You are on page 1of 12

Performance characteristics of a Kodak computed radiography system

Carla D. Bradford, Walter W. Peppler, and James T. Dobbins III

Citation: Medical Physics 26, 27 (1999); doi: 10.1118/1.598781


View online: http://dx.doi.org/10.1118/1.598781
View Table of Contents: http://scitation.aip.org/content/aapm/journal/medphys/26/1?ver=pdfcov
Published by the American Association of Physicists in Medicine

Articles you may be interested in


Comparison of different computed radiography systems: Physical characterization and contrast detail analysis
Med. Phys. 37, 440 (2010); 10.1118/1.3284539

Characterization of noise sources for two generations of computed radiography systems using powder and
crystalline photostimulable phosphors
Med. Phys. 34, 3345 (2007); 10.1118/1.2750973

Performance evaluation of computed radiography systems


Med. Phys. 28, 361 (2001); 10.1118/1.1350586

Physical imaging performance of a compact computed radiography acquisition device


Med. Phys. 25, 354 (1998); 10.1118/1.598212

Relative speeds of Kodak computed radiography phosphors and screen-film systems


Med. Phys. 24, 1621 (1997); 10.1118/1.597969
Performance characteristics of a Kodak computed radiography system
Carla D. Bradford and Walter W. Peppler
Department of Medical Physics, The University of Wisconsin, Madison, Wisconsin 53706
James T. Dobbins III
Thoracic Imaging Research Division, Department of Radiology, Duke University Medical Center, Durham,
North Carolina 27710
~Received 26 January 1998; accepted for publication 27 October 1998!
The performance characteristics of a photostimulable phosphor based computed radiographic ~CR!
system were studied. The modulation transfer function ~MTF!, noise power spectra ~NPS!, and
detective quantum efficiency ~DQE! of the Kodak Digital Science computed radiography ~CR!
system ~Eastman Kodak Co.-model 400! were measured and compared to previously published
results of a Fuji based CR system ~Philips Medical Systems-PCR model 7000!. To maximize
comparability, the same measurement techniques and analysis methods were used. The DQE at four
exposure levels ~30, 3, 0.3, 0.03 mR! and two plate types ~standard and high resolution! were
calculated from the NPS and MTF measurements. The NPS was determined from two-dimensional
Fourier analysis of uniformly exposed plates. The presampling MTF was determined from the
Fourier transform ~FT! of the system’s finely sampled line spread function ~LSF! as produced by a
narrow slit. A comparison of the slit type ~‘‘beveled edge’’ versus ‘‘straight edge’’! and its effect
on the resulting MTF measurements was also performed. The results show that both systems are
comparable in resolution performance. The noise power studies indicated a higher level of noise for
the Kodak images ~;20% at the low exposure levels and 40%–70% at higher exposure levels!.
Within the clinically relevant exposure range ~0.3–3 mR!, the resulting DQE for the Kodak plates
ranged between 20%–50% lower than for the corresponding Fuji plates. Measurements of the
presampling MTF with the two slit types have shown that a correction factor can be applied to
compensate for transmission through the relief edges. © 1999 American Association of Physicists
in Medicine. @S0094-2405~99!00101-7#

Key words: computed radiography ~CR!, digital radiography, modulation transfer function ~MTF!,
noise power spectrum ~NPS!, detective quantum efficiency ~DQE!

I. INTRODUCTION spectra ~NPS!, and detective quantum efficiency ~DQE! of


the Kodak system were measured and compared to the pub-
Photostimulable phosphor ~PSP! based computed radiogra- lished results presented by Dobbins et al.4 of this Fuji sys-
phy ~CR! has gained increasingly widespread acceptance tem. To provide an accurate comparison, the same measure-
since it was first commercialized in 1983. Numerous publi- ment techniques and analysis methods used by Dobbins et al.
cations have shown CR to be an acceptable and sometimes in the assessment of the Fuji systems were applied in this
more favorable choice over conventional screen/film systems study of the Kodak system.
for a variety of applications.1–3 Advantages with CR include:
Convenient storage of radiological data,3 increased flexibility
in image processing, and consistent reproducibility along
with a greater dynamic range, wider exposure latitude, and II. MATERIALS AND METHODS
reduced patient exposure.1,2
A. Equipment
For the first nine years after the commercialization of CR,
the Fuji Photo Film Company had primarily developed the The Kodak Digital Science CR System 400 consists of the
majority of plates and readers available. A number of studies storage phosphor reader, two types of storage phosphor
had been conducted on this equipment to characterize its screens, a bar code scanner for entering patient and exam
physical performance.4–6 By 1992 two other manufacturers information, and a workstation for reviewing the radio-
~Eastman Kodak Co., Agfa Gevaert Group! had introduced graphs. There have been two models of the storage phosphor
their own CR systems. To date there is little published docu- reader: Model 3110 and Model 400. These models differ
mentation evaluating the characteristics of these new mainly in their detection systems. We studied only the newer
systems.7 This paper serves to study the physical perfor- Model 400 storage phosphor reader.
mance of the Kodak Digital Science CR system 400 reader8 The two screen types available are general purpose ~GP-
and compare these results with a CR system manufactured 25! and high resolution ~HR!. Table I lists the screen sizes
by Fuji Photo Film ~sold by Philips Medical Systems, PCR- studied for each screen type. Table II lists the corresponding
7000!. The modulation transfer function ~MTF!, noise power image, pixel and study file sizes for the two CR screen sizes

