η for this detector is
determined by the type of gas used, its pressure, the thickness of the detector along the
direction of x-ray travel and the window attenuation. Spatial resolution is determined by
the homogeneity of the ion drift speed, ion diffusion and the spacing between the collecting
electrodes.
Figure 16. Schematic representaion of a photostimulable phosphor digital radiography system.
6. Photostimulable phosphors
Probably the most successful detectors for digital radiography to date have been
photostimulable phosphors, also known as storage phosphors. These phosphors are
commonly in the barium fluorohalide family, typically BaFBr:Eu2+ , where the atomic
energy levels of the europium activator determine the characteristics of light emission.
X-ray absorption mechanisms are identical to those of conventional phosphors. They differ
in that the useful optical signal is not derived from the light that is emitted in prompt
response to the incident radiation, but rather from subsequent emission when electrons and
holes are released from traps in the material (Takahashi et al 1984, von Seggern et al
1988). The initial x-ray interaction with the phosphor crystal causes electrons to be excited
(figure 6(c)). Some of these produce light in the phosphor in the normal manner, but the
phosphor is intentionally designed to contain traps which store the charges. By stimulating
the crystal by irradiation with red light, electrons are released from the traps and raised to the
conduction band of the crystal, subsequently triggering the emission of shorter-wavelength
(blue) light. This process is called photostimulated luminescence. Early analysis of these
systems was provided by Hillen et al (1987) and a comparison with screen–film systems was
performed by Sanada et al (1991). Recent reviews of photostimulable phosphor imaging
were presented by Kato (1994) and Bogucki et al (1995).
In the digital radiography application, the imaging plate is positioned in a light-tight
cassette or enclosure, exposed and then read by raster scanning the plate with a laser to
release the luminescence (figure 16). The emitted light is collected and detected with a
photomultiplier tube whose output signal is digitized to form the image.
The energy levels in the crystal are critical to the effective operation of the detector
(figure 6(c)). The energy difference between the traps and the conduction band ET must be
small enough so that stimulation with laser light is possible, yet sufficiently large to prevent
random thermal release of the electron from the trap. Finally, the energetics should provide
for a wavelength of the emitted light that can be efficiently detected by a photomultiplier
and for adequate wavelength separation between the stimulating and emitted light quanta to
avoid contaminating the measured signal. The electrons liberated during irradiation either
produce light promptly or are stored in traps. Because the ‘prompt’ light is not of interest
in this application, the efficiency of the storage function can be improved by increasing
the probability of electron trapping. On the other hand, when these electrons are released
by the stimulating light during readout, the probability of their being retrapped instead of
producing light, would then be higher, so that the efficiency of readout would be reduced.
The optimum balance occurs where the probabilities of an excited electron being retrapped
or stimulating fluorescence are equal. This causes the conversion efficiency to be reduced
by a factor of four compared to the same phosphor without traps, i.e. a factor of two from
the prompt light given off during x-ray exposure and another factor of two from unwanted
retrapping of the electrons during readout.
In addition, the decay characteristics of the emission must be sufficiently fast such that
the image can be read in a conveniently short time while capturing an acceptable fraction of
the emitted energy. In practice, depending on the laser intensity, the readout of a stimulable
phosphor plate yields only a fraction of the stored signal. This is a disadvantage with respect
to sensitivity and readout noise, but it can be helpful by allowing the plate to be ‘pre-read’,
i.e. read out with only a small part of the stored signal, to allow automatic optimization of
the sensitivity of the electronic circuitry for the main readout.
6.1. Strengths and limitations of stimulable phosphors
The photostimulable phosphor is an excellent detector for digital radiography in that, when
placed in a cassette, it can be used with conventional x-ray machines. Large-area plates
are conveniently produced, and because of this format, images can be acquired quickly.
The plates are reusable, have linear response over a wide range of x-ray intensities, and
are erased simply by exposure to a uniform stimulating light source to release any residual
traps.
One limitation of this type of detector is that because the traps are located throughout the
depth of the phosphor material, the laser beam providing the stimulating light must penetrate
into the phosphor. Scattering of the light within the phosphor causes release of traps over
a greater area of the image than the size of the incident laser beam. This results in loss of
spatial resolution, which is aggravated if the plate is made thicker to increase η. An ideal
solution to this problem would be a phosphor which was non-scattering for the stimulating
light and both non-scattering and non-absorbing for the emitted light. Limitations also arise
primarily from the readout stage. This is mechanically complex, and efficient collection of
the emitted light requires great attention to design. This can result in a secondary quantum
sink, especially at high spatial frequencies, causing a reduction of DQE(f ).
In practice, stimulable phosphor systems are widely used for both emergency and bedside
radiography, where the variable readout sensitivity allows for compensation for under and
overexposure problems often experienced with screen-film radiography because it is often
not possible to employ automatic exposure control in these applications. As well, this
technology has been implemented both in the form of portable cassettes which are carried
between the x-ray unit(s) and a central reader and also in dedicated units incorporating the
plates, the reader and an erasing light source.
Figure 17. Image intensifier-based digital radiography system.
7. Image intensifier based detectors
Image intensified fluoroscopy is currently the only dose-efficient x-ray procedure that allows
real-time visualization and interaction. The resulting real-time images are usually displayed
using a video system (conventional or CCD) optically coupled to the x-ray image intensifier.
Radiographs or sequences of radiographs are obtained during fluoroscopic procedures by
recording XRII output by means of a small-format (photofluorographic) or ciné camera that
receives the intensifier image via an optical splitter. More recently, instant radiography and
instant ciné have been made possible by digitization of the video signal.
An image intensifier (XRII) digital radiography system is shown schematically in
figure 17. This resembles the imaging chain used in fluoroscopy, with the exception that
the video camera is capable of integrating the signal, thus facilitating digitization of the
optical image from the XRII. The XRII (DeGroot 1994) absorbs the incident x-ray image,
amplifies it in an essentially noise-free manner and outputs it as an optical image which is
then distributed by lenses to the video camera (Rowlands 1994).
X-rays are converted to light in the large input phosphor screen typically of 12.5 cm
to 40 cm in diameter. The fluorescence illuminates a photocathode evaporated directly on
the phosphor and liberates electrons. The electrons are accelerated through a large potential
difference (typically 25 kV) and electrostatically focused by the electrodes onto a small
(2.5 cm diameter) output phosphor.
The input window is necessary to preserve the vacuum within the intensifier. It should
be as transparent to x-rays as possible to avoid loss of image-forming quanta and to minimize
scattering in the window which causes loss of contrast due to veiling glare. Initially input
windows were made of glass but they are now generally metallic—convex aluminium or
concave titanium.
The function of the input phosphor is to provide a high value of η and g1 and to convey
the light in as sharp an image as possible to the photocathode. CsI(Na) is universally used
for this purpose (DeGroot 1994). Typically, thicknesses of 300–400 µm are employed. Its
relatively high effective atomic number and high packing density lead to good quantum
detection efficiency in the diagnostic energy range. However, the unique advantage of CsI
is that it can be evaporated in such a way that it acts as a fibre optic light guide. In
figures 3 and 4 we compare η for CsI with that of screens made of various phosphors and
to photoconductor detectors (see also Vosburg et al 1977).
The purpose of the photocathode is to convert light photons to electrons efficiently. The
greatest efficiency is obtained when the spectral sensitivity of the photocathode is matched
to the phosphor spectrum. The photocathode is extremely sensitive to contamination so it
has to be made in situ, within the otherwise completed XRII after a very high vacuum has
been established. The vacuum has to be maintained continuously thereafter.
An important factor in the achieving of gain within an XRII is the 25 keV energy
each electron released from the photocathode receives from the electrostatic field before
striking the output phosphor. Gain is also achieved by electronic image demagnification.
The demagnification factor of any XRII can be changed by appropriate modification of the
relative potentials applied to each electrode. The gain can be reduced, without change of
field size, by decreasing all the applied voltages in the same proportion.
The function of the output phosphor is to convert the incident electron image to a
visible light image. It should do this with the greatest possible efficiency and with the
least blurring. The inner surface of the phosphor layer is overlaid by a thin, opaque, layer
of aluminium which helps to maintain stable electrostatic operating conditions and prevent
light from the output phosphor illuminating the photocathode layer on the input phosphor.
A major problem arising in the output phosphor is veiling glare through halation, where
the light intended to be emitted to the imaging lens instead gets caught within the glass
substrate of the phosphor and can re-enter the phosphor layer.
There are great advantages inherent in the design of the XRII which help to make the
operation of XRII-based imaging systems x-ray quantum limited. First, the intimate coupling
of the phosphor and photocathode provides much higher collection efficiency of the light
than with lens or fibre optic methods. Secondly, as shown in figure 12(c), the collection
and focusing of emitted electrons by the electrostatic field is also very efficient and these
two factors outweigh any inefficiency of the photocathode. Thirdly, the acceleration of the
electrons within the tube provides a high gain which more than compensates for subsequent
losses in the imaging system. Finally, the demagnification of the image in the tube allows
efficient lens coupling to the next stage, generally a video camera.
7.1. Video camera

