Radiation Protection Dosimetry Advance Access published March 4, 2008 Radiation Protection Dosimetry (2008), pp. 1–5
doi:10.1093/rpd/ncn047
QUALITY ASSURANCE OF COMPUTED AND DIGITAL RADIOGRAPHY SYSTEMS C. Walsh1,2, *, D. Gorman1, P. Byrne1, A. Larkin1, A. Dowling1 and J. F. Malone1 1 Haughton Institute, Dublin 8 & The Adelaide and Meath Hospital incorporating the National Children’s Hospital, Dublin, 2 Haughton Institute & St. James’s Hospital, Dublin 8, Ireland
Computed radiography (CR) and digital radiography (DR) are replacing traditional film screen radiography as hospitals move towards digital imaging and picture archiving and communication systems (PACS). Both IPEM and KCARE have recently published quality assurance and acceptance testing guidelines for DR. In this paper, the performance of a range of CR and DR systems is compared. Six different manufacturers are included. Particular attention is paid to the performance of the systems under automatic exposure control (AEC). The patient is simulated using a range of thicknesses of tissue equivalent material. Image quality assessment was based on detector assessment protocols and includes pixel value measures as well as subjective assessment using Leeds Test Objects. The protocols for detector assessment cover a broad range of tests and in general detectors (whether DR or CR) performed satisfactorily. The chief limitation in performing these tests was that not all systems provided ready access to pixel values. Subjective tests include the use of the Leeds TO20. As part of this work, suggested reference values are provided to calculate the TO20 image quality factor. One consequence of moving from film screen to digital technologies is that the dynamic range of digital detectors is much wider, and increased exposures are no longer evident from changes in image quality. As such, AEC is a key parameter for CR and DR. Dose was measured using a standard phantom as a basic means of comparing systems. In order to assess the AEC performance, exit doses were also measured while varying phantom thickness. Signal-to-noise ratios (SNRs) were calculated on a number of systems where pixel values were available. SNR was affected by the selection of acquisition protocol. Comparisons between different technologies and collation of data will help refine acceptance thresholds and contribute to optimising dose and image quality.
INTRODUCTION Film screen radiography is being replaced by computed radiography (CR) and digital radiography (DR). The digital images produced by these modalities may be linked to PACS bringing in benefits of storage, retrieval and image distribution, as well as image processing to improve image quality. In the case of digital detectors, improvements in detective quantum efficiency may also be achieved with the potential of reduced dose, image quality improvements or both. Any new technology brings with it new challenges in terms of its control and quality assurance (QA). KCARE(1,2) have developed protocols for both CR and DR receptors; IPEM(3) have expanded their Xray system tests to encompass digital technologies; AAPM have also published a protocol for CR QA(4). In general, the generators and X-ray tubes in the radiographic systems used in CR and DR remain the same as their film screen system counterparts and QA of the X-ray tube and generators in digital systems follows the standard methods(3). However, when automatic exposure control (AEC) is selected, the X-ray output is linked (directly or indirectly) to the detector performance and this must be considered. The Medicines and Healthcare products Regulatory Agency in the UK (MHRA) in advice published on the web notes: . . . ‘the wide *Corresponding author:
[email protected]
exposure dynamic range of such (DR and CR) systems may have the disadvantage that, if the . . . AEC . . . develops a fault or the output calibration drifts, the dose increase/decrease may not be identified readily. Also, the wide exposure dynamic range means there is significant potential for the initial setup of such systems not to be optimised(5).’ This paper reviews the performance of a range of CR and DR systems. QA of monitors (soft copy) performed as part of specific protocols is not included in this work. However, monitor function also forms a part of the QA results presented here, where subjective evaluations of image quality are made at a workstation/review monitor. Performance under AEC is reviewed for DR systems.
MATERIAL AND METHODS Image quality (IQ) assessments based on KCARE protocol(1,2) were performed on 23 (Kodak, AGFA, Fuji) CR systems, and 10 (Philips, Siemens, Mecall) DR systems. All systems were less than 3 y in service. Leeds Test Objects were used for some of the IQ tests. Dose measurements were taken with Radcal 6 and 60 cm3 dose chambers, which were calibrated to international standards. For AEC testing, PMMA was used to simulate a range of patient thicknesses. High-purity aluminium and copper filters were also used in some tests.
