Application of QC_DR Software for Acceptance Testing and Routine

Quality Control of Direct Digital Radiography Systems: Initial

Experiences using the Italian Association of Physicist in Medicine
Quality Control Protocol
Andrea Nitrosi,1 Marco Bertolini,1 Giovanni Borasi,1 Andrea Botti,1 Adriana Barani,1 Stefano Rivetti,2
and Luisa Pierotti3

Ideally, medical x-ray imaging systems should be QA protocol. The intrinsic nature of digital images
designed to deliver maximum image quality at an
produced by DDR systems allows the “online” use
acceptable radiation risk to the patient. Quality assur-
ance procedures are employed to ensure that these
of automated quality control software.
standards are maintained. A quality control protocol for
direct digital radiography (DDR) systems is described METHODS AND MATERIALS
and discussed. Software to automatically process and
analyze the required images was developed. In this
paper, the initial results obtained on equipment of In order to establish a generally acceptable
different DDR manufacturers were reported. The proto- baseline performance of the systems, a total of 14
col was developed to highlight even small discrepancies DDR systems from four manufacturers were
in standard operating performance. periodically tested and their results compared in
KEY WORDS: Quality control, direct digital radiography, the framework of the Digital Quality Assurance
automatic software, quality assurance, automated Task Group of the Italian Association of Physics in
measurement, image quality analysis Medicine (AIFM; Table 1). All the tests were done
in a clinical environment.
In order to verify the temporal trend of the state
INTRODUCTION of calibration, the protocol was repeated periodi-
cally on the same system (a Philips Digital
Diagnost): the first evaluation just after the initial
I n recent years, radiographic systems have
undergone the “digital revolution,” but the
quality control protocols for digital imaging
calibration and after 1, 2, 3, and 4 weeks (for this

1
equipment still remain in a quite early stage of From the Servizio di Fisica Sanitaria, Azienda Ospedaliera
development. In principle, the implementation of a Santa Maria Nuova, Reggio Emilia, Italy.
2
From the Servizio di Fisica Sanitaria, AUSL, Modena,
quality assurance (QA) program for direct digital
Italy.
radiography (DDR) equipment could be derived 3
From the Servizio di Fisica Sanitaria, AOU Sant’Orsola-
from existing national and international stand- Malpighi, Bologna, Italy.
ards.1–6 The execution of a quality assurance Correspondence to: Andrea Nitrosi, Servizio di Fisica
protocol in a clinical environment should be rapid, Sanitaria, Azienda Ospedaliera Santa Maria Nuova, Reggio
and the analysis and the verification of the results Emilia, Italy; tel: +39-0522-296076; fax: +39-0522-296392;
e-mail: [email protected]
should be obtained in very short time. In addition,
Copyright * 2008 by Society for Imaging Informatics in
each department could have different equipment Medicine
(manufacturer, software release, etc.). These Online publication 3 September 2008
aspects require a highly flexible and automated doi: 10.1007/s10278-008-9150-z

656 Journal of Digital Imaging, Vol 22, No 6 (December), 2009: pp 656Y666

QC_DR SOFTWARE FOR ACCEPTANCE TESTING AND ROUTINE QUALITY CONTROL 657

Table 1. List of DDRs Tested and Calibration Conditions

Manufacturer and product name No. of systems No. of controls Calibration conditions

Philips Digital Diagnost 6 24 SID 110/150 cm; 70 kVp; 21 mm Ala; no grid

GE Definium 4 16 SID 180 cm; 70 kVp; 20 mm Ala; no grid
Siemens Axiom Aristos FX 2 8 SID 180 cm; 70 kVp; 0.5 mm Cu+1 mm Ala; no grid
Kodak DR7500 2 4 SID 110/183 cm; 70 kVp; 0.5 mm Cu+1 mm Ala; no grid
a
Intended as additional x-ray tube filtration