27 Med. Phys. 26 „1…, January 1999 0094-2405/99/26„1…/27/11/$15.00 © 1999 Am. Assoc. Phys. Med. 27
28 Bradford, Peppler, and Dobbins III: Characteristic of a radiography system 28

TABLE I. Kodak screen type and sizes studied. slowscan directions for each screen type and size studied.
~Other publications have referred to these two directions by
Screen Type Screen size
various terms. The ‘‘fastscan’’ direction has been also
GP-25 35 cm343 cm termed the ‘‘laser scan’’ or ‘‘scan’’ direction. The ‘‘slows-
18 cm324 cm can’’ direction has been also described as the ‘‘plate scan’’
HR 18 cm324 cm
or ‘‘subscan’’ direction.! This was accomplished by attach-
ing a 10 mm slit camera ~Radiation Measurements, model:
07-624! to the front surface of the cassette with the side
studied. ~The study file contains the 12-bit logarithmic data, marked ‘‘focus side’’ facing the focal spot. The slit’s long
look-up tables, display images and masks.! axis was slightly angled ~;5°–8°! with respect to the central
The Kodak Model 400 reader is able to scan an area axes of the CR plate to provide adequate sampling for the
smaller than the actual screen size. This option allowed us to composite LSF. Shielding ~2 mm of lead! was placed around
study two different scan conditions with the use of one the slit camera to block any forward-directed scatter. The
screen. For example, to measure the system’s performance cassette/slit assembly was oriented vertically on the tabletop
for the 18 cm324 cm plate, a specific portion of the perpendicular to the edge of the table, and therefore, the
35 cm343 cm plate was exposed. When this larger plate was x-ray beam, using a square corner and positioned at an SDD
then scanned, the 18 cm324 cm option was selected. By us- of ;170 cm. The vertical orientation eliminated backscatter
ing one plate for two different sizes, effects from slight sen- since there was nothing behind the cassette. This simple cas-
sitivity variations in the phosphor from one screen to another sette alignment method was quite adequate and was con-
were removed.8 Unfortunately, this process could be done firmed by the consistency of the maxima of the line spread
for just the GP-25 screen type since only the smaller functions obtained. One at a time the HR (18 cm324 cm)
18 cm324 cm size was available for the HR screen. and GP-25 (35 cm343 cm) screens were exposed. After a
Before every measurement, each screen was flash erased ten min delay the screens were read and then erased. The
to remove any accumulated signals due to extraneous back- GP-25 screen was exposed a second time at this same dis-
ground radiation and/or the naturally occurring isotopes tance for the 18 cm324 cm GP-25 measurement. Determina-
present in the phosphor. After each screen exposure, there tion of the mAs factors were based on the goal to produce a
was a ten min delay before reading to minimize the effects of suitable LSF down to the 0.1% level.
latent image decay.9 The screens were cleaned prior to use. Due to the slit camera’s construction ~see Fig. 1!, the
The x-ray source used to expose each plate was a Phillips relief edges on each jaw ~4°! allowed a tolerance in the ac-
x-ray tube ~Model SRO 33/100! with a Phillips Maximus curacy of alignment. However, due to transmission through
C850 ‘‘Classic’’ ~three phase, twelve pulse! generator. For the beveled edges, the width of this slit was effectively larger
all plate exposures, the x-ray unit was set to 70 kVp and than the nominal 10 mm. To determine the effect of the bev-
confirmed within 2 kV by an RMI kVp meter ~Model 240A!. eled edges on the resulting MTF curves, repeat MTF mea-
Filtration of 0.5 mm Cu was applied during all measure- surements ~in both the fastscan and slowscan directions!
ments. The aluminum equivalent half-value layer for this fil- were conducted using a straight edge slit with the 100 mm/
tered spectrum was 7 mm. Exposure levels ~‘‘in air’’! were pixel HR Kodak plate. Comparisons between the MTF
measured with an ADCL ~Accredited Dosimetry Calibration curves made by the two slit types produced a correction fac-
Lab! calibrated exposure meter ~ionization chamber: Kei- tor that was applied to the less accurate MTF curves obtained
thley, Model 96035; electrometer: Keithley, Model 602!. with the beveled slit. The correction factor is described be-
The x-ray tube was aimed horizontally with the central axis low.
of the beam parallel to the edge of the x-ray table and the The greater availability of this commercial slit and its
cassette was mounted vertically on the table perpendicular to more forgiving alignment requirements ~as opposed to the
the beam using a square corner and the edge of the table. straight edge slit! will allow others to more easily reproduce
this MTF measurement method. Due to these advantages, we
chose to use the beveled slit for all of our MTF measure-
B. MTF measurements
The method of Fujita et al.6 as adapted by Dobbins et al.4
was applied to measure the presampling MTF for each plate
type and size. The line spread function ~LSF! was obtained
by acquiring images of a narrow slit in both the fastscan and

TABLE II. Image, pixel, and study file sizes for the two CR screen sizes
studied.

Screen size Image size Pixel size Study size

35 cm343 cm 204832500 0.172 cm 11.5 MB


18 cm324 cm 179232392 0.100 mm 9.7 MB FIG. 1. Cross section of the beveled slit camera ~Radiation Measurements,
Inc.!.

