Flat panel CT detectors for sub-second volumetric scanning

Richard E. Colbeth, Ivan P. Mollov, Pieter G. Roos, Edward G. Shapiro

Varian Medical Systems, Ginzton Technology Center,
2599 Garcia Ave, Mountain View, CA, USA, 94043

ABSTRACT

This paper explores the potential of flat panel detectors in sub-second CT scanning applications. Using a PaxScan
4030CB with 600um thick CsI(Tl), a central section of the panel (16 to 32 rows), was scanned at frame rates up to
1000fps. Using this platform, fundamental issues related to high speed scanning were characterized. The offset drift of
the imager over 60 seconds was found to be less than 0.014 ppm/sec relative to full scale. The gain stability over a 10
hour period is better than +/- .45%, which is at the resolution limit of the measurement. Two different types of lag
measurements were performed in order to separate the photodiode array lag from the CsI afterglow. The panel lag was
found to be 0.41% 1st frame and 0.054% 25th frame at 1000fps. The CsI(Tl) afterglow, however, is roughly an order of
magnitude higher, dominating the lag for sub-second scans. At 1000fps the 1st frame lag due to afterglow was 3.3% and
the 25th frame lag was 0.34%. Both the lag and afterglow are independent of signal level and each follows a simple
power law evolution versus time. Reconstructions of anatomical phantoms and the CATPHAN 500 phantom are
presented. With a 2 second, 1200 projection scan of the CATPHAN phantom at 600fps in 32 slice mode, using
120kVp and CTDI100 of 43.2mGy, 0.3% contrast resolution for a 6mm diameter target, can be visualized. In addition,
15lp/cm spatial resolution was achieved with a 2mm slice and a central CTDI100 of 10.8mGy.

Keywords: Flat Panel Detector, FPD, CT, Cone Beam CT, amorphous silicon, high frame rate

1. INTRODUCTION

The purpose of this work is to understand the potential of flat panel technology in sub-second volumetric CT. Since
much of the experimental work is done with a commercially available FPD (the 4030CB) [1], this work also
demonstrates the high frame rate capability of the imager and will hopefully stimulate research on new cone beam CT
(CBCT) applications for FPDs.

Traditional 3rd generation, 64 slice CT scanners are available from multiple manufacturers and a prototype 256 slice
scanner is in clinical trials [2]. The detectors used in these scanners are highly optimized to their task, having dynamic
range on the order of 20 bits and highly absorbing scintillators with low afterglow. However, traditional CT detectors
are also relatively expensive, since each pixel has its own electronics channel, the scintillators are pixelated through a
mechanical process and the underlying visible photon detectors are made from specialized arrays of crystalline Si
photodiodes [3].

In contrast, Flat Panel Detectors (FPDs) utilize multiplexing inside the array, so that the electronics channels are one per
data column. FPDs have typically used grown, columnar CsI, but mechanically pixelated scintillators have been used
in MV applications [4]. A single flat panel detector can be used for multiple applications, including fluoroscopy,
radiography, cine, angiography, DSA and more recently Cone Beam CT. Flat Panel Detectors (FPDs) generally have
many more rows of detectors than even the most advanced multi-slice CT detector and simultaneously, resolution well
beyond that available in CT scanners today. However, FPDs began in relatively low frame rate, low dynamic range
applications. Recently, we have demonstrated expanded dynamic range flat panel detectors in Cone Beam applications,
which approach the contrast resolution of traditional CT scanners [1]. And although the total time required to scan an
organ is roughly the same, the frame rate of flat panel CBCT is generally so low (30fps) that motion artifacts are a
significant issue.

Medical Imaging 2005: Physics of Medical Imaging, edited by Michael J. Flynn, 387
Proceedings of SPIE Vol. 5745 (SPIE, Bellingham, WA, 2005)
1605-7422/05/$15 · doi: 10.1117/12.595631

Downloaded From: http://proceedings.spiedigitallibrary.org/ on 07/15/2016 Terms of Use: http://spiedigitallibrary.org/ss/TermsOfUse.aspx

 In this paper, we explore technology issues surrounding the application of FPDs to sub-second CT scanning, including
maximum frame rate, lag, afterglow and detector stability. High frame rate CT acquisition is demonstrated and image
quality results are discussed.