system, the manufacturer requires a monthly Test Objects Used

calibration).
The image information acquired from a DDR MTF and Lag Tool
system is not immediately suitable for image
display. The images must be corrected in order to The test device for the determination of the
remove some artifacts and to correct their appear- modulation transfer function (MTF) and the
ance to obtain a diagnostic radiogram.7 magnitude of lag effects is described by the IEC
All manufacturers strongly suggest a periodic 62220-1 and consists of a 1.0-mm-thick tungsten
repetition of detector calibration (“flat field”). The plate, 100 mm long and 75 mm wide.2
results are influenced by the filtration, the presence
of the grid, and the distance between the x-ray Low-Contrast Detectability Tool
source and the detector. This correction is very
critical and, if not carefully done, can contribute to The low-contrast Leeds phantoms and Artinis
image degradation. For this reason, the detector M.S. CDRAD low-contrast phantoms are used to
testing should be done under the same conditions create automated contrast–detail curves.
of the calibration stage as indicated by the
manufacturer. Tests Performed
The exposure level on the detector should be
chosen in the same range as in the clinical practice. The set of quality control procedures described
This exposure level is called the “normal level”2 are basically derived by the IEC standards and
and should be specified by the manufacturer should represent an “operative” protocol transla-
(otherwise estimated on the basis of clinical image tion in the clinical environment. The protocol
analysis). adopted proposes additional controls (e.g., unifor-
All the evaluations should be performed on the raw mity tests, defective pixel analysis, etc.) in order to
preprocessed images (digital images obtained after check clinical relevant images quality aspects.
flat field, dark noise, and defective pixel corrections, The complete lists of quality controls testing
but without any post processing filtering). parameters proposed are reported in Table 2. In the
For one manufacturer (Philips), it was impossi- second column of the table, the sections where they
ble to obtain, in clinical conditions, nonprepro- are more extensively described are reported, too.
cessed (unfiltered) images.
If the pixel values are not linear with respect to Conversion Function
air kerma, they should be linearized using the
proper response curve.8 Purpose To establish the relationship between
In this study, the images were acquired under detector dose and pixel value in a range compat-
various exposure conditions using the test objects ible with clinical conditions.2
indicated below.
The data processing software9 was developed Method A calibrated dosimeter is positioned
with IDL rel. 6.4 (ITT Visual Information So- approximately at half the source-to-detector dis-
lutions, Boulder, CO, USA); all the images were tance using correct acquisition geometry in order
evaluated using this software. This software, to minimize the backscattered radiation from
named QC_DR, is freely available at http://www. layers behind the detector. The detector surface
qcdr.org. exposure is then computed at a reference point
658 NITROSI ET AL.

Table 2. List of DDR Quality Controls at least 200×200 pixels positioned in the center of
Testing parameter Section in the article the detector and in each image sector.
The measured values should correspond to the
Conversion function Conversion function
offset (electronic noise signal) of the detector.
Dark image analysis Dark image
Nonuniformity Nonuniformity It is important to note the different approach
Noise analysis Noise adopted from the main flat panel detector (FPD)
Artifacts Defective pixels analysis manufacturers in order to correct the image offset
Spatial resolution Spatial resolution using the self-scanned nature of the detectors.
Low contrast Low-contrast detectability
For example, the offset correction for the Trixell
Lag Lag
detector is typically done automatically by mea-
suring the image response just before or after an
exposure, allowing the system to correct variations
in dark current with time and temperature.
applying the inverse square law and, if necessary,
detector position correction factor. Pass/Fail Criteria For each ROI, the ratio between
Acquire constant intensity x-ray exposure the mean values and the normal level should not
images across the full area of the detector (flat differ more than 2% in order to comply with the
image), using at least eight different tube loadings conversion function control limits.2,11
(in milliampere second).10 Choose the minimum
possible milliampere second loading in order to Nonuniformity
obtain the kerma level on the surface of the
detector corresponding to the minimum kerma Purpose To test the flat field correction by evaluat-
level achievable, and choose the maximum milli- ing a uniform flat image and the conversion function
ampere second so as to avoid any saturation effects consistency over the full area of the detector.
given by the detector manufacturer specifications.
Fit the mean pixel value in function of kerma by Method Set milliampere second in order to have an
a model function (specified by the manufacturer) incident kerma level on the surface of the detector
in order to obtain a linearized image with respect near equivalent to the normal exposure level.
to the air kerma. Acquire a full-field uniform image.
Evaluate the mean signal and its standard
Pass/Fail Criteria Evaluate accuracy and precision deviation (SD) in consecutive 3×3 cm square
of the conversion function: as stated in the IEC ROIs overlapped by 1.5 cm each, omitting the
62220-1 standard, the fit result has to fulfill a edge of the image within about 3 cm (equal to
“final R2 ≥0.99 and no individual experimental about the 2% of the source-to-image distance in
data point deviates from its corresponding fit result collimator light-to-x-ray field total misalignment
by more than 2% relatively.”2 along either the length or the width of the x-ray
field);12 in addition, in good practice boundaries,
Dark Image areas should not have any diagnostic significance.
The effective analyzed area must be at least the
Purpose To assess the intensity of intrinsic noise 80% of the whole detector area.
in the system. Assess the nonuniformity evaluation locally (or
differential) and globally for the signal and signal
Method Place a lead attenuator (of at least 2 mm to noise ratio (SNR) intensity as explained below.
thick) at the x-ray output source. Close the Local signal nonuniformity (LSNU) should be
collimators as much as possible. Set the kilo- evaluated as the mean signal intensity difference
voltage peak and milliampere second as low as between two consecutive ROIs (in each spatial
possible (e.g., 40 kVp and 0.5 mAs) direction):11 !