Medical Physics, Vol. 26, No. 1, January 1999


29 Bradford, Peppler, and Dobbins III: Characteristic of a radiography system 29

TABLE III. Plate correction factors.

Screen type Screen size Cal0 Cal1

GP-25 35 cm343 cm 20.067 1.005


18 cm324 cm 20.122 0.995
HR 18 cm324 cm 20.102 1.011

ments. In order to compare with the results from Dobbins


et al. for the Fuji manufactured CR system, the necessary
correction factor was applied.
FIG. 2. Example of a transverse profile plot along the length of the slit. The
C. NPS measurements maxima occur when the slit’s central axis passes directly through the center.
The minima occur when the axis passes exactly between two pixel centers.
Images with uniform exposure ~flat field images! were
acquired at: 30, 3, 0.3, and 0.03 mR for the NPS measure-
ments. These exposure levels were determined by first set- A. MTF analysis
ting the mAs factor and then adjusting the distance between In addition to the presampling MTF, the expectation value
the x-ray source and ionization chamber until the desired of the modulation transfer function, EMTF, was also com-
exposure was detected. The mAs factor for these four expo- puted from the slit images. The EMTF is the average of the
sure levels were 250, 25, 2.5, and 0.25 mAs, respectively. digital MTF over all possible phases of the sampling comb.10
The distance between the source and reference point of the The noise equivalent quanta ~NEQ! and subsequently the
detector ~SDD! ranged between 1.6 and 1.9 m. The measured detective quantum efficiency ~DQE! of the system were de-
exposure values were within 2% for the 30, 3 and 0.3 mR termined using this EMTF curve. Calculation of the EMTF
desired values and within 3% for the 0.03 mR value. The 3% consisted of the following steps:
accuracy for the 0.03 mR exposure measurement was accom-
plished by taking ten successive exposures ~during the expo- ~i! calculation of slit angle based on transverse slit profile
sure measurement step only! at the 0.03 mR setting provid- plots
ing the chamber with an integral reading of ;0.3 mR. ~ii! computation of initial composite LSF & extrapolation
Once the SDD was determined, a cassette was placed on of LSF tails
the tabletop at this predetermined distance. The cassette was ~iii! computation of presampling MTF
oriented vertically on the table with nothing behind it that ~iv! correction for transmission through beveled edges
would produce scatter. Similar to the MTF measurements, ~v! computation of EMTF
the same two screens were used for the three screen sizes. Transverse profile plots are helpful to determine the
The above procedure was then repeated ~with the same two amount of usable data supplied in the slit image. Figure 2
screens! for each exposure level. shows an example of such a plot. Each data value in this plot
represents the maximum pixel value in each transverse pro-
III. ANALYSIS file ~perpendicular to the slit! along the slit length. For ex-
ample, the maxima in the plot represent the points when the
All images acquired by the Kodak reader were sent to a
slit’s central axis passes directly through the pixel center and
network workstation ~Kodak, PDS! and subsequently to the
the minima occurs when the axis passes exactly between two
research computer by an ethernet connection. The raw data
pixel centers. The angle of the slit was determined by the
for each image were first converted to screen exposure val-
distance in pixels between two minima that were chosen at
ues by the following relationships:
each end of the slit image.
E ~ x,y ! calc510@~ pixel2 value2C ! /1000# , Based on the slit’s angle, the orthogonal distance between
the slit’s central axis and each pixel sampling site was deter-
log10@ E ~ x,y ! corrected# 5cal 01 ~ cal 1 ! •log10@ E ~ x,y ! calc# .
mined. Projecting these pixel values along a single transverse
The constant C is 2000 for the GP-25 screen and 1700 for axis produced a finely sampled composite LSF. To provide
the HR screen. The calibration factors, cal0 and cal1, were uniform sampling along this axis, the LSF was resampled by
used to correct the calculated exposure values. They were linear interpolation where the effective sampling distance
determined from a log plot of measured exposure ~based on was the product of the pixel width and the sine of the slit’s
ionization chamber measurements! versus calculated expo- angle. The normalized LSF was then extended down to the
sure from the plate. Table III lists the values determined for 0.01% level by exponentially extrapolating past the 1% level
these correction factors. The analysis methods described be- as explained by Fujita et al.6 Figure 3 displays an example of
low are the same as those employed by Dobbins et al. This the initial and extrapolated line spread functions. These
was done to maximize the comparability of results. A more curves were normalized to 1.0 at the maximum. Choosing a
complete description of the analysis methods can be found in range of values on each LSF to fit to and extrapolate is a bit
the previous report.4 subjective. To ensure that the same extrapolation criteria