2. METHODS AND MATERIALS

2.1 Imager and Experimental Setup

The frame rate in FPD’s is fundamentally limited by the required row readout time, since all the pixels in a given row
are captured in parallel. While the RC time constant of the pixel sets the minimum possible line time, there are also
timing requirements associated with electronic capture of the pixel’s signal charge, the subsequent analog-to-digital
conversion and data transmission. For our electronics the minimum row readout time is 30 to 40 µsecs. In order to
increase the frame rate of the FPD, the total number of rows scanned during the frame must be restricted to the quotient
of the desired frame time divided by the minimum row time.

4030CB

1.24 x 40 cm
(16 x 1024)
to
2.48 x 40cm
(32 x 1024)

Rotation
Stage
Collimator
&
G242
x-ray tube

Figure 1. Cone Beam CT test stand. The high speed central strip location is marked.

The experimental work utilized a commercially available 4030CB with custom configuration files designed to scan a
narrow strip in the center of the detector containing a reduced number of rows. This is a straightforward extension of
the zoom capability of the imager and it is possible to switch between these high speed strip modes and normal full field
of view (FOV) modes in just a few frames. It is desirable to continuously discharge the unused pixels to some level, so
that they don’t saturate from dark current build up or scattered x-rays. Thus there is some overhead time associated
with the read time of the central strip. For our setup, it was possible to scan a central 16 row section of the 4030CB at
frame rates up to 1000fps.

Figure 1 shows the panel, x-ray tube and rotation stage. Dynamic Gain Switching (DGS) was used in order to achieve
the highest frame rates with extended bit depth [1]. DGS modes have low and high sensitivity ranges. The panel
switches to the appropriate signal sensitivity automatically, on a pixel-by-pixel basis, depending on the signal level.
The resulting linear range of the detector is in excess of 80,000 counts, or 16.5 bits. For most of the measurements the
pixels were 2x4 binned, resulting in a 388 x 776um pixels, in rows 1024 pixels across. The number of rows (slices) was
varied from 16 to 32 depending on the frame rate and the particular experiment. The panel was externally triggered by
a pulse generator and the choice of frame rate was continuously variable up to 1000fps. The source-to-imager distance
(SID) was 150cm and the source-to-axial distance (SAD) was typically 100cm. The Varian G242 x-ray tube has 0.4
and 0.8mm spot sizes.

388 Proc. of SPIE Vol. 5745

Downloaded From: http://proceedings.spiedigitallibrary.org/ on 07/15/2016 Terms of Use: http://spiedigitallibrary.org/ss/TermsOfUse.aspx

 2.2 Lag Measurements
Two types of lag measurements were performed with the goal of separating the array lag from the CsI afterglow. In
method 1, the x-ray beam was delivered continuously during the scanning, using the radiography mode of the generator.
The lag signal was calculated as the residual signal in the post exposure frames divided by the average signal just prior
to the beam termination. A silicon photodiode coupled to a CdWO4 crystal was used to characterize the beam rise and
fall times. Figure 2 shows a comparison of the panel signal to the measured x-ray signal. Notice that the panel follows
the beam rise time reasonably well, but lags the sharp cut off of the beam. Because the beam and the panel are
asynchronous, the signal in the frame where the beam terminated varied considerably, depending on the exact amount of
radiation captured. For all the lag measurements acquired with this technique, the post exposure frames were
considered to begin after the x-ray beam had fully terminated.

1.2

1.0

0.8
Signal

0.6 1st lag

frame
0.4
Photodiode w/ CdWO
0.2
Panel

0.0

-0.2
-5 0 5 10 15
time (msec)

Figure 2. X-ray beam signal compared the average signal from the panel at 1000fps.

The above measurement certainly contains both the array lag and the signal from the CsI afterglow. From the data sheet
for our CsI coating, the afterglow is expected to decay to approximately 0.1% after 10msec. With this in mind a special
version of firmware was constructed in which a single row could be held in accumulation mode, until the afterglow
decayed sufficiently, and then scanned repeatedly at high frame rate. Figure 3 shows the experimental timing for the
2000fps (0.5msec) line scanning. Note that the 1st reading of the line contains the x-ray signal plus the accumulated
signal from the CsI afterglow. This 1st reading, in theory, does not contain the lag of the array, since the amount of
trapped charge should remain in steady state while the signal population on the photodiode remains relatively constant.
The trapped charge depopulation and lag decay start when the scanning begins. The array lag was calculated from
residual signal starting with the 2nd frame. Using this pulsed fluoroscopy method it was possible to use extensive
averaging to reduce the noise in the measurement.