The video camera is optically coupled to the output of the XRII. The characteristics of video
cameras are dominated by the characteristics of the optical sensor chosen. For example,
the vidicon used in many fluorososcopic video cameras employs a light sensitive target
made of low-density antimony trisulphide (Sb2 S3 ). Other photoconductors with different
properties are also available. The Plumbicon utilizes a PbO target, while the target of the
Saticon is composed of amorphous selenium. These cameras have reduced lag compared
to the vidicon, making them more suitable for applications where good temporal response
is important such as digital angiography. A comprehensive review of video camera tube
design and the emergence of the vidicon as the dominant type is provided by McGee (1979).
All modern video sensors store charge representing the image at each pixel continually and
over the whole active area simultaneously (i.e. the pixels sense light in parallel). Parallel
detection is essential to a sensitive camera, but a practical system is usually read out in
pixel serial form.
Figure 18 illustrates the basis of operation of the vidicon tube camera. Vidicons are
Figure 18. Operating principle of vidicon camera tube.
vacuum tube devices in which an electron beam scans the target and continually restores
the free surface to the potential of the cathode of the electron gun, conventionally regarded
as ground (Thompsett 1979). The illuminated side of the target is maintained at a potential
Va through a transparent electrode. Thus Va appears across the target layer, giving rise to
the electric field, E. The target is photoconductive, i.e. an excellent insulator in the dark,
but readily allowing the passage of charge carriers (electrons and holes) freed by the action
of light. The electric field, E, within the photoconductor causes electrons to drift to the
transparent electrode while holes are drawn to the free surface. Thus a positive latent charge
image forms on the free surface of the illuminated target.
Readout of the latent image is performed by the scanning electron beam. During the time
the beam dwells on a particular pixel, a charge equal to the latent image charge at that pixel
flows from the beam to the target to restore the target to ground potential. A preamplifier
connected to the target forms the video signal from this current. The scanning pattern must
be highly uniform to avoid geometrical distortion and signal distortion (shading).
X-ray noise generated in the XRII and transferred optically by the sensor (σx2 ) is modified
by the MTF of the XRII tube, the coupling optics and the video sensor. However, there
are other sources of noise generated further along the imaging chain and so are independent
of these transfer functions. In well-designed vidicon cameras the only significant noise of
this type is amplifier noise, (σA2 ) most of which arises in the first stage of the preamplifier.
The spectrum of the amplifier noise is ‘triangular’ or peaked at higher spatial frequencies
and thus x-ray noise is dominant at low frequencies and amplifier noise at high frequencies.
The level of σx2 with respect to σA2 is affected by the setting of the optical aperture of the
video camera. The noise in a vidicon arises in the preamplifier, not the tube itself. There
is a passive contribution from the capacitance to ground of the target but this is dependent
solely on the tube design and independent of the photoconductor type. Arnold and Scheibe
(1984) have published the results of an investigation of the noise of XRII systems with
video cameras for the application to digital subtraction angiography.
Fluoroscopic cameras have to be modified for application to digital radiography. It is
desirable that the camera have a linear response to the incident light, i.e. γ = 1.0, where
γ is the slope of the graph of the logarithm of video output versus the logarithm of light
intensity. Two disadvantages of the Sb2 S3 vidicon for digital radiography applications are
its nonlinearity with light intensity (γ ∼ 0.7) as well as an excessively large dark current.
Plumbicon and Saticon tubes have essentially linear response to light intensity and negligible
dark current and are, therefore, preferred for digital radiography.
Digital radiography video cameras are typically operated in pulse progressive readout
(PPR) mode (Baily 1980). In PPR, prior to the x-ray pulse the video camera tube target
is ‘scrubbed’ by continuous scanning. When initiating an x-ray exposure, the camera is
‘blanked’ (i.e. the scanning electron beam current is reduced to zero at the end of the next
full frame of scrubbing) and then the pulsed x-ray exposure is made. The image is then
progressively read off the video camera target by restoring the beam current at the beginning
of the video frame and digitizing at a rate and bandwidth compatible with the required image
quality.
The sensitivity of the digital camera, and therefore the amount of radiation used per
frame, can be controlled by varying the efficiency of optical coupling between the intensifier
and camera tube by the adjustment of an optical diaphragm. Opening the diaphragm will
permit utilization of reduced exposure per frame which will increase the quantum noise.
Typical PPR systems are ‘dose efficient’ in the range from 10–100 µR/frame (25 cm mode)
and 20–200 µR/frame (15 cm mode) (Rowlands et al 1989).
By using a video camera in PPR mode it is possible to capture images efficiently
over a wide range of x-ray exposure times. Video systems with 1024 × 1024 pixels,
coupled to intensifiers of 15 or 25 cm input diameter have been shown to be capable
of clinical image quality equivalent to that of 100 mm photofluorography (Hynes et al
1989). Digital correction of vignetting and structural mottle, coupled with automatic image
enhancement algorithms, can further improve the acceptability of PPR digital images. For
general radiography, larger intensifier fields are necessary with increased matrix sizes to
maintain resolution.
8. Charge-coupled devices (CCDs)
The CCD has been discussed above in relation to its use as a readout device for phosphor-
based detectors. More detail on CCD operation is provided here. The charge-coupled device
was developed in 1970 (Boyle and Smith 1970). Because of its compactness and dynamic
range characteristics, it has virtually replaced the vacuum camera tube in commercial and
home video and it has found many applications in digital imaging. This has largely come
about due to the development of techniques for producing extremely pure crystalline silicon
and for very large scale integration (VLSI).
CCDs are particularly well suited to digital radiography because of their high spatial
resolution capability, wide dynamic range and high degree of linearity with incident signal.
They can be made sensitive to light or to direct electronic input. A CCD is an integrated
circuit formed by depositing a series of electrodes, called ‘gates’ on a semiconductor
substrate to form an array of metal-oxide-semiconductor (MOS) capacitors (figure 19).
By applying voltages to the gates, the material below is depleted to form charge storage
‘wells’. These store charge injected into the CCD or generated within the semiconductor by
the photoelectric absorption of optical quanta. If the voltages over adjacent gates are varied
appropriately, the charge can be transferred from well to well under the gates, much in the
way that boats will move through a set of locks as the potentials (water heights) are adjusted.
In area CCDs, a ‘frame transfer’ system (figure 20(a)) is employed to obtain rapid
readout. Charge is initially accumulated on ‘detector’ pixels and then transferred to an
array of ‘storage’ pixels from which the signal can be read line by line. Alternatively,
‘interline readout’ CCDs (figure 20(b)) have a line of optically shielded storage and transfer
Figure 19. Structure of a CCD array, illustrating motion of stored charge in one direction as
the potential wells are adjusted under control of the gate electrode voltages.
pixels adjacent to each column of detector elements. The charge is rapidly unloaded into
the storage column, freeing the detector elements to accumulate new signal, and transferred
down the storage column elements to a master output register which sequentially receives
signal from each storage column.
These modes of operation are used for the small-format-area imagers which may be
coupled to XRIIs or to phosphors through fibre optics or lenses. Although they provide
rapid readout, these systems require a storage area that is approximately equal to the active
photodetector area. In the case of interline devices, because the storage area is immediately
adjacent to the detector columns, this may cause the effective fill factor of the detector to
be reduced. Area format CCDs are available in sizes varying from 256 × 256 pixels or less
to 2048 × 2048 or more. However, real-time readout (30 frames/s) is currently restricted to
devices of 1000 × 1000 pixels or less.
For scanning systems, as discussed above, it is usually more practical to operate the CCD
in time delay integration (TDI) mode (figure 20(c)). Here, a storage section is not required
as the charge is simultaneously integrated and shifted down the CCD detector columns
toward the horizontal readout register. This type of analogue integration is desirable as
it is relatively noise free. In addition, because all of the detector elements in a column
contribute to each image pixel imaged by that column, the image produced by TDI is
relatively insensitive to a few pixels in the column that may suffer from abnormally low or
high sensitivity.
In any CCD, the charge is transferred ‘bucket brigade’ style over many adjacent
elements. It is, therefore, critical that the efficiency of each transfer is extremely high.
Lack of transfer efficiency can cause a serious loss of spatial resolution in the detector. If
the signal must be shifted across n elements and the efficiency per transfer is ε, then the
overall charge transfer efficiency is εn . Even if ε is 0.999, the efficiency falls to 90%
Figure 20. Typical readout configurations of CCDs showing (a) the frame transfer, (b) interline
transfer and (c) time delay integration (TDI) devices. In (a) and (b) readout storage areas which
are shielded from illumination are required.
over 100 transfers and 37% over 1000 transfers. The effect of less-than-perfect transfer
efficiency is smearing of the image in the readout direction. In commercial CCDs, values
of ε as high as 0.999999 are achievable.
Also of importance is the well storage capacity of the device. Depending on pixel size,
capacities of 300 000 to several million electrons are possible. CCDs designed for video
applications tend to be designed to have extremely small (15 µm) pixel dimensions. For
medical applications, a larger size (25–100 µm) is generally desirable because this offers
greater well capacity and better conforms with other constraints on spatial resolution.
It is important that the CCD be designed with appropriate ‘anti-blooming’ protection
to prevent degradation of the image if some of the charge wells are overfilled. This can
occur in situations where the x-ray detector is exposed to the unattenuated x-ray beam, for
example at the edges of the patient. When the CCD is used in TDI mode, the well capacity
must be such that the integral charge over all stages of integration can be accommodated.
For example, if each detector element accumulates 50 000 electrons per row and there are
64 rows in the CCD over which integration will take place, the well capacity must be 3.2
million electrons.
9. Possible replacements for the XRII
Digital systems based on the use of XRIIs have several disadvantages: the bulky nature
of the intensifier, which often impedes the clinician by limiting access to the patient and
prevents the acquisition of some important radiographic views; loss of image contrast due
to x-ray and light scatter within the tube i.e. veiling glare; geometric (pincushion) distortion
of the image also largely due to the curved input phosphor and ‘S’ distortion which is
attributable to the earth’s magnetic field.