Medical Physics, Vol. 26, No. 1, January 1999


30 Bradford, Peppler, and Dobbins III: Characteristic of a radiography system 30

FIG. 3. Initial and extrapolated composite line spread functions in the


fastscan direction for the HR Kodak plate ~100 mm/pixel!.
FIG. 4. Presampling MTF curves in the fastscan and slowscan directions for
the Kodak GP-25 ~100 mm/pixel and 172 mm/pixel! and Kodak HR ~100
were used in this and the Fuji studies, one of the slit images mm/pixel! plates.
taken with the Kodak system ~172 mm/pixel, GP-25 plate,
fastscan direction! was re-evaluated by Dobbins. A compari-
son of the resulting MTF curves show differences of less NEQ~ f !
than 1.5%. DQE~ f ! 5 ,
SNR2ideal
The Fourier transform ~FT! of the extrapolated LSF pro-
duced the presampling MTF. A factor to correct for the where
transmission through the beveled edges was determined by EMTF2~ f ! • ~ large area signal! 2
comparing the MTF curves produced by the beveled and NEQ~ f ! 5
straight-edged slits. After this correction, the result was then NPS~ f !
normalized to 1.0 at the zero frequency value. The digital EMTF2
MTF, which includes the effects of the sampling interval, 5 .
NPS~ f ! normalized
was obtained by convolving the corrected presampling MTF
with a comb function in the frequency domain.10 Finally the The ideal SNR represents the signal-to-noise ratio of the in-
EMTF was computed by averaging the digital MTF over all cident x-ray beam. We have attempted to obtain the same
phases of the sampling comb relative to the slit. x-ray spectrum as used by Dobbins et al.4 by using the same
To reduce the uncertainty in the MTF measurements, the model of x-ray tube, kV, and filtration in all of our measure-
LSF and MTF calculations were repeated three times along ments. Therefore, the same ideal SNR2 value (2.7
three different segments of the slit ~along three profile plot 3105/@ mm2 mR# ) was used in this study. ‘‘Large area sig-
cycles!. The average of these three curves was then applied nal’’ represents the mean exposure value ~mR! calculated
in subsequent calculations. from the 64 regions of the flat field image. The normalized
NPS is the NPS divided by the square of the ‘‘large area
B. NPS analysis signal.’’

The central 102431024 region of each flat field image


was divided into sixty-four 1283128 sub-regions. Within IV. RESULTS
each sub-region the data were first fitted to a planar ramp A. MTF, NPS, and DQE results
function that was subsequently subtracted from the corre-
sponding data to remove any low-frequency nonuniformity Figure 4 compares the measured presampling MTF curves
effects ~e.g., heel effect!. An ensemble average of the mean- ~the average of three segments! for the three Kodak plate
square Fourier amplitudes for all 64 regions was then com- configurations studied. For each plate configuration, the
puted and used to determine the resulting two-dimensional MTF curves corresponding to the slowscan direction were
~2D! NPS. The one-dimensional NPS curves presented in higher that the fastscan direction. At all frequencies, the HR
this paper ~see results section! represent an average of eight plate out performed the GP-25 plate for both scan conditions.
slices through the 2D NPS data ~four slices on each side of At frequencies greater than three cycles/mm, the MTF curves
the central axis!.4 To ensure the accuracy of these results, the for the HR plate are approximately twice as large as the MTF
integral of each 2D NPS was compared to the variance of its curves for the GP-25 plate. Among the two GP-25 scan con-
pixel values and confirmed to be identical. ditions studied, the 100 mm/pixel sampling performed better
than the 172 mm/pixel sampling in the fastscan direction
~;20% greater at 2.6 cycles/mm! and worse in the slowscan
C. DQE analysis
direction ~;10% lower at 2.6 cycles/mm!.
The DQE of the image detector was calculated by the Figure 5 displays the NPS profiles in each direction for
following equations: the three Kodak configurations studied. Similar to the MTF

Medical Physics, Vol. 26, No. 1, January 1999


31 Bradford, Peppler, and Dobbins III: Characteristic of a radiography system 31

FIG. 6. Two-dimensional NPS for the Kodak GP-25 ~100 mm/pixel! plate at
~a! 0.03, ~b! 0.3, ~c! 3, and ~d! 30 mR. ~Scaling for each image has been
independently adjusted.!

relatively flat except for a few peaks that are most prominent
at the lowest exposure levels ~i.e., the 0.03 mR images!.
Because the noise peaks are most prominent at the low ex-
posure levels, they most likely represent a fixed pickup
noise, perhaps from a switching power supply that is used in
the unit. The KESPR reader electronics operate with the
same digitization rate ~i.e., the same analog bandwidth! for
both pixel sizes and adjust the plate and laser scan speeds to
achieve the different size pixels. Therefore, peaks that appear
at the same relative position in both figures ~representing
different frequency scales! correspond to the same analog
frequency. Several of the peaks appear at similar positions,

FIG. 5. Normalized NPS in the fastscan and slowscan directions for the ~a!
Kodak GP-25 ~172 mm/pixel! plate, ~b! Kodak GP-25 ~100 mm/pixel! plate,
and ~c! Kodak HR ~100 mm/pixel! plate.

plots, the NPS curves also show a difference between the


two directions. For all plates studied, there was more noise in
the slowscan direction at frequencies greater than 1 cycle/
mm. Figures 6 and 7 display the two-dimensional NPS at the
four exposure levels for the GP-25 plate with 100 mm/pixel
and 172 mm/pixel sampling, respectively. ~The characteris-
tics of the 2D NPS for the HR plate are very similar to the
GP-25 100 mm/pixel plate.! The contrast scaling on each of
FIG. 7. Two-dimensional NPS for the Kodak GP-25 ~172 mm/pixel! plate at
these images have been independently adjusted in order to ~a! 0.03, ~b! 0.3, ~c! 3, and ~d! 30 mR. ~Scaling for each image has been
best display the NPS structure. The noise power spectra are independently adjusted.!

