- Email: [email protected]
Digital Radiography of the Skeleton Using a Large-Area Detector Based on Amorphous Silicon Technology: Image Quality and Potential for Dose Reduction in Comparison with Screen-Film Radiography
Clinical Radiology (2000) 55, 615±621 doi:10.1053/crad.2000.0493, available online at http://www.idealibrary.com on
Digital Radiography of the Skeleton Using a Large-Area Detector Based on Amorphous Silicon Technology: Image Quality and Potential for Dose Reduction in Comparison with Screen-Film Radiography È L K * , M . S T R O T Z ER * , N . H O LZ K N E C H T ² , C . M A N K E* , M. L EN H A R T* , M. VO J . G M E I N W I E S E R ³ , J. L I N K * , M . R E I S E R ² , S . FE U E R B A C H * *Department of Diagnostic Radiology, University Hospital of Regensburg, Franz-Josef-Strauss-Allee 11, D-93042 Regensburg; ²Department of Diagnostic Radiology, Klinikum Grosshadern, University of Munich, Marchioninistr. 15, D-81377 Munich and ³Department of Diagnostic Radiology, Klinikum FuÈrth, Jakob-Henle-Strasse 1, D-90766 FuÈrth, Germany Received: 9 August 1999 Revised: 19 January 2000 Accepted: 25 January 2000 AIM: The purpose of this study was to evaluate a large-area, ¯at-panel X-ray detector (FD), based on caesium-iodide (CsI) and amorphous silicon (a-Si) with respect to skeletal radiography. Conventional images were compared with digital radiographs using identical and reduced radiation doses. MATERIALS AND METHODS: Thirty consecutive patients were studied prospectively using conventional screen-®lm radiography (SFR; detector dose 2.5 mGy). Digital images were taken from the same patients with detector doses of 2.5, 1.25 and 0.625 mGy, respectively. The activematrix detector had a panel size of 43 ´ 43 cm, a matrix of 3 ´ 3K, and a pixel size of 143 mm. All hard copies were presented in a random order to eight independent observers, who rated image quality according to subjective quality criteria. Results were assessed for signi®cance using the Student's t-test (con®dence level 95%). RESULTS: A statistically signi®cant preference for digital over conventional images was revealed for all quality criteria, except for over-exposure (detector dose 2.5 mGy). Digital images with a 50% dose showed a small, statistically not signi®cant, inferiority compared with SFR. The FD-technique was signi®cantly inferior to SFR at 75% dose reduction regarding bone cortex and trabecula, contrast and overall impression. No statistically signi®cant differences were found with regard to over- and under-exposure and soft tissue presentation. CONCLUSION: Amorphous silicon-based digital radiography yields good image quality. The potential for dose reduction depends on the clinical query. VoÈlk, M. et al. (2000). Clinical Radiology 55, 615±621. q 2000 The Royal College of Radiologists Key words: skeleton, radiography, comparative studies, digital, amorphous silicon, dose reduction.
Recent developments in digital radiography are based on ¯atpanel detectors consisting of a scintillator (e.g. caesium iodide, CsI) and a matrix of active electrical circuit elements of amorphous silicon [1±3]. Phantom studies and ®rst clinical experience yielded promising results for image quality and potential dose reduction [4±7]. However, these studies were performed with a prototype detector of limited size (15 ´ 15 cm). The goal of this study was to evaluate a large-area (43 ´ 43 cm), active-matrix detector based on CsI and amorphous silicon with respect to image quality in skeletal radiography and the effects of dose reduction. Screen-®lm radiography was the standard of reference. Author for correspondence and guarantor of study: Markus VoÈlk, M.D., Department of Diagnostic Radiology, University Hospital of Regensburg, D-93042 Regensburg, Germany. 0009-9260/00/080615+07 $35.00/0
MATERIALS AND METHODS
Imaging Systems The digital X-ray unit was an experimental system (Siemens, Forchheim, Germany). It included an X-ray tube with a focal spot size of 0.6 and 1.0 mm. A ¯at-panel detector (Trixell, Moirans, France) was horizontally mounted in a table behind a static anti-scatter grid (80 lp/cm, ratio 15:1) with a focal distance of 115 cm. The detector used in this study had a matrix size of 3 ´ 3K pixel elements with a pixel size of 143 mm, providing an active area of 43 ´ 43 cm. This large active area, suf®cient for most examinations, was made possible by tiling four individual quarter panels together [3]. This tiling did not result in visually perceptible artefacts. The scintillator consisted of a layer of about 500 mm thallium-doped q 2000 The Royal College of Radiologists
616
CLINICAL RADIOLOGY
caesium iodide (CsI:Tl). The CsI, with its needle-like crystalline structure, together with the small pixel size, results in a high detective quantum ef®ciency (DQE) of more than 60% at zero line pairs per millimetre and a good modulation transfer function (MTF) which is comparable to the MTF of screen-®lm systems of speed 400 [3]. Detective quantum ef®ciency is the most important single parameter which re¯ects image quality and depends on spatial frequency. The MTF values at low spatial frequencies were optimized by a ®lter with a kernel size of about 3 cm and a gain of 15% (harmonization). Finally, images were high-pass ®ltered with a 7 ´ 7 (pixel) kernel. The detector converts the analog signal into a digital signal with a 14-bit resolution. The digital images were transfered to a Magic View workstation (Siemens), where automatic windowing and levelling of the digital images were performed. However, adjustment of density and contrast had to be optimized individually, because the automatic algorithm was still an experimental version. Digital raw data processing consisted of three steps. A nonlinear look-up table (LUT) was applied. Organ-speci®c LUTs were used to reach the same image impression as in conventional radiography. A digital global gain was applied to achieve the same operating point for different dose levels (speed 400, 800 and 1600). The high pass gain was 50% for speed 400, 15% for