While it is very easy to find fault with the XRII, its longevity points to the fact that
it must be doing most things well. Perhaps its most important feature is that it is x-ray
quantum limited over a huge range of input exposure levels. This is possible because it
is an electrostatic imaging device. To couple the image to the next stage efficiently, it is
important that the image be reduced from the large format at the input to the XRII. In
comparing the electrostatic and optical format reduction as shown in figure 12, it can be
seen that in the optical situation much of the light emitted from the screen is lost and the
proportion of light lost increases rapidly as the demagnification factor is increased (see also
figure 13). In an electrostatic system such as the XRII, the light from the phosphor is
efficiently coupled to the photocathode and the emitted electrons are efficiently collected
and accelerated to provide amplification. This avoids a secondary quantum sink, often a
problem with optical coupling.
9.1. X-icon system

An alternative approach which eliminates the need for the conventional XRII is the X-
icon. This is a directly x-ray sensitive, large-area video camera. In principle, the whole
fluoroscopic imaging chain (XRII, optical distributor and multiple optical devices) could be
replaced by a large-area X-icon whose signal is distributed electronically and this goal has
been pursued by several investigators (Keller and Ploke 1955, Nishida and Okamoto 1968,
Jacobs 1980). Luhta and Rowlands (1993) have reviewed the history of X-icon development
and described their progress toward this goal. Their system, illustrated in figure 21, is a
single-stage device housed within a Pyrex glass vacuum vessel and containing a flat layer
of approximately 500 µm thick amorphous selenium as the x-ray transducer. Otherwise,
its operation is identical to that of the optical video camera described earlier. Due to the
reduced number of stages compared with an XRII/video system it has the potential for
higher resolution. It is inherently a flat-field device and thus problems of distortion and
shading should be of a much less serious nature than for an XRII. X-ray sensitive vidicons
have been investigated previously but none satisfied all the necessary requirements for
medical application. Even if the technical features were adequate, an X-icon could not have
been successful in the past because there was no useful system for (i) storing sequences
of images created by video to replace the ciné camera or (ii) storing single high-quality
small-format radiographs equivalent to those produced by the 100 mm photofluorographic
camera. However, this gap has been bridged by the current availability of high-resolution
video tape and disc recorders and digital frame stores.
10. Flat-panel systems
A flat-panel digital detector, in principle, could perform all current radiological modalities—
radiography, fluoroscopy and fluorography. It could provide high image quality and an
instant readout. Existing x-ray equipment could be easily adapted to employ such detectors.
The technology of large-area active matrix arrays which would form the readout structure
for a flat-panel system has been developed for liquid crystal displays (LCD) for over a
decade.
Active matrix LCDs (AMLCDs) have been made using amorphous (hydrogenated
amorphous silicon (a-Si:H) (Piper et al 1986, Powell 1989), polycrystalline (poly-Si)
or cadmium selenide (CdSe) semiconductors (Brody et al 1984). During recent years,
several manufacturers in Japan, Europe and North America have invested heavily in this
Figure 21. Direct conversion selenium vacuum tube imager (X-icon) capable of both
radiographic and fluoroscopic operation. The evacuated Pyrex housing encloses the target of
amorphous selenium deposited on an aluminium substrate. Electrons emitted by the gun are
deflected by the scanning yoke and impinge on the target. Focus is established with the aid of
the two-part G3 electrode. The supressor mesh is necessary to prevent loss of image stability
under high illumination. This instability is due to the large target potential VT required by the
thick amorphous selenium layer. The signal is obtained directly from the target current and fed
to the pre-amplifier.
development and it is becoming the preferred technology for lap-top computer displays.
Each display panel consists of two sheets of glass with a uniform layer of liquid crystal in
between. One sheet is the active matrix itself (i.e. a large-area integrated circuit consisting
of a large number of thin film field-effect transistors (TFTs) connected to individual pixel
electrodes in a matrix). The other sheet has a uniform electrode layer.
Two general approaches for flat-panel digital x-ray detectors are currently under
investigation. In the first a phosphor layer is used to absorb x-rays and the resultant
light photons are detected by a large-area photodiode array read out with active devices
(for example, thin film transistors or diode switches) integrated onto the plate at each pixel
(Antonuk et al 1991, Fujieda et al 1993). In the second approach (sometimes called the
direct method), x-rays are detected in an amorphous selenium layer and the resulting charges
released are collected on individual pixel electrodes. Finally, the readout occurs using the
active matrix as in the indirect method.
The potential advantages of such self-scanned, readout systems include their
compactness permitting better access to patients than bulky devices such as conventional
XRIIs. Since they are flat, they can be expected to be largely free from veiling glare,
geometrically uniform. Unlike XRIIs they are immune to stray magnetic fields. These
properties facilitate quantitative image analysis, registration and clinical comparison of
images from other modalities, 3D reconstruction applications such as cone beam volume
CT (Ning et al 1991), and use in magnetic environments such as MRI rooms.
10.1. Phosphor flat-panel detectors
Several groups are developing large-area photodetector arrays composed of individual