Medical Physics, Vol. 26, No. 1, January 1999


32 Bradford, Peppler, and Dobbins III: Characteristic of a radiography system 32

Kodak plate configurations and four exposure levels. Over-


all, the DQE curves decrease as exposure level increases.
However, the curves for the standard resolution plate ~GP-
25! decrease much more quickly compared to the high reso-
lution ~HR! plate @Fig. 9~d! vs Fig. 9~a!#. At 0.03 mR, the
100 mm/pixel GP-25 plate outperforms the other plates at
lower frequencies and is equivalent to the HR plate at higher
frequencies ~.2.7 cycles/mm!. At 30 mR, the HR plate is
comparable to the 100 mm/pixel GP-25 plate at the lowest
frequencies but significantly better at all higher frequencies
~.1.2 cycles/mm!.

B. Effects from quantum and system noise


In Fig. 5~a! ~standard resolution plate, 172 mm/pixel! the
0.03 and 0.3 mR NPS lines for the Kodak-400 system differ
approximately by a factor of 10. At this low exposure the
noise power is inversely proportional to the exposure level.
For the other two Kodak configurations studied, the plots
display a factor smaller than ten between the corresponding
0.03 and 0.3 mR NPS curves. This suggests that at very low
exposures ~,0.3 mR!, quantum noise dominates for the 172
mm/pixel GP-25 Kodak plate while system noise is more
prevalent for both 100 mm/pixel GP-25 and HR plates. The
differences between the 0.3 and 3 mR, and 3 and 30 mR set
of NPS curves are reduced, indicating that system noise is a
more dominant factor at the higher exposure levels.
As expected the DQE curves at 0.03 and 0.3 mR for the
standard resolution Kodak plate ~172 mm/pixel! are very
similar. This is due to the dominant role of quantum noise
for this plate at the low exposure levels @see Fig. 8~a!#. As
explained above, the NPS values at these two exposure lev-
els have a factor of 10 difference. During the DQE analysis,
this factor is canceled by the factor of 10 difference in the
exposure levels. The result is similar DQE curves at 0.03 and
0.3 mR. The additional system noise in all other plates stud-
ied prevents this similarity.

C. Comparison of slit type


To ensure that our results were independent of slit type
the MTF for the high resolution ~HR! plate was measured
FIG. 8. DQE curves in the fastscan and slowscan directions for the ~a! twice in each direction, once using the beveled slit and once
Kodak GP-25 ~172 mm/pixel! plate, ~b! Kodak GP-25 ~100 mm/pixel! plate, using a straight-edged slit. Figure 10 provides an example of
and ~c! Kodak HR ~100 mm/pixel! plate. the results produced. This figure compares the resulting pre-
sampling MTF for both slits in the slowscan direction using
the HR ~100 mm/pixel! plate. At frequencies less than 2
which supports the hypothesis that the noise is from pickup cycles/mm, the difference between the ‘‘straight edge’’ MTF
in the electronics. The peaks are well confined and do not data and the uncorrected ‘‘beveled edge’’ MTF data was less
likely contribute to the noise power outside their immediate than 2.5%. However, at higher frequencies this difference
neighborhood. As explained in the analysis section, the NPS quickly increased and was as large as 12.5% at the Nyquist
profiles represent an average of eight slices through the 2D limit ~5 cycles/mm!. A plot of the difference as a function of
NPS data. Since these eight slices were chosen close to the frequency qualitatively resembled a sinc function. Therefore
central axis, the noise peaks that are in the corners of the 2D we chose to model the correction factor as sinc(W•u),
NPS images do not appear on the corresponding one- where ‘‘u’’ is the frequency in cycles/mm and W represents
dimensional NPS profile while the peaks along the axis do. an effective slit width. We found the minimum error for W
Figure 8 displays the DQE curves in each direction for the 552 mm. After the correction the differences in MTF were
three Kodak configurations studied. Figure 9 compares the less than 0.8% for frequencies below 2 cycles/mm and
average DQE curves ~average of each direction! for the three slightly higher ~1.75%! for the larger frequencies.

Medical Physics, Vol. 26, No. 1, January 1999


33 Bradford, Peppler, and Dobbins III: Characteristic of a radiography system 33

FIG. 9. These four plots compare the


resulting average DQE curves for each
Kodak plate studied at ~a! 0.03, ~b! 0.3,
~c! 3, and ~d! 30 mR.