# The Author 2008. Published by Oxford University Press. All rights reserved. For Permissions, please email:
[email protected]
C. WALSH ET AL.
IQ tests were performed across the full range of systems and followed the standard KCARE protocol. AEC test protocols are less well defined and a range of approaches were tested on a subset of the DR systems in an attempt to find an optimum approach. RESULTS AND DISCUSSION Detector tests In KCARE protocol (Rev 8)(1,2), the following tests are listed under annual QC: dosimetry; dark noise; detector dose indicator consistency; uniformity; blurring and stitching artefacts; limiting spatial resolution (in one quadrant at 458 only); threshold contrast detail detectability (TCDD). The results of these tests are described briefly below. Dosimetry and dark noise The tube-current exposure time (mA s) needed to produce a range of output exposures is measured and recorded. These values are used in later tests to obtain the appropriate exposure for a given test (e.g. evaluation of TCDD is based on a 4 mGy exposure). The exposures are a function of tube and generator performance. Dark noise assesses the level of noise inherent in the system. No peculiarities were noted in the tests on any system: the results form a baseline for ongoing monitoring of performance. Dose detector index Most CR and DR systems have some form of dose detector index (DDI); essentially this is a numerical value displayed with each exposure that correlates to dose. The KCARE test assesses the consistency of this value for a defined exposure. All systems were within specification with coefficient of variation (CV) ranging from 0 to 5.2%. In the systems where a CV of 0% was measured, the resolution of the DDI was found to be poor: in other words a considerable variation in dose was required before the value incremented or decremented. The 0% variation reflects this poor sensitivity. Uniformity This test assesses the uniformity of the recorded signal from a uniformly exposed (1 mm Cu at tube head) detector. Full analysis requires access to pixel values so that the standard deviation can be measured in a given region of interest (ROI). However, where pixel values are not available a subjective assessment is possible by simply viewing the image and looking for non-uniformities and
imperfections. Viewing uniformity images is a crude but useful test. In the case of one of the digital systems assessed, the DR plate failed as it was evident from visual inspection that one of the quadrants of the detector was displayed at a different greyscale shading than the other three quadrants. Subsequent examination by the supplier revealed a detector fault (which had gone undetected in the supplier’s commissioning tests). This test is also of value in assessing CR plates. A batch of CR cassettes with intermittently faulty latches (which allowed the plates to slip out of the cassette) was discovered after speckling on uniformity images caused by dust particles was observed. Blurring and stitching artefacts The system is tested for any localised distortion or for stitching artefacts, by imaging mesh test objects (e.g. Leeds Test Objects MS4). Blurring was not observed in any of the systems tested. In digital systems, the stitching artefact (essentially a cross through the centre of the image, dividing the image into four quadrants) was evident when the MS4 Leeds Test Objects were imaged, but the artefact was processed out in uniformity images and clinical images. Limiting resolution and TCDD (Huttner and TO20) Both tests objects are widely used in subjective assessment of image quality; both suffer from considerable inter- and intra-subjective variability(6,7). The images may be viewed on machine review monitors, as film prints, or on PACS stations. The assessment takes in the whole imaging chain, as well as some tube performance characteristics ( particularly focal spot size). The Huttner provides a measure of limiting spatial resolution and performance improves with magnification. TO20 contains discs of varying density for assessing threshold contrast. The discs also vary in size, challenging the spatial resolution of the system. TO20 has been identified by Marsh et al.(8) as useful in overall image quality assessment. Limiting spatial resolution ranged from 2.8 to 4 lp mm21 and 3.15 to 5 lp mm21 for DR review monitors and CR review stations, respectively. PACS monitors showed 4–4.5 lp mm21 for images acquired on DR, and 3.15 –5.6 lp mm21 for images acquired on CR, with the higher values recorded for the high-resolution CR plates. Limiting spatial resolution of film prints taken from DR acquired images ranged from 3.55 to 4.5 lp mm21. Limiting spatial resolution should approach the Nyquist p limit; at 458 the Nyquist frequency is defined by 2/2p, where p is the pixel pitch(1,2). For the DR systems, p ¼ 0.143 mm and Nyquist frequency is 4.9 lp mm21, only slightly higher than the results of the QA tests.