photodiodes made with amorphous silicon, onto which a conventional x-ray absorbing
phosphor, such as Gd2 O2 S, is placed or thallium-doped caesium iodide (CsI:Tl) is grown
(Perez-Mendez et al 1989, Fujieda et al 1991).
The principle of operation of an amorphous silicon detector is shown schematically in
figure 22. The detector pixels are configured as photodiodes (figure 22(a)) which convert
the optical signal from the phosphor to charge and store that charge on the pixel capacitance.
Being low-noise devices, the photodiodes provide a very large dynamic range, of the order
of 40 000. A typical thin film transistor readout array is shown in figure 22(b). The signal
is read out by activation of scanning control lines for each row of the device, connected
to the gates of TFTs located on each detector pixel. An entire row of the detector array is
activated simultaneously and the signal is read on lines for each column in the array which
connects all the TFT sources in that column to a low-noise charge amplifier. The amplified
signals from the columns are then multiplexed and digitized. This allows fast detector
readout and requires a number of electronic channels equal to the number of columns of
the array. Both radiographic (Antonuk et al 1992) and fluoroscopic (Schiebel et al 1994)
systems have been described.
Alternatively, instead of TFT readout various diode switching schemes can be used
(Chabbal et al 1996, Graeve et al 1996). The advantage of the diode approach is that since
the photodiode has to be made anyway, the switching diode can be made at the same time
without increase in the number of processing steps. The disadvantages of diode readout are
a strong nonlinearity and large charge injection.
The area allocated to each pixel of the array must contain the photodiode, switching
device and control and signal lines so that the fill factor is less than 100%. This potential loss
of x-ray utilization efficiency becomes proportionately greater as the pixel size is decreased
and provides a challenge for the application of this technology to very high-resolution
applications.
The advantage of utilizing CsI as the x-ray absorber is that it can be grown in columnar
crystals which act as fibre optics (figure 7(b)). When coupled to the photodiode pixels,
there is little lateral spread of light and, therefore, high spatial resolution can be maintained.
In addition, unlike conventional phosphors in which diffusion of light and loss of resolution
become worse when the thickness is increased, CsI phosphors can be made thick enough
to ensure a high value of η while maintaining high spatial resolution.
10.2. Solid state electrostatic systems
There are several advantages in the use of solid state electrostatic systems such as the
photoconductor, amorphous selenium (a-Se), or the semiconductor, silicon, in high-purity,
single-crystal form rather than phosphors. The basis of most medical x-ray imaging systems
is a phosphor layer or ‘screen’ shown in figure 7(a). As discussed earlier, x-rays absorbed
in the screen release light which must escape to the surface to create an image and lateral
spread of light is determined by diffusion. Thus the blur diameter is comparable to the
screen thickness. This blurring causes a loss of high-frequency image information which is
fundamental (Sandrik and Wagner 1982) and largely irreversible. The loss can be alleviated
by using a phosphor such as CsI which can be grown in the form reminiscent of a fibre optic
(figure 7(b)). However, the separation between fibres is created by cracking and as a result
the channelling of light is not perfect (Spekowius et al 1995). An alternative and possibly
Figure 22. (a) Principle of amorphous silicon detector (cross-sectional view) showing thin film
transistor (TFT) and the n-i-p a-Si:H photodiode with a transparent conductive indium tin oxide
(ITO) electrode connected to a common bias of ∼ 5 V. The other electrode of the photodiode
is connected to the drain (D) of the TFT which acts as a switch, controlled by the potential on
the gate (G), to connect the photodiode to the source (S). (b) Amorphous silicon matrix readout
array of pixel elements (photodiodes) and thin film transistor switches which can be used to
create a large-area detector. The scanning contol addresses the array line by line and the charge
signals are read out along columns of the device connected to charge amplifiers.
preferable method is to use a structureless electrostatic layer (Brodie and Gutcheck 1982)
as shown in figure 7(c). X-rays interacting in the photoconductor plate release electrons
and holes, which, because they are charged, can be guided directly to the surfaces of the
photoconductor by the applied electric field. The latent charge image on the photoconductor
surface is, therefore, not blurred significantly even if the plate is made thick enough to absorb
most incident x-rays (Que and Rowlands 1995).
Amorphous selenium (a-Se) is the most highly developed photoconductor for x-ray
applications. Its amorphous state makes possible the maintenance of uniform imaging
characteristics to almost atomic scale (there are no grain boundaries) over large areas. The
primary function of the a-Se layer is to attenuate x-rays, generate free electron–hole pairs
(in proportion to the intensity of the incident x-rays) and collect them at the electrodes.
To achieve a high value of η, the detector must be of adequate thickness (figure 4). High
efficiency in converting absorbed x-ray energy into free electron–hole pairs requires high
electric fields. Finally, the number of bulk traps in the layer must be small so that virtually
all the freed carriers reach their appropriate electrode. Each surface must have an electrode
attached to permit collection of charge from the a-Se while preventing entry of charge from
the electrodes into the a-Se. This is called a blocking contact, which must be maintained
even under very high electric fields (Schaffert 1980). Finally the surface of the a-Se at which
the image is formed must have a very small transverse conductivity, otherwise the image
charge could migrate laterally and destroy the resolution. The small transverse conductivity
is achieved by introducing a high density of traps in the a-Se very close to the image
interface. (Pai and Springett 1993).
The first medical application of a-Se, xeroradiography (Boag 1973)—where a latent
charge image on the surface of an a-Se plate was read out with toner—was a technical
and commercial success in its day. Though xeroradiography (see Jeromin 1988) is no
longer competitive, this is probably more because of the toner readout method used at that
time, rather than the underlying properties of a-Se (Brodie and Gutcheck 1985). Thus,
by employing an electronic readout, a-Se again becomes feasible as the basis of a clinical
imaging system.