V. DISCUSSION and DQE ~there were a few minor modifications to accom-


modate differences in the transformation to exposure values
A. Comparison of Kodak and Fuji CR systems
and differences in image dimensions!.
In order to compare the results presented in this paper Figure 11 compares the measured presampling MTF
with the results presented by Dobbins et al.4 on the Fuji- curves ~the average of three segments! for the Kodak-400
7000 CR system, all measurement and analysis methods per- system with Dobbins’ results for the Fuji-7000 system. ~For
formed in this study were the same as used by Dobbins. the PCR model 7000 reader, Dobbins et al.4 had measured a
Except for the slit type, no changes or refinements have been variety of plate models for each pixel size and resolution
made. We attempted to create the same x-ray spectra by type. For this comparison we chose only the latest plate mod-
applying the same model x-ray tube/generator kVp, and fil- els in his study that corresponded the closest to the pixel
tration. Of course there may be some slight differences in the sizes of the Kodak plates in this study.! Similar to the Kodak
x-ray spectra since different equipment was used and the plates, the Fuji MTF curves corresponding to the slowscan
x-ray tubes may have had different usage histories. It is not direction were higher than the fastscan direction. However,
clear how much this would affect results. The same com- there was a much larger difference between the slowscan and
puter programs were used for the computation of MTF, NPS, fastscan MTF curves for the Kodak plates ~e.g., at 4
cycles/mm for the high resolution 100 mm/pixel plate, the
Kodak fastscan MTF curve was 39% lower than its slowscan
curve as compared to a difference of 9.3% for the Fuji MTF
curves!. At larger frequencies ~.2 cycles/mm!, the Kodak
MTF curves were higher than the corresponding Fuji curves
in the slowscan direction and lower in the fastscan direction.
The average values at higher frequencies were slightly
greater for Kodak than for Fuji. The differences between the
average MTF curves ~average of the two directions! of each
CR system were less than 12% for both the large ~172 mm/
pixel Kodak and 200 mm/pixel Fuji! standard resolution
and 100 mm/pixel ~Fuji and Kodak! high resolution plates.
As for the 100 mm/pixel standard resolution plates, the two
systems produced average MTF curves within 10% for fre-
FIG. 10. An example of the effect of the sinc correction on the beveled slit quencies less than 3.3 cycles/mm. At the larger frequencies
presampling MTF data. ‘‘Uncorrected premtf’’ and ‘‘corrected premtf’’ re- the average Kodak MTF curve is 20% larger ~the absolute
fer to the beveled slit presampling MTF data before and after the sinc
difference is less than 0.04!.
correction, respectively. ‘‘Straight-edge premtf’’ refers to the straight edged
slit presampling MTF data. The above data were measured with the HR Figure 12 compares the NPS profiles between the Kodak-
Kodak plate ~100 mm/pixel! and in the slowscan direction. 400 and Fuji-7000 systems. Each curve in these plots is an

Medical Physics, Vol. 26, No. 1, January 1999


34 Bradford, Peppler, and Dobbins III: Characteristic of a radiography system 34

FIG. 11. Comparison of average presampling MTF curves for the Kodak-
400 and Fuji-7000 readers and the ~a! type V ST ~200 mm/pixel! Fuji and FIG. 12. Comparison of average NPS profiles at 0.03, 0.3, 3, and 30 mR for
GP-25 ~172 mm/pixel! Kodak plates, ~b! type IIIN ST ~100 mm/pixel! Fuji the Kodak-400 and Fuji-7000 readers and the ~a! type V ST ~200 mm/pixel!
and GP-25 ~100 mm/pixel! Kodak plates, and ~c! type V HR ~100 mm/pixel! Fuji and GP-25 ~172 mm/pixel! Kodak plates, ~b! type IIIN ST ~100 mm/
Fuji and HR ~100 mm/pixel! Kodak plates. pixel! Fuji and GP-25 ~100 mm/pixel! Kodak plates, and ~c! type V HR ~100
mm/pixel! Fuji and HR ~100 mm/pixel! Kodak plates. ~NPS data at 0.03 mR
for the Fuji 7000 reader and type V HR plate were not available.!

average of the NPS profiles for the two directions. Generally, trend. The leveling off of Kodak values at higher frequencies
the Kodak NPS curves are higher than the corresponding is also not as prevalent for the larger plates.
Fuji curves for each plate type and exposure level studied. The DQE curves for both systems are compared in Fig.
For all plates studied, differences in noise power between the 13. Each curve represents an average of the two directions.
two systems tend to increase with increasing exposure level As expected from the NPS calculations, differences in DQE
and frequency. For 100 mm pixel sampling, differences be- between the two systems tend to increase with increasing
tween the Kodak and Fuji NPS curves ~both standard and exposure. The exception occurs at 0.3 mR where the DQE
high resolution plates! remain relatively constant up to 2.5–3 curves for each system are a bit closer than the 0.03 mR
cycles/mm @see Figs. 12~b! and 12~c!#. In this frequency curves. The Kodak DQE curves are ;30%–40% lower at
range the Kodak NPS curves are about 20%–40% higher at 0.03 mR, 20%–35% lower at 0.3 mR, 30%–50% lower at 3
0.3 mR, 40%–70% higher at 3 mR and 100%–170% higher mR and 50%–70% lower at 30 mR.
at 30 mR. At higher frequencies the Kodak curves begin to
level out while the corresponding Fuji curves continue to
B. General discussion
drop. For the larger sampling distance ~172 and 200 mm
pixels! the Kodak NPS curves are again higher and a similar A very wide range of exposures was used for these ex-
trend of greater difference with increasing exposure is seen periments. This was done to test the system at very high and
@see Fig. 12~a!#. Only the 0.3 mR curves deviate from this very low exposure levels. In routine clinical practice, expo-