speed 800, and 0 (i.e., no high pass ®ltering) for speed 1600. Digital post-processing resulted in MTF values of 30% at 2.5 lp/mm and 4% at 8 lp/mm, respectively. Digital images were printed on Ektascan DHG laser ®lms using an 8 bit Ektascan XLP laser printer (Eastman Kodak, Rochester, NJ, U.S.A.). Due to the limited laser-®lm format of 35 ´ 43 cm, the 43 ´ 43 cm digital radiographs had to be printed using a 25% size reduction. A screen-®lm system with a speed of 400 was used as reference (Lanex Regular/T-MAT Plus DG ®lm, Eastman Kodak). The MTF value of this system is 20% at 2.8 lp/mm and 4% at 6.2 lp/mm. Calculation of ®lm speed was based on the radiation dose required to produce an optical density of 1.0 above base plus fog (1000 Gy/dose [Gy] speed) [8]. Measurement of radiation dose was performed using a DALi ionization chamber (PTW, Freiburg, Germany). Digital and conventional images were taken using a phototimer (Iontomat, Siemens). The ionization chambers were placed identically for digital and conventional radiographs. The detector dose was 2.5 mGy for conventional imaging. Digital radiographs were generated with detector doses of 2.5 mGy, 1.25 mGy and 0.625, respectively.
Data Acquisition This clinical study was approved by the ethics committee. Thirty consecutive patients were examined prospectively after written consent was obtained. Patients under 50 years of age, pregnant women and patients who were seriously ill were excluded from the study. A total of seven patients refused to participate. Conventional images were acquired no more than 30 min before the digital images. Identical projections and ®elds-ofview were applied. Kilovoltage was identical for conventional and digital studies. All 30 patients received conventional radiographs with standard dose according to speed 400, as well as digital images with identical radiation exposure plus 50% and
75% dose reduction. All images were acquired by one of two experienced radiographers (professional experience exceeded 20 years in both). The applied kilovoltage and amperage values were registered for all studies. None of the investigations had to be repeated due to technical reasons, speci®cally exposure errors. A variety of anatomic regions were investigated: spine, n 7; pelvis and hip, n 5; extremities, n 16; ribs, n 1; clavicles, n 1. The study included both patients with normal ®ndings (n 13) and patients with disease (n 17). Fractures were present in 10 patients, of whom nine had metallic implants (screws, protheses, plates). One patient was in a cast without a metallic implant. Five patients had osteoarthrotic changes. One patient presented a metastatic lesion ®lled with bone cement. One patient showed a amputation of D1 of the right foot.
Data Analysis Hard copies were presented in a random order to eight independent observers (six board certi®ed general radiologists, and two residents with 5 years of professional experience). Potential bias from the direct comparison of conventional and digital images of the same patient was excluded. Nevertheless, both investigations produced obviously different image appearances. Thus, the observers were not blinded for digital and conventional radiographs. No information about patient history or exposure parameters was given. Seven different image features were evaluated: (1) bone cortex; (2) bone trabecula; (3) soft tissue; (4) overall contrast; (5) over-exposure (density); (6) under-exposure (density); and (7) overall impression. Image quality was ranked on a subjective four-point scale: (1) excellent; (2) good (minor limitations); (3) moderate (major limitations); (4) bad visualization (non-diagnostic). Intermediate scores at 0.5 intervals were allowed. The quality criteria and ranking scale were derived from the literature [9±12]. The subjective evaluation of the above-mentioned features depended on several objective parameters simultaneously, such as spatial and contrast resolution. Mean values were calculated for each feature, each observer, and each conventional and digital radiograph. The conventional radiographs were compared with the digital radiographs by comparing the averages using the two-tailed Student's t-test for paired samples (95% con®dence level). A total of 6720 observations were analysed. RESULTS
The quality of images obtained with the amorphous silicon detector at a speed of 400 were rated signi®cantly better in six of seven features compared with those obtained by conventional screen-®lm radiography when the same exposure parameters were applied (Table 1) (Fig. 1). Only over-exposure was rated as slightly better (statistically not signi®cant). Digital images with a 50% dose reduction were slightly, but statistically not signi®cantly, worse in most aspects compared with conventional screen-®lm radiographs. Soft tissue representation, over-exposure, under-exposure and overall impression showed a small, statistically not signi®cant, preference when a digital speed of 800 was simulated (Table 1).
617
DIGITAL RADIOGRAPHY OF THE SKELETON
Table 1 ± Mean values of quality scores (data from 30 patients and eight observers) Category/feature
SFR reference dose
FD reference dose
FD 50% of standard dose
FD 25% of standard dose
Bone cortex Bone trabeculae Soft tissue Overall contrast Over-exposure (density) Under-exposure (density) Overall impression
1.35 (0.18) 1.48 (0.23) 1.54 (0.24) 1.46 (0.10) 1.31 (0.13) 1.34 (0.12) 1.50 (0.14)
1.17 (0.10)* 1.28 (0.19)* 1.24 (0.12)* 1.31 (0.17)* 1.21 (0.20) 1.19 (0.09)² 1.25 (0.11)²
1.37 (0.14) 1.53 (0.18) 1.37 (0.17) 1.52 (0.13) 1.25 (0.25) 1.32 (0.09) 1.49 (0.14)
1.52 (0.15)² 1.82 (0.17)² 1.41 (0.20) 1.61 (0.17)* 1.27 (0.28) 1.37 (0.16) 1.68 (0.11)*
Standard deviations (SD) in parentheses. SFR screen-®lm radiography (speed 400); FD digital ¯at-panel detector; P-value of differences (lower or greater) determined between FD and SFR with the two-tailed Student's t-test for paired samples without the Bonnferoni correction. * P < 0.05; ² P < 0.005; all other comparisons between FD and SFR were not signi®cant.