Acquire a constant intensity x-ray exposure max
ROIi;j
ROIi1;j1

images across the full area of the detector (“dark LSNU ¼ max ð1Þ
ROIi;j
image”). Evaluate the mean signal on the surface
of the detector in five regions of interest (ROI) of where ROIi;j is the mean value of the (i,j)-th ROI.
QC_DR SOFTWARE FOR ACCEPTANCE TESTING AND ROUTINE QUALITY CONTROL 659

Global signal nonuniformity (GSNU) should be Table 3. Reference Limits Proposed for Nonuniformity Tests
evaluated as the difference between the maximum Test Reference limit (%)
and the minimum mean signal intensity found in
Local signal nonuniformity G4
all the ROIs:
Global signal nonuniformity G8
    Local SNR nonuniformity G8
max ROIi;j
min ROIi;j
GSNU ¼      ð2Þ Global SNR nonuniformity G20
max ROIi;j þ min ROIi;j
2
where ROIi;j has the same definition given with respect to the simple difference in the local
previously. average difference, taking into account the error
Similarly, the local SNR nonuniformity propagation theory in the ratio of mean value and
(LSNRNU) should be evaluated as signal to noise SD (approximately 4 ɛROI).
intensity difference between two consecutive ROIs Analogously, the global maximum percentage
(in each spatial direction): signal and SNR nonuniformity errors were


0
 1 assessed to be 8% and 20%, respectively.

ROI

max


ROIi;j

ROIi1;j1


ROI
B ð i;j Þ ð i1;j1 Þ C
LSNRNU ¼ maxB
@
C
A Noise
ROIi;j

ðROIi;j Þ
Purpose To evaluate the system noise in terms of
ð3Þ
both electronic and spatial characteristics.
where ROIi;j is the mean value of the (i,j)-th ROI
and σ(ROIi,j) is the SD of the (i,j)-th ROI. Method Noise power spectrum (NPS): NPS has the
Global SNR nonuniformity (GSNRNU) should advantage of characterizing the spatial frequency
be evaluated as the difference between the maxi- content of image noise. The method for its
mum and the minimum SNR intensity found in all calculation is explained in the IEC 62220-1
the ROIs: standard.2

GSNRNU Relative standard deviation (RSD) analysis: The

    trend of the RSD versus the exposure can give a
ROI ROI
max
ROIi;j
min
ROIi;j useful insight into noise properties. For this
ð i;j Þ ð i;j Þ
¼     purpose, all the images should be converted into
ROIi;j ROIi;j
max
ROI þ min
ROI dose acquired for the conversion function test.10,11
ð i;j Þ ð i;j Þ 2
The RSD, which is the ratio between the SD and
ð4Þ the average signal value in a ROI of at least 400×
400 pixels positioned in the center of the images,
where ROIi;j and σ(ROIi,j) have the same defini- should be measured.
tions given previously. The trend of the mean square RSD versus the
exposure level should be fitted using the following
Pass/Fail Criteria The reference limits for signal function:
and SNR intensity nonuniformity evaluation are 
2  
TOT
reported in Table 3. RSD2 ¼ ¼ þþ 2 ð5Þ
The rationale of the proposed limits is derived x x x
by the error propagation theory. As stated in IEC where x is the x-ray exposure, α represents the
62220-1, “no individual experimental data point contribution of the quantum statistical noise, β is
deviates from its corresponding fit result by more the multiplicative contribute to noise related to a
than 2% relatively,” than assuming 2% the fixed-pattern noise unresolved by flat field correc-
maximum percentage error in the mean value for tion, and γ is the additive factor that can be
one ROI (ɛROI), the maximum percentage error in interpreted as the noise connected to electronics.13
the difference for two adjacent ROIs is two times
ɛROI, i.e., 4%. Local differences maximum per- Pass/Fail Criteria NPS There are no established
centage error in the SNR is assumed to be double criteria or limits for this test that should be
660 NITROSI ET AL.