Medical Physics, Vol. 26, No. 1, January 1999


35 Bradford, Peppler, and Dobbins III: Characteristic of a radiography system 35

had larger differences in the two scan directions. All the


results presented in this report were obtained with the up-
graded filter.
For the measurements on the standard resolution plate the
same 35343 cm plate was used for both the large ~172 mm!
and small ~100 mm! pixel sizes. For the small pixel size this
was accomplished by indicating that the system should scan
only a portion of the plate since the system is not capable of
performing a small pixel scan of an entire 35343 cm plate.
Therefore, differences in the results for the two pixel sizes
are not due to any differences in the plate itself. The fact that
the DQE results for the small pixel size are better than the
large pixel size indicates that small pixel scanning of the
entire 35343 cm plate would provide some benefit, albeit
with a significant increase in the image data size and in-
creased scan time. Alternatively, these results suggest that
there is room for improvement in the acquisition of the large
pixel data.
The NPS curves show very similar trends between Kodak
and Fuji, except that the Kodak curves flatten out at higher
frequencies whereas the Fuji NPS curves continue to drop. In
fact, in Figs. 12~b! and 12~c!, the Kodak and Fuji NPS~0!
values are almost the same at low exposure. That fact,
coupled with the fact that the Kodak curves flatten out at
high frequencies, implies that the difference between the two
systems appears to be a higher background white noise
source in the Kodak. This would account for the flattening
out at high frequency, and would also account for why the
higher exposure NPS values differ more than the low expo-
sure curves. Such an additive white noise component could
be the result of system noise, for example.
It was not the intention of this study to measure the meth-
odological uncertainty of the measurement methods pre-
sented. However, duplicate flat field and slit images were
acquired on different 35 cm343 cm GP-25 plates ~for both
FIG. 13. Comparison of average DQE at 0.03, 0.3, 3, and 30 mR for the the 172 mm/pixel and 100 mm/pixel reader settings!. A com-
Kodak-400 and Fuji-7000 readers and the ~a! type V ST ~200 mm/pixel! Fuji parison among the duplicate MTF curves show average per-
and GP-25 ~172 mm/pixel! Kodak plates, ~b! type IIIN ST ~100 mm/pixel!
cent differences of 2.4%62.4% for the 172 mm/pixel GP-25
Fuji and GP-25 ~100 mm/pixel! Kodak plates, and ~c! type V HR ~100
mm/pixel! Fuji and HR ~100 mm/pixel! Kodak plates. ~DQE data at 0.03 mR plate and 0.5%61.4% for the 100 mm/pixel GP-25 plate.
for the Fuji 7000 reader and type V HR ~100 mm/pixel! plate were not Repeat measurements of NPS were made using different
available.! GP-25 plates for the 35 cm343 cm and 18 cm324 cm sizes.
The integral NPS showed average percent differences of
9.5%, 7.3%, 4.2%, and 19% for 0.03, 0.3, 3, and 30 mR
sures would be expected to be in the range of the middle two exposures, respectively. We feel that these difference values
exposures utilized for this report ~0.3–3 mR!. In this more overestimate the methodological uncertainty since they in-
clinically relevant exposure range the differences in the clude plate variability.
DQE’s for the two systems were somewhat reduced. The A previous report7 has also compared the performance of
DQE for the Kodak system was 20%–35% lower at 0.3 mR the Kodak and Fuji computed radiography systems. There
and 30%–50% lower at 3 mR. are several ways in which the measurements and methodol-
The difference in system response for the two scan direc- ogy used in the previous report are different from those in
tions was significantly larger for the Kodak system than the this report. The previous report used a range of x-ray tech-
Fuji system. Our early measurements demonstrated an even niques ~70–150 kVp! and only one exposure value ~0.3 mR!
larger difference for the Kodak system than what was pre- compared with the present report which used one kVp ~70!
sented in this report. We discussed our early findings with and a range of exposure values. The previous report com-
Kodak and obtained an upgrade to the antialiasing filter. Af- pared the Kodak Digital Science CR System 400 ~also
ter the installation of the new filter the difference in the two known as the Kodak Ektascan model 400! to the Fuji
scan directions was significantly reduced, but remains larger 9501-HQ system ~as opposed to the PCR 7000 system! and
than for the Fuji system. Earlier versions of the Fuji system used Fuji type-Va plates instead of type V plates that were