Figure 3. Timing of the experimental procedure used to measure the array lag.

Proc. of SPIE Vol. 5745 389

Downloaded From: http://proceedings.spiedigitallibrary.org/ on 07/15/2016 Terms of Use: http://spiedigitallibrary.org/ss/TermsOfUse.aspx

 2.3 Offset and Gain Stability
The offset and gain stability were considered over time scales relevant to commercial CT scanners. Normally an offset
correction is done just prior to the scan and is required to remain stable during volumetric data acquisition. Typical
acquisition times are on the order of 20 to 30 seconds. For our measurement dark files were collected at 1000fps.
Every 30th frame was sampled and 2000 frames were collected, which spanned a 60 second time period. Generally the
offset is temperature dependent. In this case no special precautions were taken to stabilize the temperature of the panel.
As can be seen in Figure 1, there is a fan blowing on the imager, but it is out in the open room air subject to ambient
temperature changes.

The gain stability is of interest over a period of a day, since CT scanners are typically gain calibrated once a day. The
gain stability was measured with the 4030CB in full FOV mode. A flat, open field image with 90% recursive filtering
was periodically acquired in pulsed fluoroscopy mode, with an x-ray technique of 120kVp, 50mA, 5msec and 1.25mm
Cu plus 1mm Al filtration. A dosimeter was mounted on the front surface of the imager and allowed to accumulate
greater than 20mR for each reading. Using the dosimeter readings and the average counts from the panel, the sensitivity
in counts/uR is calculated and tracked over time. The estimated standard deviation in the measurement, due to
dosimetry error and noise in the imager flat file, is +/- 0.45%.

2.4 DQE
The DQE is a measure of the efficiency with which a system utilizes incoming quanta. It requires knowledge of
the Modulation Transfer Function (MTF), the Noise Power Spectrum (NPS) and the average x-ray flux. A useful
practical definition applicable for linear systems is [5]:
d * MTF 2 (u )
( )
2

DQE q, u =
q * NPS q, u ( ) (1)

where the DQE is expressed as a function of the spatial frequency u and the photon flux q , which is expressed
2
as phtotons / mm / mR ; d is the average pixel signal expressed in the same units as those used in the
calculation of NPS.

We calculated the photon flux using the x-ray spectral modeling methodology of Boone and Seibert [6] for the
actual conditions used in our experiments, and also cross-checked the calculation method against one of the
values defined in ISO 92361 obtaining agreement within ~1.5%.

The NPS was obtained by a method similar to that described by many authors including Dobbins [7]. Flat fields
taken at known x-ray exposures, were processed in regions, typically 128x128 pixels. For each individual region,
the 2D FFT is calculated and the results averaged for many regions. The normalization of the FFT is checked to
ensure that the squared Fourier amplitudes sum to the pixel variance. The 2D NPS is calculated from the 2D FFT
2
 
by scaling by  1 where f inc is the frequency increment given by f inc = 1 /( N * pixel _ side) and N
 f inc 
is the size of the region used (128 typically). Providing there is no directional preference indicated in the data the
1D NPS is obtained using all of the points in the 2D plane, excluding those on-axis, which are most susceptible to
artifacts.

Finally, it is necessary to correct for any lag in the imager, as discussed by Granfors and Aufrichtig [8]. The lag
compensation was done by scaling the DQE by (1 − lag1 ) , where lag 1 is the first frame lag.
2

In the data presented below, a pulsed fluoroscopy source was used. The tube voltage was 70kVp with a beam
hardness of 7.1mm Al half value layer (RQA5). For a given DQE vs spatial resolution curve, every data point is
plotted and the scatter in the data indicates the uncertainty in the measurements. Because of the size of our
ionization chamber, the DQE was measured in the full FOV, 2x2 binning, analog gain=1, DGS mode at 30fps.