At a simulated speed of 1600 the amorphous silicon detector was signi®cantly inferior in all respects except soft tissue representation (Fig. 2), and over-exposure; these two features were rated as slightly superior to conventional screen-®lm radiographs. Under-exposure was rated as slightly inferior to conventional screen-®lm radiographs (Table 1).
DISCUSSION
In this study, screen-®lm radiographs of the skeleton were compared with hard copies of digital images obtained at different radiation doses (simulated speeds of 400, 800 and 1600). Subjective criteria of image quality were applied.
(a)
(b)
(c)
(d)
Fig. 1 ± A 66-year-old woman with breast cancer. (a) Conventional screen-®lm radiography of the hip shows a defect in the right os ilium ®lled with bone cement after resection of a metastatic lesion. (b) Digital images with a simulated speed of 400. (c, d) Digital images with a simulated speed of 800 (c), and 1600 (d). Diagnostic information is equivalent on all images compared with conventional screen-®lm radiography, but noise is increased at a simulated speed of 1600 (75% dose reduction).
618
CLINICAL RADIOLOGY
(a)
(b)
(c)
(d)
Fig. 2 ± A 65-year-old man with diabetes mellitus type IIb. (a) Conventional screen-®lm radiography shows a calci®ed interdigital artery between D1 and D2. (b) Digital images with a simulated speed of 400. Visualization of the calci®ed artery is signi®cantly sharper on digital image with a simulated speed of 400 (b) than on conventional screen-®lm radiography (a). (c, d) Digital images with a simulated speed of 800 (c) and 1600 (d). Decreasing signal-to-noise ratio without loss of diagnostic information is observed at a simulated speed of 800 (c) and 1600 (d).
619
DIGITAL RADIOGRAPHY OF THE SKELETON
(a)
(b)
(c)
(d)
Fig. 3 ± A 64-year-old woman with fracture of the left ankle. (a) Conventional screen-®lm radiography shows a stabilized fracture of the left ankle with osteosynthetic materials. (b) Digital images with a simulated speed of 400. (c, d) Digital images with a simulated speed of 800 (c) and 1600 (d). Even at a simulated speed of 1600 (75% dose reduction) the position of the multiple osteosynthetic materials can be demonstrated suf®ciently.
620
CLINICAL RADIOLOGY
The presentation of soft tissue was superior to conventional radiographs in all digital images (speeds 400, 800 and 1600). Bone cortex and bone trabeculae as quality criteria for osseous structures were signi®cantly better in digital images with a speed of 400, slightly inferior at a speed of 800, and signi®cantly inferior at a speed of 1600. The post-processing optimizes the favourable properties of digital imaging, such as large dynamic range and high contrast resolution. Image quality with respect to over-exposure and under-exposure were rated superior or almost equivalent compared with conventional screen-®lm radiographs. However, edge enhancement increases perceived image noise resulting in inferior overall impression, thus its application is limited in radiation dose-reduced images, especially at a simulated speed of 1600 as demonstrated in the current study (Table 1). The image quality of the skeletal radiographs obtained by the digital system at a speed of 400 yielded a statistically signi®cant preference for the digital system in six out of seven features. At a simulated speed of 800 (50% dose reduction) no statistically signi®cant differences in image quality compared with conventional screen-®lm radiographs (speed 400) were found. Previously performed phantom and clinical studies with a small prototype of the amorphous silicon detector support the hypothesis that a dose reduction of up to 50% may be applied without loss of diagnostic con®dence [4,7]. At a dose reduction of 75% (speed 1600) the overall impression was signi®cantly inferior compared with conventional screen-®lm radiographs. This is based on the inferior signal-to-noise ratio. However, this ratio would be suf®cient in many clinical circumstances, such as routine follow-up studies after fractures, if alignment is to be considered (Fig. 3), bone grafting, or arthroplasty; in stress examinations to assess instability (e.g. of the ankle, lumbar or cervical spine); and for orthopaedic measurements. In this study, image quality was rated subjectively. Kappa statistics were not applicable because the relatively large number of observers and response categories would result in an extremely large number of possible multivariate responses [13]. A potential but unavoidable source of bias arises from the fact that conventional and digital images could be easily identi®ed due to their different hard copy appearances. This study was exclusively based on the evaluation of hard copies as otherwise presentation of the different imaging modalities in a random order would not have been possible. The features of soft-copy viewing with the possibility of digital post-processing (windowing and levelling as well as digital magni®cation) were not used. Additionally, the full information of the 14-bit images could not be represented on the 8-bit laser printer hard copies. It must be pointed out that the hard copies used in this study adversely compromised the amorphous silicon detector device in terms of size reduction and lack of soft copy tool manipulation. The digital images were processed to provide optical density and contrast resolution similar to those of conventional radiographs. To achieve this, digital raw data processing was applied in three steps to perform suf®cient contrast resolution: ®rst a non-linear LUT was applied, then harmonization and ®nally edge enhancement were performed. An active detector size of at least 43 ´ 43 cm is required to