Neitzel et al (1994) described a commercially available a-Se drum scanner for chest
radiography. This is similar to earlier work by Xerox (Jeromin and Klynn 1979) and
Philips at their Research Laboratories in Aachen (Hillen et al 1988) in which readout is
accomplished by scanning a linear array of tiny electrometers over the electrostatic image
on the selenium surface. May and Lubinski (1993) described a method to read out an a-Se
plate with a phosphor-coated toner and laser scanner. They showed highly detailed images
which were thought to be suitable for mammography. Cook et al (1994) have published
preliminary descriptions of their recent work on laser discharge readout of a-Se. This work
is related to much earlier publications by Korn et al (1978), Zermeno et al (1979), and
DeMonts and Beaumont (1989) as well as the air-gap readout method (Rowlands et al
1991) and readout with a condensed state dielectric (Rowlands and Hunter 1995).
Lee et al (1996) have described a flat-panel method for radiography, based on the use of
an active matrix readout method for a-Se and a similar approach has also been advocated by
Zhao and Rowlands (1992, 1995) for radiography and fluoroscopy. The potential features
of this method are: high image quality, real-time readout rate and compact size. The basic
concept is shown in figure 23. During x-ray exposure, energy is absorbed by the a-Se
layer and the charge created is drawn by the internal electric field ESe to the surfaces. The
image charge is collected by the pixel electrode and accumulated onto the pixel capacitance
(i.e. self-capacitance and an integrated storage capacitor). The pixel electrode and storage
capacitor are connected to the TFT switch of each pixel. The readout device can be similar
to that used with amorphous silicon (figure 22(b)). The external scanning control circuit
generates pulses to turn on all the TFT switches on a row of the array and transfers charge
from the pixel capacitors to the readout rails (columns). The charge is then collected and
amplified by an amplifier on each rail and the data for the entire row are multiplexed out.
(The amplifiers and multiplexer are in another single-crystal silicon integrated circuit which
is wirebonded to the array.) This sequence is repeated for each row of the array. The readout
can be in real-time thus this approach has the potential to be used both in radiography and
fluoroscopy. Fahrig et al (1995) have analysed the factors influencing DQE in a-Se x-ray
detectors.
11. Other direct conversion systems
The advantages of direct conversion of the x-ray signal to electric charge are obvious.
Problems associated with secondary quantum sinks are eliminated and bulky optical coupling
elements are not required. One material which can be used for direct conversion is crystalline
silicon. Silicon has been used for x- and gamma-ray spectroscopy as well as dosimetry. A
Figure 23. Principle of amorphous selenium area detector (see also figure 22(b) for the readout
configuration). The a-Se layer is evaporated directly onto an active matrix array in which the
pixel elements are simply electrodes. An electric field ESe is established across the selenium
by applying an appropriate high voltage to the common top electrode on the a-Se layer. The
charge released by radiation is collected by the electric field on a capacitor made, for example,
by overlapping the pixel electrode and an adjacent gate line.
simple P-I-N photodiode construction can also be configured as an x-ray imaging detector.
Silicon produces an electron–hole pair for each 3.6 eV deposited in the crystal (table 1),
so a 40 keV quantum will produce 11 100 electron–hole pairs. These can be efficiently
collected from the depleted region of the diode by an electric field placed across this
region. Such a detector can be operated at room temperature, but the thermal noise
characteristics can be improved considerably by cooling to a few degrees celsius. The
photodiodes can be produced with very fine pitch, essentially limited only by the number
of quanta that are to be collected in a pixel, i.e. by the acceptable noise level and patient
dose.
As seen in figure 4, silicon, because of its low Z and density, is not an efficient
x-ray absorber at higher energies, and to achieve an acceptable value of η, even at
mammographic energies, the detector must be at least 1 mm thick. For high-resolution
detectors (p = 50 µm), this implies that quanta striking the detector entrance surface at
some distance from the central ray of the x-ray beam will, because of the non-normal
incidence, be likely to be absorbed in an adjacent detector element, causing geometrical
broadening of the line spread function. To avoid this problem, the detector array can be
arranged along a curved surface, with the radius of curvature equal to the source image
plane distance (SID) as shown in figure 24(a). One method of producing this type of
detector (figure 24(b)) is as a hybrid between a pixellated array of photodiodes and a TDI
CCD readout formed on a separate substrate. The two matrices are joined on a pixel-
by-pixel basis by a series of microscopic indium ‘bumps’. Thus charge liberated in the
photodiodes is collected and integrated down CCD columns and finally digitized. Detectors
of this design were initially used for imaging in the infrared spectrum and have been
shown to provide very high spatial resolution and other desirable imaging characteristics
when modified for use with x-rays (Henry et al 1995). It is likely that in the future,
more x-ray absorbent detector materials for the photodiodes such as zinc cadmium telluride,
mercuric iodide or thallium bromide (Shah et al 1989) will be available and will offer
higher η for thinner detector structures making such direct conversion detectors much more
practical.
(b)
Figure 24. (a) Detectors employing low-Z, low-density materials must be thick to achieve
reasonable values of η. To avoid loss of resolution due to geometric spread of signal the
detectors must be curved so that the x-rays are at normal incidence. (b) A hybrid detector that
can employ crystalline silicon or zinc cadmium telluride as the x-ray absorber. The detector
pixels are each connected to elements of a CCD in the readout layer below.
12. Conclusion
There are a number of possible approaches for producing detectors for digital radiography,
although there may not be a single technology that is optimal for all applications. The
technology of choice will depend on spatial resolution and geometry requirements as well
as the necessary imaging speed and acceptable cost. Optimization will require careful
attention to the specification of the clinical imaging task and the key image information that
is required of the examination. Close interaction with the radiologist and the clinician will
help ensure that the imaging technology appropriately meets these requirements.