performed in the commissioning/acceptance test, A defective detector element (“defective pixel”)

but it could be useful in order to evaluate any grid is an element of the ROI where the absolute value
suppression tool on the images if the calibration deviates more than 20% from the mean pixel value
of the detector has to be performed with the grid of the ROI elements.10 Count the total number of
inserted. If the calibration of the detector has defective detector elements (defective pixels).
to be performed with the grid, the presence of
discrete spikes at spatial frequencies corre- Pass/Fail Criteria After detector calibration, in
sponding to the interline spacing of the grid normal conditions, no defective detector element
should indicate an imperfect grid suppression. shall be counted.
Presence of discrete spikes in NPS should not be
observed with respect to the acceptance test, also Spatial Resolution
if blotches or small single-point artifacts do not
have enough power to demonstrate a measurable Purpose To measure the detector presampling
change in the NPS.14 MTF in order to ensure that the hardware is
RSD As for NPS, we do not propose any criteria performing properly and is not degrading the
or limits. α, β, and γ values should be registered resolution of the image below original equipment
and compared with the reference value fixed in performance levels. This test is important for
the acceptance testing control. This test should systems with moving parts in the image chain
give information about the DDR electronics (i.e., computed radiography or scanning sys-
functionality and about flat field correction tems);14 so in static DDRs systems, it could be
effectiveness. assessed only at the commissioning stage.

Method Follow the IEC 62220-1 requirements.2

Defective Pixels Analysis
Pass/Fail Criteria The IEC 62220-1 standard does
Purpose To check the accuracy of the defective not propose acceptability limits. The presampling
pixels/lines correction in the preprocessing stage.15 MTF should not be different from the minimum
Ask the manufacturer for a “defective pixel map,” reference values supplied from the manufacturer,
which specifies the dead pixels location, individual taking into account experimental error propagation.
pixels, pixel clusters, and lines of contiguous
pixels that fail to produce a useable output value.16 Low-Contrast Detectability
In early flat panel plates, defective pixels may
represent about 0.3% of the total number of pixels. Purpose To assess and control the ability of the
Improvements in panel manufacturing procedures system to demonstrate low-contrast object.
are steadily reducing this number.
Defective pixels are acceptable if they are not Method Place the phantom (N.A. CDRAD or Leeds
noticeable on the image, which usually occurs only phantoms) on the detector. Follow the instructions
if there are large clusters of defective pixels or they provided by the phantom manufacturer.
occur in multiple adjacent rows or columns. In the commissioning test, acquire at least six
Manufacturers have specifications for how many images moving the phantom slightly before each
defective pixels are acceptable and on how they exposure (to obtain images with different relative
are eventually clustered and, for this reason, position of the details on the detector elements) at
noncorrectable. If the defective pixel map is not appropriate milliampere second to deliver a dose
accessible, different ways to evaluate this are nearly equivalent to the normal exposure level (e.g.,
possible.7 2.5 μGy). Repeat acquisitions following the same
conditions described before changing milliampere
Method Acquire a flat, full-area image at a dose second in order to deliver about one fourth of the
near equivalent to the normal exposure level (e.g., normal exposure level and four times the normal
2.5 μGy on the detector); for this purpose, the exposure level (e.g., 0.6 and 10 μGy on the
image used before the nonuniformity test can be detector).2,17 In routine quality control test, acquire
utilized. Divide the image into 1×1 cm ROIs. one image at the normal exposure level.
QC_DR SOFTWARE FOR ACCEPTANCE TESTING AND ROUTINE QUALITY CONTROL 661

Table 4. Constancy DDR Quality Controls and Relative Frequency  a temporary change in the sensitivity of the
Test performed Frequency detector determined by prior exposure history
(multiplicative lag).
Conversion function Biannual
Dark image analysis Biannual
Nonuniformity Biannual Method Follow the IEC 62220-1 Annex A:
Noise components Annual
“Determination of lag effects.” Two tests are
Artifacts Biannual
Low contrast Biannual proposed: additive and multiplicative lag effects
Lag Biannual test.