allow both vertical (35 ´ 43 cm) and horizontal (43 ´ 35 cm) imaging orientations without detector rotation [14]. In our study no differences were noted in the different sites examined. Even with a detector size of 43 ´ 43 cm, some special radiographs, e.g. long-leg images, cannot be acquired and have to be performed with conventional screen-®lm technique. Detective quantum ef®ciency (DQE) is generally accepted as the best single objective indicator of image ®delity. The investigated detector proved to have a DQE of approximately 60% at 0.lp/mm (70 kVp, ®ltering with 21 mm Al) which is superior to screen-®lm radiography and commercially available storage-phosphor systems by a factor of almost two. A pixel size of 143 mm resulted in a maximum spatial resolution of 3.5 lp/mm. This spatial resolution is suf®cient for skeletal radiographs [15,16]. An exception is subperiostal bone resorption, which requires a pixel size of 80±100 mm [17]. Our ®ndings suggest that amorphous silicon technologybased digital radiography of the skeleton might replace conventional skeletal radiography without loss of image quality even at a dose reduction of up to 50%. This feature could be important, especially in paediatric radiology. A dose reduction of 75% is reserved for speci®c clinical queries, where the lower signal-to-noise ratio is not of clinical importance. Acknowledgments. We thank the engineers and physicists from Siemens AG Medical Engineering (Forchheim, Germany), as well as our technologists for their excellent technical support. REFERENCES 1 Antonuk LE, Yorkston J, Huang W, et al. A real-time, ¯at-panel, amorphous silicon, digital x-ray imager. RadioGraphics 1995;15:993± 1000. 2 Antonuk LE, El-Mohri Y, Siewerdsen JH, Yorkston J, Huang W, Scarpine VE, Street RA. Empirical investigation of the signal performance of a high resolution, indirect detection, active matrix, ¯at-panel imager (AMFPI) for ¯uoroscopic and radiographic operation. Med Phys 1997;24:51±70. 3 Chaussat C, Chabball J, Ducourant T, Spinnler V, Vieux G, Neyret R. New CsI/a-Si 17" ´ 17" X-ray ¯at panel detector provides superior detectivity and immediate direct digital output for general radiography systems. SPIE Med Imag 1998;3336:45±55. 4 VoÈlk M, Strotzer M, Gmeinwieser J, et al. Flat-panel X-ray detector using amorphous silicon technology: reduced radiation dose for the detection of foreign bodies. Invest Radiol 1997;32:373±377. 5 Strotzer M, Gmeinwieser J, Spahn M, et al. Amorphous silicon (a-Si), ¯at-panel, X-ray detector versus screen-®lm radiography: effect of dose reduction on the detectability of cortical bone defects and fractures. Invest Radiol 1998;33:33±38. 6 Strotzer M, Gmeinwieser J, VoÈlk M, FruÈnd R, Seitz J, Feuerbach S. Detection of simulated chest lesions with reduced radiation dose: comparison of conventional screen-®lm radiography and a ¯at-panel X-ray detector based on amorphous silicon (a-Si). Invest Radiol 1998;33:98±103. 7 Strotzer M, Gmeinwieser J, VoÈlk M, et al. Clinical application of a ¯atpanel X-ray detector based on amorphous silicon technology: image quality and potential for dose reduction in skeletal radiography. Am J Roentgenol 1998;171:23±27. 8 Haus AG. The AAPM/RSNA physics tutorial for residents: measurement of screen-®lm performance. RadioGraphics 1996;16:1165±1181. 9 Woodard PK, Slone RM, Gierada DS, Reiker GG, Pilgram TK, Jost RG. Chest radiography: depiction of normal anatomy and pathologic structures with selenium-based digital radiography versus conventional screen-®lm radiography. Radiology 1997;203:197±201. 10 Heesewijk HPM, Neitzel U, van der Graaf Y, de Valois JC, Feldberg MAM. Digital chest imaging with a selenium detector: comparison with
DIGITAL RADIOGRAPHY OF THE SKELETON
conventional radiography for visualization of speci®c anatomic regions of the chest. Am J Roentgenol 1995;165:535±540. 11 Floyd CE, Baker JA, Chotas HG, Delong DM, Ravin CE. Seleniumbased digital radiography of the chest: radiologists' preference compared with ®lm-screen radiographs. Am J Roentgenol 1995;165:1353± 1358. 12 Swee RG, Gray JE, Beabout JW, McLeod RA, Cooper KL, Bond JR, Wenger DE. Screen-®lm versus computed radiography imaging of the hand: a direct comparison. Am J Roentgenol 1997; 168:539±542. 13 Landis JR. The measurement of observer agreement for categorial data. Biometrics 1977;33:159±174.
621
14 Chotas HG, Dobbins JT, Ravin CE. Principles of digital radiography with large-area, electronically readable detectors: review of the basics. Radiology 1999;210:595±599. 15 Jonsson A, Laurin S, Karner G, et al. Spatial resolution requirements in digital radiography of scaphoid fractures: an ROC analysis. Acta Radiol 1996;37:555±560. 16 Murphey MD, Quale JL, Martin NL, et al. Computed radiography in musculoskeletal imaging: state of the art. Am J Roentgenol 1992;158: 19±27. 17 Murphey MD. Digital skeletal radiography: spatial resolution requirements for detection of subperiostal resorption. Am J Roentgenol 1989;152:541±546.