390 Proc. of SPIE Vol. 5745

Downloaded From: http://proceedings.spiedigitallibrary.org/ on 07/15/2016 Terms of Use: http://spiedigitallibrary.org/ss/TermsOfUse.aspx

 2.5 CTDI Dosimetry
Dose measurements were taken with a Victoreen 660 Survey Meter and Acrylic CTDI phantoms that are intended for
CT dosimetry. The detector probe is a 10cm x 1 cm diameter open air ionization chamber. The head acrylic cylinder is
16 cm in diameter, 15 cm length and the body acrylic cylinder is 32 cm diameter. The SAD of the acrylic phantoms was
100 cm, which is about double the typical CT scanner value. The CTDI100 (100mm chamber) at the center and periphery
was measured and normalized to 100 mAs for both acrylic phantoms and air scans and the results are shown in Table 1.
The CTDIW was calculated using the 1/3 center plus 2/3 peripheral dose for both head and body techniques. The beam
filtration was flat at 4.8 mm Al equivalents, which is similar to diagnostic scanners at the center. Shaped filters were not
used for these tests. The mGy/mAs values in the table are a factor of 3 to 4 lower than typical CT scanners due to our
large SAD and flat beam filtration.

Table 1. CTDI measurements at a technique of 120 kV @ 200 mAs, slice thickness of 22.5 mm, beam
filtration of 4.8 mm Al equivalents and SAD of 100 cm.
Phantom CTDI100 Center CTDI100 Periphery CTDIW
(mGy/100 mAs) (mGy/100 mAs) (mGy/100 mAs)
Head acrylic 4.5 5.8 5.4
Body acrylic 1.5 4.9 3.8
Air 7.8

3. RESULTS
3.1 DQE in Dynamic Gain Switching Mode
As discussed briefly above, in Dynamic Gain Switching (DGS) mode the panel switches between high and low
sensitivity on a pixel-by-pixel basis. The output data contains a 14bit measurement plus a gain bit for each pixel.
The offset & gain normalization is then applied post image capture. Figure 4 shows the Signal and SNR vs.
mAs/frame for the DGS mode running with 16 rows, 2x4 binning, analog gain equal 1 and 1000fps. Figure 5 is
the measured DQE with beam quality RQA5, for the DGS mode with 2x2 binning and analog voltage gain
= 1 in full FOV mode at 30fps.
1000
80,000

70,000

60,000
100
50,000
Signal

SNR

40,000

30,000
10
20,000

10,000

0
1
0.000 0.050 0.100 0.150 0.200 0.250 0.300 0.350
0.001 0.010 0.100 1.000
mAs/frame mAs/frame

Figure 4. Signal and SNR vs. mAs/frame for 2x4 binning,, 16 x 1024 DGS mode running at 1000fps.

Proc. of SPIE Vol. 5745 391

Downloaded From: http://proceedings.spiedigitallibrary.org/ on 07/15/2016 Terms of Use: http://spiedigitallibrary.org/ss/TermsOfUse.aspx

 0.8

0.7

0.6
5.1 nGy
0.5
13.1 nGy
DQE

0.4 23.8 nGy

62 nGy
0.3
241 nGy
0.2

0.1

0
0 0.2 0.4 0.6 0.8 1 1.2 1.4
Spatial Frequency (lp/mm)

Figure 5. DQE of the DGS mode with full FOV 2x2, Gain=1 at 30fps. The beam quality was RQA5.

3.2 Offset and Gain Stability

The offset and gain stability were measured as described above and the results are shown in Figures 6 and 7. The offset
drift of the panel over 60 seconds was 0.0011 counts (adu’s) per second. Assuming an 80,000 count max signal range,
the offset drift is 0.014ppm/sec. In a 60 second scan the offset can be expected to drift approximately 0.8 ppm. Again
there was no temperature control of the panel, which is normally applied to CT detectors, so this represents an upper
bound.

1713

y = 0.0011x + 1711.7
1712.5
Average Offset

1712
(adu's)

1711.5

1711

1710.5
0 10 20 30 40 50 60
Time (seconds)
Figure 6. Offset drift versus time at 1000fps.

The gain stability measurements seen in Figure 7, show that the peak-to-peak variation in sensitivity over a 10 hour
period was 0.9%. However the standard deviation of the measurement is +/- 0.45%, so the measured gain variations
appear to be limited by the uncertainty in the experiment.

392 Proc. of SPIE Vol. 5745

Downloaded From: http://proceedings.spiedigitallibrary.org/ on 07/15/2016 Terms of Use: http://spiedigitallibrary.org/ss/TermsOfUse.aspx

 99.00

Sensitivity (counts/uR)
98.00

average = 96.9
97.00

96.00

95.00
1 10 100 1000
Time (minutes)

Figure 7. Gain variation over a 10 hour period.