Pass/Fail Criteria As stated by the IEC 62220-1

standard,2 the lag effects must contribute less than
0.5% of the effective exposure. In Table 4, the
tests and the frequency of constancy quality
controls are listed.10
Signal difference to noise (SDNR) is defined by:


QC_DR Software

ROIsignal
ROIbackground

SDNR ¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ð6Þ

2signal þ
2background QC_DR software, following the procedures
stated before, can automatically perform the image
where ROIsignal and ROIbackground are the mean analysis necessary to obtain the conversion function
values of the ROI within the target and of the and to evaluate the RSD, the dark image, the signal
background region near the target, respectively, and SNR nonuniformity, the defective pixels, and
and
2signal and
2background are the variances of the the lag effects. The analysis is performed based on
ROI within the target and of the background area DICOM or more generally raw images (in this latter
near the target, respectively. Measure the SDNR case, the user must fill the fields needed to open/
for each target group. elaborate the image(s), e.g., image type, width and

Pass/Fail Criteria There are no established criteria

or limits for this test that should be performed in
the commissioning/acceptance test. This test rep-
resents a baseline for future measurements and it
could allow an objective assessment of system
performance by comparison of current and histor-
ical data.
In the following routine quality control tests
(after detector calibration), no statistically signifi-
cant difference should be measured from the
baseline acceptance test.

Lag

Purpose To evaluate the severity of any artifact

due to previous exposure to the detector. This
effect derived from:
 a delay in generated signal being read that can
cause some of the signal generated in a prior
image to appear superimposed on the signal
generated during the current image acquisition Fig 1. Depiction of the QC_DR software graphical user
(additive lag); interface.
 Table 5. Results of Constancy Quality Control Tests (The Error Corresponds to ±1 SD)
662

Conversion function Dark image analysis Nonuniformity Artifacts Lag

2
Test performed R Max data point deviation (%) Mean signal (%) LSNU (%) GSNUS (%) LSNRNU (%) GSNRNU (%) No. of defective pixel(s) Additive Multiplicative

Results 0.999±0.001 1.69 0.9±0.6 0.7±0.2 3.6±2 4.6±1 15.5±2 0 0.05±0.04 0.13±0.08
Reference limit proposed 90.99 G2 G2 G4 G8 G8 G20 0 G0.5 G0.5

Table 6. Weekly Tested System Test Results (the Error Corresponds to ±1 SD)

Conversion function Dark image analysis Nonuniformity Artifacts Noise components Lag

Test performed R2 Max data point deviation (%) Mean signal (%) LSNU (%) GSNUS (%) LSNRNU (%) GSNRNU (%) No. of defective pixel(s) Quantum statistical noise (α) Multiplicative contribute (β) Additive factor (γ) Additive Multiplicative

After calibration 0.9999 0.09 0.9±0.3 0.26 0.74 4.21 12.07 0 2.5E-06 9.4E-05 8.6E-04 0.02 0.05
After 1 week 0.9999 0.19 0.9±0.3 0.93 1.49 4.41 12.01 0 2.6E-06 9.5E-05 8.5E-04 0.04 0.14
After 2 weeks 0.9998 0.45 1.2±0.3 0.36 1.43 4.95 13.72 0 2.5E-06 9.1E-05 8.6E-04 0.06 0.27
After 3 weeks 0.9999 0.32 1.4±0.3 0.32 1.46 4.4 13.71 0 2.6E-06 9.8E-05 8.9E-04 0.07 0.30
After 4 weeks 0.9999 0.30 1.1±0.3 0.44 1.27 4.05 14.08 0 2.7E-06 1.0E-04 9.1E-04 0.06 0.25
Reference limit 90.99 G2 G2 G4 G8 G8 G20 0 – – – G0.5 G0.5
NITROSI ET AL.
QC_DR SOFTWARE FOR ACCEPTANCE TESTING AND ROUTINE QUALITY CONTROL 663