Digital Radiography of the Skeleton Using a Large-Area Detector Based on Amorphous Silicon Technology: Image Quality and Potential for Dose Reduction in Comparison with Screen-Film Radiography È L K * , M . S T R O T Z ER * , N . H O LZ K N E C H T ² , C . M A N K E* , M. L EN H A R T* , M. VO J . G M E I N W I E S E R ³ , J. L I N K * , M . R E I S E R ² , S . FE U E R B A C H * *Department of Diagnostic Radiology, University Hospital of Regensburg, Franz-Josef-Strauss-Allee 11, D-93042 Regensburg; ²Department of Diagnostic Radiology, Klinikum Grosshadern, University of Munich, Marchioninistr. 15, D-81377 Munich and ³Department of Diagnostic Radiology, Klinikum FuÈrth, Jakob-Henle-Strasse 1, D-90766 FuÈrth, Germany Received: 9 August 1999 Revised: 19 January 2000 Accepted: 25 January 2000 AIM: The purpose of this study was to evaluate a large-area, ¯at-panel X-ray detector (FD), based on caesium-iodide (CsI) and amorphous silicon (a-Si) with respect to skeletal radiography. Conventional images were compared with digital radiographs using identical and reduced radiation doses. MATERIALS AND METHODS: Thirty consecutive patients were studied prospectively using conventional screen-®lm radiography (SFR; detector dose 2.5 mGy). Digital images were taken from the same patients with detector doses of 2.5, 1.25 and 0.625 mGy, respectively. The activematrix detector had a panel size of 43 ´ 43 cm, a matrix of 3 ´ 3K, and a pixel size of 143 mm. All hard copies were presented in a random order to eight independent observers, who rated image quality according to subjective quality criteria. Results were assessed for signi®cance using the Student's t-test (con®dence level 95%). RESULTS: A statistically signi®cant preference for digital over conventional images was revealed for all quality criteria, except for over-exposure (detector dose 2.5 mGy). Digital images with a 50% dose showed a small, statistically not signi®cant, inferiority compared with SFR. The FD-technique was signi®cantly inferior to SFR at 75% dose reduction regarding bone cortex and trabecula, contrast and overall impression. No statistically signi®cant differences were found with regard to over- and under-exposure and soft tissue presentation. CONCLUSION: Amorphous silicon-based digital radiography yields good image quality. The potential for dose reduction depends on the clinical query. VoÈlk, M. et al. (2000). Clinical Radiology 55, 615±621. q 2000 The Royal College of Radiologists Key words: skeleton, radiography, comparative studies, digital, amorphous silicon, dose reduction.
Recent developments in digital radiography are based on ¯atpanel detectors consisting of a scintillator (e.g. caesium iodide, CsI) and a matrix of active electrical circuit elements of amorphous silicon [1±3]. Phantom studies and ®rst clinical experience yielded promising results for image quality and potential dose reduction [4±7]. However, these studies were performed with a prototype detector of limited size (15 ´ 15 cm). The goal of this study was to evaluate a large-area (43 ´ 43 cm), active-matrix detector based on CsI and amorphous silicon with respect to image quality in skeletal radiography and the effects of dose reduction. Screen-®lm radiography was the standard of reference. Author for correspondence and guarantor of study: Markus VoÈlk, M.D., Department of Diagnostic Radiology, University Hospital of Regensburg, D-93042 Regensburg, Germany. 0009-9260/00/080615+07 $35.00/0
MATERIALS AND METHODS
Imaging Systems The digital X-ray unit was an experimental system (Siemens, Forchheim, Germany). It included an X-ray tube with a focal spot size of 0.6 and 1.0 mm. A ¯at-panel detector (Trixell, Moirans, France) was horizontally mounted in a table behind a static anti-scatter grid (80 lp/cm, ratio 15:1) with a focal distance of 115 cm. The detector used in this study had a matrix size of 3 ´ 3K pixel elements with a pixel size of 143 mm, providing an active area of 43 ´ 43 cm. This large active area, suf®cient for most examinations, was made possible by tiling four individual quarter panels together [3]. This tiling did not result in visually perceptible artefacts. The scintillator consisted of a layer of about 500 mm thallium-doped q 2000 The Royal College of Radiologists
616
CLINICAL RADIOLOGY
caesium iodide (CsI:Tl). The CsI, with its needle-like crystalline structure, together with the small pixel size, results in a high detective quantum ef®ciency (DQE) of more than 60% at zero line pairs per millimetre and a good modulation transfer function (MTF) which is comparable to the MTF of screen-®lm systems of speed 400 [3]. Detective quantum ef®ciency is the most important single parameter which re¯ects image quality and depends on spatial frequency. The MTF values at low spatial frequencies were optimized by a ®lter with a kernel size of about 3 cm and a gain of 15% (harmonization). Finally, images were high-pass ®ltered with a 7 ´ 7 (pixel) kernel. The detector converts the analog signal into a digital signal with a 14-bit resolution. The digital images were transfered to a Magic View workstation (Siemens), where automatic windowing and levelling of the digital images were performed. However, adjustment of density and contrast had to be optimized individually, because the automatic algorithm was still an experimental version. Digital raw data processing consisted of three steps. A nonlinear look-up table (LUT) was applied. Organ-speci®c LUTs were used to reach the same image impression as in conventional radiography. A digital global gain was applied to achieve the same operating point for different dose levels (speed 400, 800 and 1600). The high pass gain was 50% for speed 400, 15% for speed 800, and 0 (i.e., no high pass ®ltering) for speed 1600. Digital post-processing resulted in MTF values of 30% at 2.5 lp/mm and 4% at 8 lp/mm, respectively. Digital images were printed on Ektascan DHG laser ®lms using an 8 bit Ektascan XLP laser printer (Eastman Kodak, Rochester, NJ, U.S.A.). Due to the limited laser-®lm format of 35 ´ 43 cm, the 43 ´ 43 cm digital radiographs had to be printed using a 25% size reduction. A screen-®lm system with a speed of 400 was used as reference (Lanex Regular/T-MAT Plus DG ®lm, Eastman Kodak). The MTF value of this system is 20% at 2.8 lp/mm and 4% at 6.2 lp/mm. Calculation of ®lm speed was based on the radiation dose required to produce an optical density of 1.0 above base plus fog (1000 Gy/dose [Gy] speed) [8]. Measurement of radiation dose was performed using a DALi ionization chamber (PTW, Freiburg, Germany). Digital and conventional images were taken using a phototimer (Iontomat, Siemens). The ionization chambers were placed identically for digital and conventional radiographs. The detector dose was 2.5 mGy for conventional imaging. Digital radiographs were generated with detector doses of 2.5 mGy, 1.25 mGy and 0.625, respectively.