3.3 Lag and Afterglow

Figure 8 shows the lag data taken with method 1 (afterglow + array lag), for panel input signals varying from 1% to
75% of full scale. Independent of signal level, the residual signal evolved according to the same power law function.
For method 2 (array lag only), the array lag was found to be independent of input signal level and source energy, which
is shown in Figure 9. The photodiode array lag and the CsI afterglow are shown in Figure 10. The curve labeled CsI
afterglow actually contains both the afterglow and the array lag. But since the array lag is roughly an order of
magnitude lower, this curve is a good approximation to the afterglow. Also notice that the projected afterglow signal at
175msec is an order of magnitude less than the 1st frame array lag, which is consistent with our assumption that the
afterglow is negligible after this length of time. Finally, we can see that at 1000fps the afterglow is predicted to be the
dominant source of residual signal over an entire 0.5 second scan.

100.00%

10.00%
Residual Signal
(%)

1.00%

0.10%

0.01%
0.001 0.01 0.1
Time (seconds)

Figure 8. CsI afterglow vs. time for input levels from 1% to 75% of the maximum linear input dose.

Proc. of SPIE Vol. 5745 393

Downloaded From: http://proceedings.spiedigitallibrary.org/ on 07/15/2016 Terms of Use: http://spiedigitallibrary.org/ss/TermsOfUse.aspx

 1.00%

70kVp, 390uR, 2670cts

90kVp, 816uR, 5124 cts
Lag (%) 120kVp, 1033uR, 8511 cts

0.10%

0.01%
1 10 100
Frame #

Figure 9. Array lag measured with Method 2 at 2000fps.

10.00%
1000fps

1.00%
Residual Signal

y = 0.0651x-0.9025
(%)

0.10%

Array Lag -0.2612

y = 0.0006x
CsI(Tl) Afterglow

0.01%
0.10 1.00 10.00 100.00 1000.00
Time (msec)

Figure 10. Residual signal due to photodiode array lag (Method 2) and CsI afterglow (Method 1), at 1000fps.

394 Proc. of SPIE Vol. 5745

Downloaded From: http://proceedings.spiedigitallibrary.org/ on 07/15/2016 Terms of Use: http://spiedigitallibrary.org/ss/TermsOfUse.aspx

 3.4 Phantom Results
In addition to the physics measurements a number of high frame volumetric scans have been performed. The
experimental setup was limited both in rotation speed and the x-ray flux. The maximum rotation speed of our stage is
0.5 revolutions/sec. The generator and x-ray tube are intended for CBCT applications where the rotation time is on the
order of 30 seconds and the x-ray dose is delivered in pulsed fluoroscopy mode. In order to approach the exposure
required for sub-second CT scans, the high speed scans used extended 2 second radiography shots. This limited us to a
maximum of 400mAs at 120kVp. Because of the geometric repeatability of the experimental apparatus, for some of
the scans it was possible to sum multiple scans to increase the total CTDI.

Figure 11 shows an axial slice from a high frame rate scan of a chest phantom, and for comparison, an image of the
same phantom taken on a commercial CT scanner. In order to generate a comparable level of dose, the flat panel
data is the sum of four individual scans. The flat panel scan has not been corrected for beam hardening, the effects
of afterglow, or scatter.

(a) (b)

Figure 11. (a) Chest phantom scanned over 2 sec, 1200 views, taken at 120kVp, 2x4 binning, 32x1024
matrix. An additional 2x1 binning during reconstruction, created pixels 0.776mm square. The
CTDIW = 57mGy. (b) A scan for the same phantom taken on a commercial scanner is shown on
the right for comparison.

In Figures 12 and 13 are the results of scans of the CATPHAN 500 phantom. In the resolution phantom 15lp/cm can be
resolved. This is less than expected, since scans of the same phantom at 30fps have achieved greater than 20lp/cm.
Since no lag correction has been applied, the degradation in resolution may be attributable to the CsI afterglow, but this
is still under investigation. The scan of the low contrast phantom revealed that the 0.3% contrast objects could be
visualized for target diameters as low as 6mm. Increasing the re-binning in the axial plane did reduce the high
frequency noise observed, but had no significant affect on the observable contrast resolution.