LSNU (%) GSNUS (%) LSNRNU (%) GSNRNU (%) No. of defective pixel(s) Quantum statistical noise (α) Multiplicative contribute (β) Additive factor (γ) Additive Multiplicative

0.11
0.12
height, offset, little or big endian, pixel size, and so

G0.5
on). An output file can be saved in a text file
Lag

(common separated value format) for further anal-

0.035
0.034
G0.5 ysis or test documentation. A simplified user guide
is disposable on the website.9 In Fig. 1, a depiction
of the graphical user interface (GUI) of the QC_DR
1.4E-03
4.8E-04
–

software is presented.

RESULTS AND DISCUSSION

4.3E-04
1.3E-04
Noise components

All the MTF, NPS, and detective quantum

–

efficiency measured in the acceptance test of the

checked systems conform to the manufacturer’s
specification and were in good agreement with the
results published in the literature.8,18–24 The results
of quality control constancy tests on all systems,
2.9E-05
8.7E-06

considering the reference values and tolerances

–
Table 7. Test Results of the System Before and After Calibration

proposed, are listed in Table 5.

The proposed baseline performance limits for
constancy quality control tests on calibrated units are
derived from the IEC standard and are satisfied for
Artifacts

all the systems tested. Only the conversion function

54
0
0

limits on the maximum data point deviation were

close to the limit proposed, in particular those
referred to low-dose mean pixel value signals. This
10.24
12.49

was probably also due to the intrinsic dosimeter

G20

sensitivity.
The proposed baseline performance could rep-
3.22
5.90
Nonuniformity

resent the acceptable limit levels for all DDR

G8

systems. Achievable level limits, lower than the

baseline ones, should be adopted for each own
2.56
2.05

system on the basis of acceptance and routinely

G8

quality controls results.

0.56
0.77

As we mentioned before, on a Philips Digital

G4

Diagnost system for which the manufacturer

Dark image analysis

Mean signal (%)

specifications require at least a monthly flat

0.12
5.32

fielding correction, the tests were repeated also 1,

G2

2, 3, and 4 weeks after the calibration. In Table 6,

the data referred to this system (called “weekly
Max data point deviation (%)

tested system”) for which the tests were repeated

are reported.
Conversion function

About 90 days after last calibration

14.74

When visiting the center which was operating a

1.7
G2

Values in italics are out of limits

Siemens Axiom Aristos FX system, we found that

the technicians did not execute for quite a long
time (about 90 days) the routine (monthly)
0.9965
After calibration 0.9999

required calibration procedure.

Reference limit 90.99
R2

As could be foreseen, the system was found

severely miscalibrated. After a first set of evalua-
Not calibrateda
Test performed

tion, the system was properly calibrated. In Table 7,

the data referred to this system before and after the
proper calibration procedure are reported.
a
664 NITROSI ET AL.

Fig 2. Contrast–detail curves obtained, taking into account two systems from the same manufacturer (error bars are referred to
standard error).

Conversion Function

For a complete constancy quality test, the time All DDR systems widely pass the reference
required was about 15 min per DDR system and the limits proposed both in terms of the Pearson
utilization of the automated software allowed the correlation coefficient (R2) and of the maximum
complete analysis of the images in less than 10 min. percentage experimental data point deviation. For

Fig 3. Contrast–detail curves measured for the same system just after calibration and repeated each week before the following
manufactured requested calibration procedure.
QC_DR SOFTWARE FOR ACCEPTANCE TESTING AND ROUTINE QUALITY CONTROL 665