Acknowledgments
The authors gratefully acknowledge the financial support of their research on digital
radiography by the National Cancer Institute of Canada through a Terry Fox Programme
Project Grant entitled ‘Medical Imaging for Cancer’.
References
Antonuk L E, Boudry J, Kim C W, Longo M J, Morton E J, Yorkston J and Street R A 1991 Signal, noise,
and readout considerations in the development of amorphous silicon photodiode arrays for radiotherapy and
diagnostic imaging Proc. SPIE 1443 108–19
Antonuk L E, Boudry J, Wang W, McShan D, Morton E J, Yorkston J and Street R A 1992 Demonstration
of megavoltage and diagnostic x-ray imaging with hydrogenated amorphous silicon arrays Med. Phys. 19
1455–66
Arnold B and Scheibe P O 1984 Noise analysis of a digital radiography system. Am. J. Radiol. 142 609–13
Baily N A 1980 Video techniques for x-ray imaging and data extraction from roentgenographic and fluoroscopic
presentations Med. Phys. 7 472–91
Barrett H and Swindell W 1981 Radiological Imaging (New York: Academic) pp 285–8
Bendat J S and Piersol A G 1986 Random Data Analysis and Measurement Techniques 2nd Edn (New York: Wiley)
p 338
Boag J W 1973 Xeroradiography Phys. Med. Biol. 18 3–37
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)
Boyle W S and Smith G E 1970 Charge-coupled semiconductor devices Bell Syst. Tech. J. 49 587
Brodie I and Gutcheck R A 1982 Radiographic information theory and application to mammography Med. Phys.
9 79–95
——1985 Minimum exposure estimates for information recording in diagnostic radiology Med. Phys. 12 362–7
Brody P, Luo F C and Malmbert P 1984 Active-matrix addressing enhances flat panels Electronics July 12 113–7
Bunch P C, Huff K E and Van Metter R 1987 Analysis of the detective quantum efficiency of a radiographic
screen-film combination J. Opt. Soc. Am. A 4 902–9
Chabbal J et al 1996 Amorphous silicon x-ray sensor Proc SPIE 2708 499–510
Chan H P, Doi K, Galhotra S, Vborny C J, MacMahon H and Jokich P M 1987 Image feature analysis and computer-
aided diagnosis in digital radiography. 1. Automated detection of microcalcifications in mammography Med.
Phys. 14 538–48
Cook E L, Edwards J D, Nelson O L and Potts J E 1994 Performance of a high resolution radiographic detector The
Society of Imaging Science and Technology 47th Ann. Conf. and Int. Committee on the Science of Photography,
ICPS’94 (Springfield, VA: IS&T) Extended Abstracts pp 699
Cunningham I A, Westmore M S and Fenster A 1994 A spatial frequency dependent quantum accounting diagram
and detective quantum efficiency model of signal and noise propagation in cascaded imaging systems Med.
Phys. 21 417–27
DeGroot P M 1994 Image intensifier design and specifications Proc. Summer School on Specification, Acceptance
Testing and Quality Control of Diagnostic X-ray Imaging Equipment ed J A Seibert, G T Barnes and R G
Gould (Woodbury, NY: AIP) pp 429–60
DeMonts H and Beaumont F 1989 A new photoconductor imaging system for digital radiography Med. Phys. 16
105–9
DiBianca F A and Barker M D 1985 Kinestatic charge detection Med. Phys. 12 339–43
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 22 171–81
Drangova M and Rowlands J A 1986 Optical factors affecting the detective quantum efficiency of radiographic
screens Med. Phys. 13 150–7
Evans R D 1955 The Atomic Nucleus (New York: McGraw-Hill)
Fahrig R, Rowlands J A and Yaffe M J 1995 X-ray imaging with amorphous selenium: Detective quantum
efficiency of photoconductive receptors for digital mammography Med. Phys. 22 153–60
Feig S A and Yaffe M J Digital mammography, computer-aided diagnosis, and telemammography Radiol. Clin. N.
Am. 33 1205–30
Fraser R G et al 1989 Digital imaging of the chest Radiology 171 297–307
Fujieda I, Cho G, Drewery J, Gee T, Jing T, Kaplan S N, Perez-Mendez V and Wildermuth D 1991 X-ray and
charged particle detection with CsI(Tl) layer coupled to a-Si:H photodiode layers IEEE Trans. Nucl. Sci. 38
255–62
Fujieda I, Street R A, Weisfield R L, Nelson S, Nylen P, Perez-Mendez V and Cho G 1993 High sensitivity readout
of 2D a- Si image sensors Japan. J. Appl. Phys. 32 198–204
Giger M L, Ahn N, Doi K, MacMahon H and Metz C E 1990 Computerized detection of pulmonary nodules in
digital chest images: use of morphological filters in reducing false-positive detections Med. Phys. 17 861–5
Goodman L R, Wilson C R and Foley W D 1988 Digital imaging of the chest: promises and problems Am. J.
Radiol. 150 1241–51
Graeve T, Li S, Alexander S M and Huang W 1996 High-resolution amorphous silicon image sensor Proc SPIE
2708 494–8
Hejazi S and Trauernicht D P 1996 Potential image quality in scintillator CCD-based imaging systems for digital
radiography and digital mammography Proc. SPIE 2708 440–9
Henry J M, Yaffe M J, Pi B, Venzon J E, Augustine F and Tumer T O 1995 Solid state x-ray detectors for digital
mammography Proc. SPIE 2432 392–401
Hillen W, Schiebel U and Zaengel T 1987 Imaging performance of a digital storage phosphor system Med. Phys.
14 745–51
——1988 A selenium-based detector system for digital slot-radiography Proc. SPIE 914 253–61
Hounsfield G N 1973 Computerized transverse axial scanning (tomography): Part I Br. J. Radiol. 46 1016–22
Hynes D M, Rowlands J A and Edmonds E W 1989 Clinical comparison of analog and digital 100 mm
photofluorography J. Can. Assoc. Radiol. 40 262–5
Jacobs J E 1980 X-ray sensitive television camera tubes Real-time Radiological Imaging: Medical and Industrial
Applications (Philadelphia, PA: ATSM) p 90
Jeromin L S 1988 Electroradiography Encyclopedia of Medical Devices and Instrumentation vol 2, ed J G Webster
(New York: Wiley) pp 1146–68
Jeromin L S and Klynn L M 1979 Electronic recording of x-ray images J. Appl. Photo. Eng. 5 183–9
Karellas A, Harris L J, Liu H, Davis M A and D’Orsi C J 1992 Charge-coupled device detector: performance
considerations and potential for small-field mammographic imaging applications Med. Phys. 19 1015–23
Kasap S O 1991 Photoreceptors: the selenium alloys Handbook of Imaging Materials ed A S Diamond (New York:
Marcel Dekker) pp 329–77
Kato K 1994 Photostimulable phosphor radiography design considerations Specification, Acceptance Testing and
Quality Control of Diagnostic X-ray Imaging Equipment ed J A Seibert, G T Barnes and R G Gould (American
Association of Physicists in Medicine Monograph #20) (Woodbury, NY: AIP) pp 731–69
Keller M and Ploke M 1955 Visual displays of x-ray photographs using a television camera tube which responds to
x-rays Z. Angew. Phys. 7 562–71 (In German. Translation available: Document 08Q70G- 1162-GJ Associated
Technical Services, Inc., Glen Ridge, NJ, USA)
Klein C A 1968 Bandgap dependence and related features of radiation ionization energies in semiconductors J.
Appl. Phys. 39 2029
Korn D M, Johnson S P, Nelson O L and Ziegler R J 1978 A method of electronic readout of electrophotographic