Medical Physics, Vol. 26, No. 1, January 1999


36 Bradford, Peppler, and Dobbins III: Characteristic of a radiography system 36

compared in this report. Lastly, slightly different methodolo- cies. At lower frequencies the 100 mm/pixel GP-25 plate is
gies were used to measure NPS and MTF. In spite of these much better. At all exposure levels and frequencies, the 100
differences, there are some areas where the results can be mm/pixel GP-25 plate outperformed the 172 mm/pixel GP-25
compared. The MTFs measured in both studies for standard plate.
resolution plates ~both Fuji and Kodak! were very similar. These measurements have shown the Kodak-400 and
We were unable to compare the large pixel size in the laser Fuji-7000 systems to be comparable in resolution perfor-
scan direction and high resolution plates ~both Fuji and mance. Noise power studies have shown differences among
Kodak! since that data was not available in the previous re- the two systems. At low exposure levels, the Kodak images
port. Also, NPS values could not be compared since data for contain ;20% more noise power and 40%–70% more at
similar measurements was not available. We found that our higher exposure levels. The DQE for the Kodak plates is
measurements of DQE for the Kodak system were ;20% from 20%–50% lower than the corresponding Fuji plates in a
lower at all frequencies than those of Samei and Flynn7 for clinically relevant exposure range ~0.3–3.0 mR!.
both the standard and high resolution plates. This difference Measurements have shown that a beveled slit can be used
may be partially explained by a 3.5% difference in the value to measure the MTF of a system. A simple correction factor
used for the square of the ideal signal-to-noise ratio (2.61 can be applied to account for transmission through the bev-
3105 vs 2.73105 ). It is not clear to what the remaining eled jaws. The primary advantage of the beveled slit is that
difference can be attributed. the relief edges allow alignment of the slit to be a signifi-
The MTF values for two Fuji systems using standard cantly less critical issue. For the 4° relief angle of the jaws
resolution plates, measured by both Samei and Flynn,7 and the focal spot must only be positioned within 64°. At an
Dobbins et al.,4 were quite comparable. For the DQE results SSD of 1.8 meters that is 612 cm, a tolerance that was easily
there is only one curve where the two papers have reasonably accommodated by eye without the need for any alignment
comparable data. All the other curves are for different expo- mechanism. Since we also had a slit fabricated with straight
sures, kVps, or plates and readers. Comparing the PCR 7000 ~nonbeveled! jaws to compare our results, we were able to
reader from Dobbins et al. ~with ST-V plates, 200 micron find an accurate correction factor.
sampling and 0.3 mR exposure, and with both scan direc-
tions averaged! with the FCR 9501-HQ reader from Samei
and Flynn ~with ST-Va plates, 100 micron sampling, 0.3 mR
exposure, and with results for the plate scan direction only!
ACKNOWLEDGMENTS
shows approximately a 5% difference at low frequency up to
almost 20% difference at high frequencies. Since the differ- This work was supported in part by Grant No. 1 RO1
ences in the two results have a frequency dependence, meth- CA58521 from the National Cancer Institute and Grant No.
odological differences may have some effect. Since the DAMD17-94-J-4055 from the Department of the Army. The
present report used the same methodology as Dobbins et al., contents of this publication are solely the responsibility of
the methodology may account for some of the differences the authors and do not necessarily represent the official
between the present results and those of Samei and Flynn. views of the National Cancer Institute or the Department of
To correct for transmission through the beveled edges, we the Army. We would like to thank Eastman Kodak Co. for
chose to apply a sinc function correction factor to each MTF their help and the use of a 18 cm324 cm HR plate. We
curve. As shown in Fig. 10 this correction worked very well. would also like to thank the University of Wisconsin, Ac-
A more accurate description of the transmission profile, credited Dosimetry Calibration Laboratory for the use of an
through the slit, would presumably provide a more accurate ion chamber and electrometer.
correction. However, considering the small magnitude of the
MTF and DQE values involved at the higher frequencies, an
error of less than 1.75% should be accurate enough for most
situations. For the larger plate ~172 mm pixel! the correction 1
J. C. Wandtke, ‘‘Bedside chest radiography,’’ Radiology 190, 1–10
factor results in an error less than 0.8% up to the Nyquist ~1994!.
limit ~;3 cycles/mm!. The correction factor determined for 2
K. Aoki, ‘‘Computed Radiography in Pediatrics,’’ J. Digital Imaging 8,
these experiments is only applicable for this energy spec- 92–96 ~1995!.
3
trum. A different factor would be needed for different spectra F. E. Lindhardt, ‘‘Clinical experiences with computed radiography,’’ Eur.
J. Radiol. 22, 175–85 ~1996!.
since higher energy beams would penetrate the beveled jaws 4
J. T. Dobbins III, D. L. Ergun, L. Rutz, D. A. Hinshaw, H. Blume, and D.
of the slit to a greater extent and lower energy beams to a C. Clark, ‘‘DQE( f ) of four generations of computed radiography acqui-
lesser extent. sition devices,’’ Med. Phys. 22, 1581–1593 ~1995!.
5
W. Hillen, U. Schiebel, and T. Zaengel, ‘‘Imaging performance of a digi-
tal storage phosphor system,’’ Med. Phys. 14, 744–751 ~1987!.
6
VI. CONCLUSION H. Fujita, D.-Y. Tsai, T. Itoh, K. Doi, J. Morishita, K. Ueda, and A.
Ohtsuka, ‘‘A simple method for determining the modulation transfer
Among the three plates studied for the Kodak system, the function in digital radiography,’’ IEEE Trans. Med. Imaging 11, 34–39
~1992!.
100 mm/pixel HR plate outperformed both GP-25 plates at 7
E. Samei and M. J. Flynn, ‘‘Physical Measures of Image Quality in Pho-
the higher frequencies and exposure levels. At low expo- tostimulable Phosphor Radiographic Systems,’’ SPIE Proc. 3032, 328–
sures, all three plates are comparable at the higher frequen- 338 ~1997!.

Medical Physics, Vol. 26, No. 1, January 1999


37 Bradford, Peppler, and Dobbins III: Characteristic of a radiography system 37

8
T. Bogucki, D. P. Trauernicht, and T. E. Kocher, Characteristics of a tative radiographic imaging using a photostimulable phosphor system,’’
storage phosphor system for medical imaging, Technical and Scientific Med. Phys. 17, 454–459 ~1990!.
Monograph No. 6, Publication No. N-331 ~Health Sciences Division, 10
J. T. Dobbins III, ‘‘Effects of undersampling on the proper interpretation
Eastman Kodak C., Rochester, NY, 1995!. of modulation transfer function, noise power spectra, and noise equivalent
9
C. E. Floyd, H. G. Chotas, J. T. Dobbins III, and C. E. Ravin, ‘‘Quanti- quanta of digital imaging systems,’’ Med. Phys. 22, 171–181 ~1995!.

Medical Physics, Vol. 26, No. 1, January 1999

You might also like