the weekly tested system, no significant difference for the uncalibrated system, there were 54 ele-
(considering the experimental data acquisition ments. All these defective pixels were corrected
error propagation) was found. subsequently to the detector calibration.
The uncalibrated system passed the reference
limit proposed for the correlation coefficient, but Low-Contrast Detectability
the maximum percentage experimental data point
deviation was out of the limit. The limit was not The test proposed seems to have a good
met for low doses. That should be expected on the sensitivity in order to assess detectability degrada-
basis of the dark noise test results reported in the tion. Also, the calibration frequency proposed in
following paragraph. clinical use by the manufacturer seems to be able
to avoid operating outside any limits.
Dark Image In Fig. 2, we report the contrast–detail curves25,26
obtained, taking into account two systems from the
Considering the experimental data acquisition same manufacturer, one of which was uncalibrated.
error propagation, no significant difference was For the first (uncalibrated) system, the curves are
found in any systems after calibration. The dark referred to as the precalibration and postcalibration
image analysis did not reveal appreciable differ- conditions; for the second one, the results are
ences among tests performed 1, 2, 3, and 4 weeks obtained just after calibration.
after the flat fielding correction, probably due to Performing a two-way analysis of variance test
the correct calibration timing. with a 5% confidence level, it was possible to
For the uncalibrated system, dark image analysis demonstrate that the difference between two cali-
revealed a significant difference on the mean brated systems is not statistically significant
signal that resulted in more than twice the limit (F(1,48)=0.11, p=0.74). The curves are statistically
proposed. That result is probably due to an different (F(1,48)=15.84, p=0.0002) between pre-
ineffective automatic dark noise correction. calibration and postcalibration conditions.
In Fig. 2, it is worth noting that the calibration
Nonuniformity seems to improve the detection of large diameter
target rather than small target detection that is
Considering the experimental data acquisition principally limited by system’s high-frequency
error propagation, no significant difference was MTF.27 The contrast–detail curves of the weekly
found in any systems after the calibration. Similar tested system are reported in Fig. 3. Differences
results were found for the weekly tested system between the curves are not statistically significant.
and for the uncalibrated one.
Lag
Noise
For all the systems checked after calibration,
Relative Standard Deviation additive and multiplicative lag effects were in
agreement with respect to the limit proposed by
Considering the experimental data acquisition the IEC 62220-1 requirement. The results for the
error propagation, no significant difference was weekly tested system were in agreement with the
found in any systems after calibration and for the limit proposed, but there is observed a weak
system for which the controls were repeated 1, 2, 3, increasing trend with time after the calibration.
and 4 weeks after the flat fielding correction. For the
uncalibrated system, significant differences were CONCLUSIONS
found for the noise components with an increase of
about a factor of three for each of the components. Quality control tests are of fundamental impor-
tance for keeping the systems properly operating.
Artifacts Analysis The proposed quality control tests for DDR are
quite complex and require an automated software
After detector calibration and in weekly tests, no for image analysis. When the systems were
defective detector elements were detected, while calibrated, all the tests passed within the reference
666 NITROSI ET AL.