Data Acquisition This clinical study was approved by the ethics committee. Thirty consecutive patients were examined prospectively after written consent was obtained. Patients under 50 years of age, pregnant women and patients who were seriously ill were excluded from the study. A total of seven patients refused to participate. Conventional images were acquired no more than 30 min before the digital images. Identical projections and ®elds-ofview were applied. Kilovoltage was identical for conventional and digital studies. All 30 patients received conventional radiographs with standard dose according to speed 400, as well as digital images with identical radiation exposure plus 50% and
75% dose reduction. All images were acquired by one of two experienced radiographers (professional experience exceeded 20 years in both). The applied kilovoltage and amperage values were registered for all studies. None of the investigations had to be repeated due to technical reasons, speci®cally exposure errors. A variety of anatomic regions were investigated: spine, n 7; pelvis and hip, n 5; extremities, n 16; ribs, n 1; clavicles, n 1. The study included both patients with normal ®ndings (n 13) and patients with disease (n 17). Fractures were present in 10 patients, of whom nine had metallic implants (screws, protheses, plates). One patient was in a cast without a metallic implant. Five patients had osteoarthrotic changes. One patient presented a metastatic lesion ®lled with bone cement. One patient showed a amputation of D1 of the right foot.
Data Analysis Hard copies were presented in a random order to eight independent observers (six board certi®ed general radiologists, and two residents with 5 years of professional experience). Potential bias from the direct comparison of conventional and digital images of the same patient was excluded. Nevertheless, both investigations produced obviously different image appearances. Thus, the observers were not blinded for digital and conventional radiographs. No information about patient history or exposure parameters was given. Seven different image features were evaluated: (1) bone cortex; (2) bone trabecula; (3) soft tissue; (4) overall contrast; (5) over-exposure (density); (6) under-exposure (density); and (7) overall impression. Image quality was ranked on a subjective four-point scale: (1) excellent; (2) good (minor limitations); (3) moderate (major limitations); (4) bad visualization (non-diagnostic). Intermediate scores at 0.5 intervals were allowed. The quality criteria and ranking scale were derived from the literature [9±12]. The subjective evaluation of the above-mentioned features depended on several objective parameters simultaneously, such as spatial and contrast resolution. Mean values were calculated for each feature, each observer, and each conventional and digital radiograph. The conventional radiographs were compared with the digital radiographs by comparing the averages using the two-tailed Student's t-test for paired samples (95% con®dence level). A total of 6720 observations were analysed. RESULTS
The quality of images obtained with the amorphous silicon detector at a speed of 400 were rated signi®cantly better in six of seven features compared with those obtained by conventional screen-®lm radiography when the same exposure parameters were applied (Table 1) (Fig. 1). Only over-exposure was rated as slightly better (statistically not signi®cant). Digital images with a 50% dose reduction were slightly, but statistically not signi®cantly, worse in most aspects compared with conventional screen-®lm radiographs. Soft tissue representation, over-exposure, under-exposure and overall impression showed a small, statistically not signi®cant, preference when a digital speed of 800 was simulated (Table 1).
617
DIGITAL RADIOGRAPHY OF THE SKELETON
Table 1 ± Mean values of quality scores (data from 30 patients and eight observers) Category/feature
SFR reference dose
FD reference dose
FD 50% of standard dose
FD 25% of standard dose
Bone cortex Bone trabeculae Soft tissue Overall contrast Over-exposure (density) Under-exposure (density) Overall impression
1.35 (0.18) 1.48 (0.23) 1.54 (0.24) 1.46 (0.10) 1.31 (0.13) 1.34 (0.12) 1.50 (0.14)
1.17 (0.10)* 1.28 (0.19)* 1.24 (0.12)* 1.31 (0.17)* 1.21 (0.20) 1.19 (0.09)² 1.25 (0.11)²
1.37 (0.14) 1.53 (0.18) 1.37 (0.17) 1.52 (0.13) 1.25 (0.25) 1.32 (0.09) 1.49 (0.14)
1.52 (0.15)² 1.82 (0.17)² 1.41 (0.20) 1.61 (0.17)* 1.27 (0.28) 1.37 (0.16) 1.68 (0.11)*
Standard deviations (SD) in parentheses. SFR screen-®lm radiography (speed 400); FD digital ¯at-panel detector; P-value of differences (lower or greater) determined between FD and SFR with the two-tailed Student's t-test for paired samples without the Bonnferoni correction. * P < 0.05; ² P < 0.005; all other comparisons between FD and SFR were not signi®cant.