Page 2 of 5
QA OF CR AND DR SYSTEMS
Nyquist limit was 4.2 lp mm21 for standard sized CR plates and 7.4 lp mm21 for the high-resolution plates. Review monitors generally scored lower than PACS monitors. Review monitors function simply to ensure a proper image has been acquired and are not intended as diagnostic monitors. KCARE recommend calculating an image quality factor (IQF) based on TCDD measures. The IQF is based on work by Gallacher et al.(9). IQF is calculated according to Equation (1) and involves comparison between the system under test and a reference set of results. IQF ¼
n 1X HT ðAi Þ Dref 0:5 n i¼1 HTref ðAi Þ D
ð1Þ
where HT(A), threshold contrast detail index values calculated from the image; Href T (A), threshold contrast detail index values calculated from a reference image of a system known to be in good adjustment; D, the dose to the image plate, Dref, the dose to the image plate for the reference image; n, the number of details in the test object.(1,2) In the KCARE protocol, TO20 images are acquired at 70 kVp, with 1 mm Cu in the beam and the appropriate milliampere seconds to deliver 4 mGy. However, the look-up table for HT(A) values is based on 75 kVp, with 1.5 mm Cu in the beam(10). It was decided to keep the KCARE protocol for the tests. As use of the HT(A) values would introduce an undefined error into the calculation, the calculation was based on the number of targets observed in each category and the quality factor was denoted as IQF(b). All images are acquired and analysed in the same way (including the reference image) and IQF(b) is simply a comparison factor: the more targets observed, the higher the score for IQF(b) and the better the system’s contrast performance. A reference value by taking the median performance of 12 CR and DR systems based on reading film printouts has been obtained. The median (and first quartile) values are shown in Table 1. Taking each system individually and comparing it to the reference system (median value) yielded the following results: for images acquired on DR systems and displayed on PACS, IQF(b) ranged from 0.9 to 1.22; IQF(b) of DR acquired images Table 1. Suggested reference values for TO20 for each target (a–m).
Median 1st Quartile
a
b
C
d
e
F
g
h
j
K
l
M
6 5
6 5
6 5
6 6
6 5
6 4
7 6
6 6
5 5
8 7
6 6
4 4
printed to film ranged from 0.98 to 1.1. CR images viewed on PACS had IQF(b)s ranging from 0.85 to 1.05; IQF(b) for CR images printed on film ranged from 0.85 to 0.95. The images for these tests were acquired and viewed with a standard clinical protocol. Automatic exposure control A key performance indicator for AEC is receptor dose: this dose should remain constant as patient thickness varies. Measuring receptor dose in CR systems is relatively straight forward as the measurement chamber can easily be placed in the bucky tray. In DR systems, the detector is often not readily accessible. In this section, attention is restricted to DR systems (where receptor dose is more difficult to assess directly) and the problem of establishing an optimum protocol for assessing AEC performance is considered. AEC tests for DR systems are summarised in IPEM Report 91(2). However, most tests involve recoding milliampere seconds or DDI (and thus do not directly assess AEC performance) with only limited guidelines offered to assess receptor dose with varying phantom thickness. Expected values are not suggested, and acceptance thresholds allow wide variations from baseline for some of the tests. A repeatability test is described using a standard phantom and an exit dose test, which challenges AEC performance with varying phantom thickness. Repeatability A 21 mm Al plate (supplied as standard with the systems) was mounted at the tube head. The measurement chamber was placed on the table or at the face of the chest detector. SID (source to image distance) was 150 cm for chest detectors and 110 cm for table detectors. All exposures were taken at 70 kV and with AEC enabled. Six table detectors and six chest detectors were assessed. Doses measured at the table ranged from 4.0 to 5.5 mGy. Mean dose was 4.8 mGy (+0.6). At the chest detector, doses ranged from 4.0 to 6.6 mGy. Mean dose was 4.6 mGy (+1.0). This test does not challenge AEC function, but provides a useful baseline measure that can be compared directly to suppliers’ commissioning tests. Exit dose with varying phantom thickness Seven table detectors and five chest units were assessed. Exposures were under AEC at 70 kV. The grid was out. SID Table was 110 cm, SID chest was 150 cm. PMMA of thickness 5– 20 cm (varied in 5 cm steps) was placed in the beam. The PMMA was raised 3–5 cm and the measurement chamber