and electroradiographic images J. Appl. Photogr. Eng. 4 178–82
Lee D L Y, Cheung L K, Palecki E F and Jeromin L S 1996 A discussion on resolution, sensitivity, S/N ratio and
dynamic range of Se-TFT direct digital radiographic detector Proc. SPIE 2708 511–22
Luhta R and Rowlands J A 1993 X-ray sensitive video camera Proc. SPIE 1896 38–49
Maidment A D A, Fahrig R and Yaffe M J 1993 Dynamic range requirements of x-ray detectors for digital
mammography Med. Phys. 20 1621–33 (see also letter in Med. Phys. 21 1215)
Maidment A D A and Yaffe M J 1994 Analysis of the spatial-frequency dependent DQE of optically coupled
digital mammography detectors Med. Phys. 21 721–9
——1996 Analysis of signal propagation in optically coupled detectors for digital mammography: II lens and fibre
optics Phys. Med. Biol. 41 475–93
May J W and Lubinski A R 1993 High resolution computed radiography by scanned luminescent toner radiography
Proc. SPIE 1896 292–312
McGee J D 1979 The history of electronic imaging Electronic Imaging ed T P McLean and P Schagen (London:
Academic) pp 11–54
McGinty G P 1984 Video Cameras: Operation and Servicing (Indianapolis: Howard Sams)
Miller L D 1971 Transfer characteristics and spectral response of television camera tubes Photoelectronic Imaging
Devices vol 1, ed L M Biberman and S Nudelman (New York: Plenum) pp 267–90
Munro P, Rawlinson J A and Fenster 1990 Digital fluoroscopic imaging device for radiotherapy localization Int.
J. Radiat. Oncol. Biol. Phys. 18 641–9
Neitzel U 1994 Discernable gray levels and digitization requirements in digital mammography Med. Phys. 21
1213–4
Neitzel U, Maack I and Guenther-Kohlfahl S 1994 Image quality of a digital chest radiography system based on
a selenium detector Med. Phys. 21 509–16
Ning R, Barsotti J B and Kido D K 1991 Clinical image-intensifier-based volume CT imager for angiography,
Medical Imaging V: Imaging Physics Proc. SPIE 1443 236–49
Nishida R and Okamoto S 1968 An x-ray sensitive vidicon J. Inst. TV Engr. Japan 20 192
Nishikawa R M, Mawdsley G E, Fenster A and Yaffe M J 1987 Scanned projection digital mammography Med.
Phys. 14 717–27
Pai D M and Springett B E 1993 The physics of electrophotography Rev. Mod. Phys. 65 163–211
Perez-Mendez V, Cho G, Fujieda I, Kaplan S N, Qureshi S and Street R A 1989 The application of thick
hydrogenated amorphous silicon layers to charged particle and x-ray detection Mater. Res. Soc. Symp. Proc.
149 621–30 (also Lawrence Berkeley Laboratories Report LBL-26998 April 1989)
Piper W, Bigelow J E, Castleberry D E and Possin G E 1986 The demands on the a-Si FET as a pixel switch for
liquid crystal displays Proc. SPIE 617 10–5
Powell M 1989 The physics of amorphous-silicon thin-film transistors IEEE Trans. Electron Devices 36 2753–63
Que W and Rowlands J A 1995 X-ray imaging using amorphous selenium: inherent resolution Med. Phys. 22
365–74
Rabbani M, Shaw R and Van Metter R 1987 Detective quantum efficiency of imaging systems with amplifying
and scattering mechanisms J. Opt. Soc. Am. A 4 895–901
Roehrig H, Fajardo L Yu T and Schempp W V 1994 Signal, noise and detective quantum efficiency in CCD-based
x-ray imaging systems for use in mammography Proc. SPIE 2163 320–32
Rose A 1948 The sensitivity performance of the eye on an absolute scale J. Opt. Soc. Am. 38 196
Rougeot H 1993 Direct x-ray photoconversion processes Digital Imaging: AAPM 1993 Summer School Proc.
ed W Hendee and J Trueblood (AAPM Monograph 22) (Madison, WI: Medical Physics Publishing)
pp 49–96
Rowlands J A 1994 Fluoroscopic systems and viewing considerations Proc. Summer School on Specification,
Acceptance Testing and Quality Control of Diagnostic X-ray Imaging Equipment ed J A Seibert, G T Barnes
and R G Gould (Woodbury, NY: AIP) pp 483–98
Rowlands J A and Hunter D M 1995, X-ray imaging using amorphous selenium: Generalised analysis of Photo-
induced discharge (PID) readout methods Med. Phys. 22 1983–96
Rowlands J A, Hunter D M and Araj N 1991 X-ray imaging using amorphous selenium: A photoinduced readout
method for digital mammography Med. Phys. 18 421–31
Rowlands J A, Hynes D M and Edmonds E W 1989 System for digital acquisition of gastrointestinal images Med.
Phys. 16 553–60
Sanada S, Doi K, Xu X-W, Yin F-F, Giger M L and MacMahon H 1991 Comparison of imaging properties of a
computed radiographic system and screen-film systems Med. Phys. 18 414–20
Sandrik J M and Wagner R F 1982 Absolute measures of physical image quality: measurement and application to
radiographic magnification Med. Phys. 9 540
Schaffert R M 1980 Electrophotography (London: Focal) p 284
Schiebel U W, Conrads N, Jung N, Weilbrecht M, Wieczorek H, Zaengel T, Powell M J, French I D and Glasse C
1994 Fluoroscopic x-ray imaging with amorphous silicon thin-film arrays Proc. SPIE 2163 129–40
Shah K S, Lund J C, Olschner F, Moy L and Squillante M R 1989 Thallium bromide radiation detectors IEEE
Trans. Nucl. Sci. 36 199–202
Spekowius G, Boerner H, Eckenbach W and Quadfleig P 1995 Simulation of the imaging performance of x-ray
image intensifier/TV camera chains Proc. SPIE 2432 12–23
Swank R K 1973 Absorption and noise in x-ray phosphors J. Appl. Phys. 44 4199–203
Takahashi K, Kohda K and Miyahara J 1984 Mechanism of photostimulated luminescence in BaFX : Eu2+ (X =
Cl, Br) J. Lumin. 31 & 32 266
Tesic M, Mattson R A, Barnes G B, Sones R A and Stickney J B 1983 Digital radiography of the chest: design
features and considerations for a prototype unit Radiology 148 259–63
Thompsett M F 1979 Video-signal generation Electronic Imaging ed T P McLean and P Schagen (London:
Academic) pp 55–101
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 1405–12
Vosburg K G, Swank R K and Houston J M 1977 X-ray image intensifiers Adv. Electron. Electron Phys. 43 205–44
Wagenaar D J and Terwilliger R A 1995 Effects of induced charge in the kinestatic charge detector Med. Phys. 22
27–34
Yaffe M J 1993 Direct digital mammography using a scanned-slot CCD imaging system Med. Progr. Through
Technol. 19 13–21
——1994 Digital Mammography Syllabus of Categorical Course on Technical Aspects of Mammography ed A
Haus and M J Yaffe (Oak Brook, IL: Radiological Society of North America) pp 275–86
Yaffe M J and Nishikawa R M 1994 X-ray imaging concepts: noise, SNR and DQE Specification, Acceptance
Testing and Quality Control of Diagnostic X-ray Imaging Equipment ed J A Seibert, G T Barnes and R G
Gould (American Association of Physicists in Medicine Monograph #20) (Woodbury, NY: AIP) pp 109–44
Zermeno A, Kirby T, Cowart R, Marsh L, and Ong P 1979 Laser readout of electrostatic images Proc. SPIE 173
81–7
Zhao W and Rowlands J A 1992 A large area solid-state detector for radiology using amorphous selenium Proc.
SPIE 1651 134–43
——1995 X-ray imaging using amorphous selenium: Feasibility of a flat panel self-scanned detector for digital
radiology Med. Phys. 22 1595–2604