At a simulated speed of 1600 the amorphous silicon detector was signi®cantly inferior in all respects except soft tissue representation (Fig. 2), and over-exposure; these two features were rated as slightly superior to conventional screen-®lm radiographs. Under-exposure was rated as slightly inferior to conventional screen-®lm radiographs (Table 1).
DISCUSSION
In this study, screen-®lm radiographs of the skeleton were compared with hard copies of digital images obtained at different radiation doses (simulated speeds of 400, 800 and 1600). Subjective criteria of image quality were applied.
(a)
(b)
(c)
(d)
Fig. 1 ± A 66-year-old woman with breast cancer. (a) Conventional screen-®lm radiography of the hip shows a defect in the right os ilium ®lled with bone cement after resection of a metastatic lesion. (b) Digital images with a simulated speed of 400. (c, d) Digital images with a simulated speed of 800 (c), and 1600 (d). Diagnostic information is equivalent on all images compared with conventional screen-®lm radiography, but noise is increased at a simulated speed of 1600 (75% dose reduction).
618
CLINICAL RADIOLOGY
(a)
(b)
(c)
(d)
Fig. 2 ± A 65-year-old man with diabetes mellitus type IIb. (a) Conventional screen-®lm radiography shows a calci®ed interdigital artery between D1 and D2. (b) Digital images with a simulated speed of 400. Visualization of the calci®ed artery is signi®cantly sharper on digital image with a simulated speed of 400 (b) than on conventional screen-®lm radiography (a). (c, d) Digital images with a simulated speed of 800 (c) and 1600 (d). Decreasing signal-to-noise ratio without loss of diagnostic information is observed at a simulated speed of 800 (c) and 1600 (d).
619
DIGITAL RADIOGRAPHY OF THE SKELETON
(a)
(b)
(c)
(d)
Fig. 3 ± A 64-year-old woman with fracture of the left ankle. (a) Conventional screen-®lm radiography shows a stabilized fracture of the left ankle with osteosynthetic materials. (b) Digital images with a simulated speed of 400. (c, d) Digital images with a simulated speed of 800 (c) and 1600 (d). Even at a simulated speed of 1600 (75% dose reduction) the position of the multiple osteosynthetic materials can be demonstrated suf®ciently.
620
CLINICAL RADIOLOGY
The presentation of soft tissue was superior to conventional radiographs in all digital images (speeds 400, 800 and 1600). Bone cortex and bone trabeculae as quality criteria for osseous structures were signi®cantly better in digital images with a speed of 400, slightly inferior at a speed of 800, and signi®cantly inferior at a speed of 1600. The post-processing optimizes the favourable properties of digital imaging, such as large dynamic range and high contrast resolution. Image quality with respect to over-exposure and under-exposure were rated superior or almost equivalent compared with conventional screen-®lm radiographs. However, edge enhancement increases perceived image noise resulting in inferior overall impression, thus its application is limited in radiation dose-reduced images, especially at a simulated speed of 1600 as demonstrated in the current study (Table 1). The image quality of the skeletal radiographs obtained by the digital system at a speed of 400 yielded a statistically signi®cant preference for the digital system in six out of seven features. At a simulated speed of 800 (50% dose reduction) no statistically signi®cant differences in image quality compared with conventional screen-®lm radiographs (speed 400) were found. Previously performed phantom and clinical studies with a small prototype of the amorphous silicon detector support the hypothesis that a dose reduction of up to 50% may be applied without loss of diagnostic con®dence [4,7]. At a dose reduction of 75% (speed 1600) the overall impression was signi®cantly inferior compared with conventional screen-®lm radiographs. This is based on the inferior signal-to-noise ratio. However, this ratio would be suf®cient in many clinical circumstances, such as routine follow-up studies after fractures, if alignment is to be considered (Fig. 3), bone grafting, or arthroplasty; in stress examinations to assess instability (e.g. of the ankle, lumbar or cervical spine); and for orthopaedic measurements. In this study, image quality was rated subjectively. Kappa statistics were not applicable because the relatively large number of observers and response categories would result in an extremely large number of possible multivariate responses [13]. A potential but unavoidable source of bias arises from the fact that conventional and digital images could be easily identi®ed due to their different hard copy appearances. This study was exclusively based on the evaluation of hard copies as otherwise presentation of the different imaging modalities in a random order would not have been possible. The features of soft-copy viewing with the possibility of digital post-processing (windowing and levelling as well as digital magni®cation) were not used. Additionally, the full information of the 14-bit images could not be represented on the 8-bit laser printer hard copies. It must be pointed out that the hard copies used in this study adversely compromised the amorphous silicon detector device in terms of size reduction and lack of soft copy tool manipulation. The digital images were processed to provide optical density and contrast resolution similar to those of conventional radiographs. To achieve this, digital raw data processing was applied in three steps to perform suf®cient contrast resolution: ®rst a non-linear LUT was applied, then harmonization and ®nally edge enhancement were performed. An active detector size of at least 43 ´ 43 cm is required to