Page 3 of 5
C. WALSH ET AL.
was placed on the detector side of the PMMA in order to measure exit dose. With 15 cm PMMA in the beam, average exit dose is 6.6 mGy at the table and 6.0 mGy at the chest bucky. Expected receptor dose for digital systems is 3–5 mGy (varies between manufacturers). Exit dose will be slightly higher than receptor dose because of attenuation of the table (or covers for chest unit), the distance between the measurement chamber and the image receptor and differences in scatter. Exit doses are reasonably steady for each system between 10 and 20 cm phantom thickness. Doses are plotted and shown in Figure 1. Signal-to-noise ratio In addition to assessing individual parts of the imaging chain (e.g. receptor performance), it is useful to include some IQ measurements that assess image quality after the image data has passed through all stages involved in acquisition, processing and display. Assessing quality of Huttner and TO20 images on PACS workstations involves a check on the whole chain, but uses a subjective assessment method. Measuring signal-to-noise ratio (SNR) of
test images, acquired at the X-ray system being assessed, offers the possibility of a more objective IQ test. If this is performed at the PACS workstations the test encompasses the full imaging chain. Our interest is reviewing the potential of employing a simple measure to estimate SNR that could be incorporated into routine QA work. Software on PACS systems allows easy access to pixel values. In this experiment, SNR was calculated in a uniform image as a simple ratio of mean pixel value and standard deviation in a region of interest approximately 1/3 the size of the image. This technique is similar to the one used in QA of digital mammography(11). Tests were restricted to two dual detector DR systems. Images were acquired as above with 5– 20 cm PMMA in the beam. SNR values varied from 27 to 39 across a range of clinical protocols with PMMA thicknesses of 10–20 cm, but when a machine QA protocol was selected SNR range increased to 181–236 (Figure 2). The SNR increase was due to a much lower standard deviation in the ROI on the QA protocol. As the same phantom was imaged with the same exposure parameters in each case it appears likely that additional processing (smoothing) is being performed on the images acquired under the QA protocol. This result is also of interest for the KCARE detector tests and emphasises the need for caution in selecting acquisition protocols that reflect clinical performance.
CONCLUSIONS
Figure 1. Exit dose in mGy with varying PMMA thickness (cm).
Most systems passed IQ detector tests; where failures occurred, most were detected from visual inspection of uniformity images. The paper proposes reference values for calculating IQF for CR and DR systems.
Figure 2. Variance in SNR with PMMA thickness (cm). Results 1– 4 were acquired with an abdomen protocol: 1 and 3 were acquired at table detectors; 2 and 4 were acquired at chest detectors. Results 5 and 6 were acquired at a table and chest detector using a pre-programmed QA protocol.
Page 4 of 5
QA OF CR AND DR SYSTEMS
The problem of assessing AEC is explored, and values for exit dose are established for DR table and chest detectors. Clinical protocols and particularly quality protocols loaded on DR (or CR) systems may bring in different processing features, complicating the task of establishing simple, routine image quality tests, which provide assurance of clinical function. FUNDING This work was conducted partially within the frame of the European Commission 6th Framework Programme SENTINEL Contract No FP6 – 012909. REFERENCES 1. KCARE. Protocol for the QA of Computed Radiography Systems, Commissioning and Annual QA Tests. Available on http://www.kcare.co.uk/Education/ protocols.htm. 2. KCARE. Protocol for the QA of Direct Digital Radiography Systems, Commissioning and Annual QA Tests. Available on http://www.kcare.co.uk/Education/ protocols.htm. 3. IPEM Report 91. Recommended Standards for Routine Performance Testing of Diagnostic X-ray Imaging Systems, IPEM (2005).
4. AAPM Report No. 93. Acceptance Testing and Quality Control of Photostimulable Storage Phosphor Imaging Systems, Report of AAPM Task Group 10, October (2006). 5. MHRA, Radiation dose issues in digital radiography systems. Available on http://www.mhra.gov.uk/home/ idcplg?IdcService=SS_GET_PAGE&nodeId=263. 6. Tapiovaara, M. Image quality measurements in radiology, Radiat. Prot. Dosim. 117(1–3), 116– 119 (2005). 7. Walsh, C., Dowling, A., Meade, A. and Malone, J. Subjective and objective measures of image quality in digital fluoroscopy, Radiat. Prot. Dosim. 117(1– 3), 34– 37 (2005). 8. Marsh, D. and Malone, J. Methods and materials for subjective and objective measurements of image quality, Radiat. Prot. Dosim. 94(1–2), 37–42 (2001). 9. Gallacher, J., Mackenzie, A., Batchelor, S., Lynch, I. and Saunders, J. E. Use of a quality index in threshold contrast detail detection measurements in television fluoroscopy, Br. J. Radiol. 76, 464– 472 (2003). 10. Coewen, A. R., Haywood, J. M., Workman, A., Coleman, N. J., McArdle, S. and Clarke, O. F. Leeds DSF Test Objects, Instruction Manual (University of Leeds) (1987). 11. Van Engen, R., Young, K., Bosmans, H. and Thijssen, M. The European Protocol for the Quality Control of the physical and technical aspects of mammography screening, Addendum on Digital Mammography, version 1.0, (2003).
Page 5 of 5