limits. The uncalibrated system was widely outside 60601-1-3. Geneva, Switzerland: International Electrotechnical
these limits. We believe that a larger imaging trial Commission, 1994
13. Monnin P, Gutierrez D, Bulling S, Lepori D, Valley JF,
should be useful to refine the performance limits. Verdun FR: Performance comparison of an active matrix flat
The proposed protocol should provide an initial panel imager, computed radiography system, and a screen-film
framework for definition of a standardized quality system at four standard radiation qualities. Med Phys 32:343–
control program for DDRs. 350, 2005
14. Yaffe MJ, Bloomquist AK, Mawdsley GE, Pisano E,
Hendrick RE, Fajardo LL, Boone JM, Kanal K, Mahesh M,
ACKNOWLEDGMENTS Fleischman RC, Och J, Williams MB, Beideck DJ, Maidment
ADA: Quality control for digital mammography: part II
The authors gratefully acknowledge the collaboration of the recommendations from the ACRIN DMIST trial. Med Phys
Italian Association of Physics in Medicine (AIFM), especially 33:737–752, 2006
Dr. Giacomo Belli, chief of the AIFM Digital Radiography 15. Floyd CE Jr, Warp RJ, Dobbins JT 3rd, Chotas HG,
Workgroup. In addition, the authors would like to express their Baydush AH, Vargas-Voracek R, Ravin CE: Imaging character-
gratitude to Dr. Andy Smith and Dr. John Yorkston for their istics of an amorphous silicon flat-panel detector for digital
interest in this work. chest radiography. Radiology 218:683–688, 2001
16. Padgett R, Kotre CJ: Assessment of the effects of pixel
loss on image quality in direct digital radiography. Phys Med
REFERENCES Biol 49:977–986, 2004
17. Charnock P, Connolly PA, Hughes D, Moores BM:
1. Schreiner-Karoussou A: Review of image quality standards to Evaluation and testing of computed radiography systems.
control digital x-ray systems. Radiat Prot Dosim 117:23–25, 2006 Radiat Prot Dosim 114:201–207, 2005
2. International Electrotechnical Commission. Medical 18. Samei E, Flynn MJ, Chotas HG, Dobbins III, JT: DQE
electrical equipment: characteristics of digital x-ray imaging of direct and indirect digital radiographic system. Proc SPIE
devices—part 1: determination of the detective quantum 4320:189–197, 2001
efficiency. Document no. 62220-1. Geneva, Switzerland: 19. Samei E, Flynn MJ: An experimental comparison of
International Electrotechnical Commission, 2003 detector performance for direct and indirect digital radiography
3. The Kings Centre for the Assessment of Radiological Equip- systems. Med Phys 30:608–622, 2003
ment: Protocols for QA of DDR System, website: http://www.kcare. 20. Borasi G, Nitrosi A, Ferrari P, Tassoni D: On site
co.uk/content.php?page=protocols.htm&folder=Education evaluation of three flat panel detectors for digital radiography.
4. Faulkner K: The DIMOND project and its impact on Med Phys 30:1719–1731, 2003
radiation protection. Radiat Prot Dosim 3–6, 2005 21. Granfors PR, Aufrichtig R: Performance of a 41×41 cm2
5. Shepard SJ et al.: Quality Control in Diagnostic Radiol- amorphous silicon flat panel detector for radiographic imaging
ogy. AAPM Report 74, 2002 applications. Med Phys 27:1324–1331, 2000
6. Seibert JA: Acceptance Testing and Quality Control of 22. Illers H, Buhr E, Bergmann D, Hoeschen C: Measure-
Photostimulable Storage Phosphor Imaging Systems. AAPM ment of the detective quantum efficiency (DQE) of digital x-ray
Report 93, 2006 imaging devices according to the standard IEC 62220-1. Proc
7. Beutel J, Kundel HL, Van Metter RL: Handbook of SPIE 5368:17787, 2003
Medical Imaging—Volume 1. Physics and Psychophysics, 23. Borasi G, Samei E, Bertolini M, Nitrosi A, Tassoni D:
Washington: SPIE, 2000 Contrast–detail analysis of three flat panel detectors for digital
8. Samei E, Dobbins JT, Lo JY, Tornai MT: A framework radiography. Med Phys 33:1707–1719, 2006
for optimising the radiographic technique in digital x-ray 24. Neitzel U, Günther-Kohfahl S, Borasi G, Samei E:
imaging. Radiat Prot Dosim 114:220–229, 2005 Determination of the detective quantum efficiency of a digital
9. Bertolini M. QC_DR program. (http://www.qcdr.org) x-ray detector: comparison of three evaluations using a common
10. Perry N, Broeders M, de Wolf C, Törnberg S, Holland R, image data set. Med Phys 31:2205–2211, 2004
von Karsa L, Puthaar E: European Guidelines for Quality 25. Young KC, Cook JJH, Oduko JM, Bosmans H:
Assurance in Breast Cancer Screening and Diagnosis, 4th Comparison of software and human observers in reading images
edition. Luxembourg: Office for Official Publications of the of the CDMAM test object to assess digital mammography
European Communities, 2006 systems. Proc SPIE 7:614206.1–614206.13, 2006
11. Italian Association in Medical Physics (AIFM): Digital 26. Marshall NW: A comparison between objective and
Radiology Working Group. (http://www.fisicamedica.org) subjective image quality measurements for a full field digital
12. International Electrotechnical Commission. Medical mammography system. Phys Med Biol 51:2441–2463, 2006
electrical equipment: General Requirements for Safety—3. 27. Moy JP: Signal-to-noise ratio and spatial resolution in x-
Collateral Standard: General Requirements for Radiation ray electronic imagers: is the MTF a relevant parameter? Med
Protection in Diagnostic X-ray Equipment. Document no. Phys 27:86–93, 2000