allow both vertical (35 ´ 43 cm) and horizontal (43 ´ 35 cm) imaging orientations without detector rotation [14]. In our study no differences were noted in the different sites examined. Even with a detector size of 43 ´ 43 cm, some special radiographs, e.g. long-leg images, cannot be acquired and have to be performed with conventional screen-®lm technique. Detective quantum ef®ciency (DQE) is generally accepted as the best single objective indicator of image ®delity. The investigated detector proved to have a DQE of approximately 60% at 0.lp/mm (70 kVp, ®ltering with 21 mm Al) which is superior to screen-®lm radiography and commercially available storage-phosphor systems by a factor of almost two. A pixel size of 143 mm resulted in a maximum spatial resolution of 3.5 lp/mm. This spatial resolution is suf®cient for skeletal radiographs [15,16]. An exception is subperiostal bone resorption, which requires a pixel size of 80±100 mm [17]. Our ®ndings suggest that amorphous silicon technologybased digital radiography of the skeleton might replace conventional skeletal radiography without loss of image quality even at a dose reduction of up to 50%. This feature could be important, especially in paediatric radiology. A dose reduction of 75% is reserved for speci®c clinical queries, where the lower signal-to-noise ratio is not of clinical importance. Acknowledgments. We thank the engineers and physicists from Siemens AG Medical Engineering (Forchheim, Germany), as well as our technologists for their excellent technical support. REFERENCES 1 Antonuk LE, Yorkston J, Huang W, et al. A real-time, ¯at-panel, amorphous silicon, digital x-ray imager. RadioGraphics 1995;15:993± 1000. 2 Antonuk LE, El-Mohri Y, Siewerdsen JH, Yorkston J, Huang W, Scarpine VE, Street RA. Empirical investigation of the signal performance of a high resolution, indirect detection, active matrix, ¯at-panel imager (AMFPI) for ¯uoroscopic and radiographic operation. Med Phys 1997;24:51±70. 3 Chaussat C, Chabball J, Ducourant T, Spinnler V, Vieux G, Neyret R. New CsI/a-Si 17" ´ 17" X-ray ¯at panel detector provides superior detectivity and immediate direct digital output for general radiography systems. SPIE Med Imag 1998;3336:45±55. 4 VoÈlk M, Strotzer M, Gmeinwieser J, et al. Flat-panel X-ray detector using amorphous silicon technology: reduced radiation dose for the detection of foreign bodies. Invest Radiol 1997;32:373±377. 5 Strotzer M, Gmeinwieser J, Spahn M, et al. Amorphous silicon (a-Si), ¯at-panel, X-ray detector versus screen-®lm radiography: effect of dose reduction on the detectability of cortical bone defects and fractures. Invest Radiol 1998;33:33±38. 6 Strotzer M, Gmeinwieser J, VoÈlk M, FruÈnd R, Seitz J, Feuerbach S. Detection of simulated chest lesions with reduced radiation dose: comparison of conventional screen-®lm radiography and a ¯at-panel X-ray detector based on amorphous silicon (a-Si). Invest Radiol 1998;33:98±103. 7 Strotzer M, Gmeinwieser J, VoÈlk M, et al. Clinical application of a ¯atpanel X-ray detector based on amorphous silicon technology: image quality and potential for dose reduction in skeletal radiography. Am J Roentgenol 1998;171:23±27. 8 Haus AG. The AAPM/RSNA physics tutorial for residents: measurement of screen-®lm performance. RadioGraphics 1996;16:1165±1181. 9 Woodard PK, Slone RM, Gierada DS, Reiker GG, Pilgram TK, Jost RG. Chest radiography: depiction of normal anatomy and pathologic structures with selenium-based digital radiography versus conventional screen-®lm radiography. Radiology 1997;203:197±201. 10 Heesewijk HPM, Neitzel U, van der Graaf Y, de Valois JC, Feldberg MAM. Digital chest imaging with a selenium detector: comparison with
DIGITAL RADIOGRAPHY OF THE SKELETON
conventional radiography for visualization of speci®c anatomic regions of the chest. Am J Roentgenol 1995;165:535±540. 11 Floyd CE, Baker JA, Chotas HG, Delong DM, Ravin CE. Seleniumbased digital radiography of the chest: radiologists' preference compared with ®lm-screen radiographs. Am J Roentgenol 1995;165:1353± 1358. 12 Swee RG, Gray JE, Beabout JW, McLeod RA, Cooper KL, Bond JR, Wenger DE. Screen-®lm versus computed radiography imaging of the hand: a direct comparison. Am J Roentgenol 1997; 168:539±542. 13 Landis JR. The measurement of observer agreement for categorial data. Biometrics 1977;33:159±174.
621
14 Chotas HG, Dobbins JT, Ravin CE. Principles of digital radiography with large-area, electronically readable detectors: review of the basics. Radiology 1999;210:595±599. 15 Jonsson A, Laurin S, Karner G, et al. Spatial resolution requirements in digital radiography of scaphoid fractures: an ROC analysis. Acta Radiol 1996;37:555±560. 16 Murphey MD, Quale JL, Martin NL, et al. Computed radiography in musculoskeletal imaging: state of the art. Am J Roentgenol 1992;158: 19±27. 17 Murphey MD. Digital skeletal radiography: spatial resolution requirements for detection of subperiostal resorption. Am J Roentgenol 1989;152:541±546.