Yaffe 1997

Uploaded by

Copyright:

Available Formats

Yaffe 1997

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Yaffe 1997

Uploaded by

Copyright:

Available Formats

Home Search Collections Journals About Contact us My IOPscience

X-ray detectors for digital radiography

1997 Phys. Med. Biol. 42 1

Please note that terms and conditions apply.

X-ray detectors for digital radiography

M J Yaffe and J A Rowlands

Received 29 March 1996, in final form 16 August 1996

Figure 2. Schematic diagram of a digital radiography system.

To avoid degradation of image quality in the digitization process, it is important that

Important detector properties are: field coverage, geometrical characteristics, quantum

3.1. Field coverage

3.2. Geometrical characteristics

3.3. Quantum efficiency

3.4. Spatial resolution

Figure 3. Quantum interaction efficiency, η, of various thicknesses of selected phosphors. Note

Figure 4. Quantum interaction efficiency, η, of selected direct conversion detector materials.

Material Z EK (keV) w (eV) ωK (approx.)a

CdTe 48/52 26.7/31.8 4.4 0.85–0.88

3.7. Dynamic range

4. Phosphor-based detector systems

propagation of photons can be considered to be diffusive. This results in a limit to the

Figure 11. Effect of optical coupling efficiency on DQE(f ) of a phosphor–fibre optic–CCD

Figure 16. Schematic representaion of a photostimulable phosphor digital radiography system.

6.1. Strengths and limitations of stimulable phosphors

Figure 17. Image intensifier-based digital radiography system.

7. Image intensifier based detectors

7.1. Video camera

Figure 18. Operating principle of vidicon camera tube.

8. Charge-coupled devices (CCDs)

9. Possible replacements for the XRII

attributable to the earth’s magnetic field.

9.1. X-icon system

10. Flat-panel systems

10.1. Phosphor flat-panel detectors

Several groups are developing large-area photodetector arrays composed of individual

10.2. Solid state electrostatic systems

11. Other direct conversion systems

You might also like