Proc. of SPIE Vol. 5745 395

Downloaded From: http://proceedings.spiedigitallibrary.org/ on 07/15/2016 Terms of Use: http://spiedigitallibrary.org/ss/TermsOfUse.aspx

 (a) (b)

(c)

Figure 12. (a) + (c) 2 sec, 1200 view scan of CATPHAN 500 line pair per cm high resolution gauge taken at 120kVp,
2x4 binning, 32x1024 matrix. The slice thickness was 2mm and the effective patient dose at the center was 10.8mGy
CTDI100. The final pixel size for projection images was 0.388 (tangential) and 0.776 (axial). (b) The calibration target
on the right was also scanned over 2 seconds, 1200 views, but with a 0.776 x 0.776 pixel size. The effective patient
dose was 10.8mGy CTDI100 at patient center. The slice thickness was 5mm.

396 Proc. of SPIE Vol. 5745

Downloaded From: http://proceedings.spiedigitallibrary.org/ on 07/15/2016 Terms of Use: http://spiedigitallibrary.org/ss/TermsOfUse.aspx

 0.3%
6 mm
(a) (b)

Figure 13. CATPHAN low contrast phantom, 10mm slice acquired at 600fps, 2x4 pixels, 32 x 1024 matrix, 120kVp
and CTDI100 (central) = 43.2 mGy. The image on the right is reconstructed with additional 4x1 binning, pixel size
1.552 (tangential) x 0.776(axial) mm at the detector. The image on the left is reconstructed with an additional 2x1
binning for a pixel size of 0.776 x 0.776 mm.

4. CONCLUSIONS

Research and applications of flat panel based cone beam CT have increased significantly over the last few years;
however, all this work has, of necessity, been limited to relatively slow scans at low imager frame rates. In this
investigation we have characterized the behavior of a flat panel detector at frame rates compatible with sub-second
scanning. More broadly, this work explores the potential for multiplexed detectors in CT scanning.

The results are very promising. One of the early concerns was the lag behavior of the amorphous silicon photodiode
array, since at low frame rates the short term lag can be near 10% and the long-term lag on the order of 1% for many
seconds. At high frame rates, the integration time for capturing the lag is so reduced that the issue of residual signal
becomes dominated by the scintillator afterglow, which is essentially the situation in state-of-the-art CT scanners.
Stability was also a concern. Although flat panel imagers have always used offset and gain normalization like CT
detectors, the level of stability required is much higher in CT applications. Our results show that the offset stability is
extremely good and that the gain stability appears adequate over the time scales required for CT. Using the extended
dynamic range capability of the 4030CB panel, we have demonstrated 0.3% contrast resolution for a 6mm diameter
target.

Proc. of SPIE Vol. 5745 397

Downloaded From: http://proceedings.spiedigitallibrary.org/ on 07/15/2016 Terms of Use: http://spiedigitallibrary.org/ss/TermsOfUse.aspx

 ACKNOWLEDGEMENTS

The authors would like to acknowledge the valuable discussion and insight of M. Grasruck, Siemens Medical
Solutions, in applying the dynamic gain switching mode to CT.

REFERENCES

1. P.G. Roos et al., “Multiple gain ranging readout method to extend the dynamic range of amorphous silicon flat
panel imagers,” Proceedings of SPIE Medical Imaging, 5368-16, Feb. 2004, San Diego, CA.
2. 6th Annual International Symposium on Multidetector-Row CT, Ritz-Carlton, San Francisco, CA, June 23-26,
2004.
3. S. Utrup et al. “Design and performance of a 32-slice CT detector system using back-illuminated
photodiodes,” Proceedings of SPIE Medical Imaging, 5368-06, Feb. 2004, San Diego, CA.
4. E.J. Seppi et al., “Megavoltage cone-beam CT using a high efficiency image receptor,” Int. J. Radiation
Oncology Biol. Phys., Vol.55, No.3, pp. 793-803, 2003.
5. I.A. Cunningham in Handbook of Medical Imaging (HMI), Volume 1, p124, Edited by J. Beutel, H.L. Kundel
and R.L. Van Metter. SPIE press copyright 2000.
6. J.M. Boone and J.A.Seibert, Medical Physics 24, 1661 (1997)
7. Dobbins in HMI p197 et seq.
8. P.R. Granfors, R. Aufrichtig, “DQE(f) of an amorphous-silicon flat-panel x-ray detector: detector parameter
influences and measurement methodology,” Proc. SPIE Med.Img.2000, vol.3977, pg.2-13, 13-15 Feb 2000.

398 Proc. of SPIE Vol. 5745

Downloaded From: http://proceedings.spiedigitallibrary.org/ on 07/15/2016 Terms of Use: http://spiedigitallibrary.org/ss/TermsOfUse.aspx