- Email: [email protected]
Influence of Image-Capturing Parameters on Digital X-Ray Radiogrammetry
Journal of Clinical Densitometry, vol. 8, no. 1, 87–94, 2005 © Copyright 2005 by Humana Press Inc. All rights of any nature whatsoever reserved. 1094-6950/05/8:87–94/$30.00
Original Article
Influence of Image-Capturing Parameters on Digital X-Ray Radiogrammetry Joachim Böttcher,*,1 Alexander Pfeil,1 Anders Rosholm,2 Ansgar Malich,1 Alexander Petrovitch,1 Bianka Heinrich,1 Gabriele Lehmann,3 Hans-Joachim Mentzel, MD,1 Gert Hein,3 Werner Linss,3 and Werner A. Kaiser,1 1Institute
of Diagnostic and Interventional Radiology Friedrich-Schiller-University Jena, Erlanger Allee 101, 07747 Jena, Germany; and 2Pronosco A/S, Kohavevej 5, 2950 Vedbaek, Denmark, 3Department of Rheumatology and Osteology, Clinic of Internal Medicine, Friedrich-Schiller-University Jena, Erlanger Allee 101, 07747 Jena, Germany, and 4Institute of Anatomy I, Friedrich-Schiller-University Jena, Bachstrasse 18, 07740 Jena, Germany
Abstract The purpose of this study was to evaluate the importance of different image-capturing conditions, which might influence the characteristics of radiographs and, consequently, impact calculations of bone mineral density (BMD) and Metacarpal Index (MCI) using digital X-ray radiogrammetry (DXR). Radiographs of the left hand of deceased males were acquired three times using systematically varied parameters: 4–8 miliamp seconds (mA); 40–52 kV; film-focus distance (FFD); 90–130 cm; film sensitivity, 200/400; and different image modalities (conventional vs original digital radiographs as well as digital printouts). Furthermore, the interradiograph reproducibility using both conventional equipment and printouts vs originals of digital images and the intraradiograph reproducibility (either conventional or digital printouts) were evaluated. All BMD and MCI measurements were obtained with the DXR technology. The interradiograph reproducibility of DXR–BMD using conventional images under standardized conditions (6 mAs; 42 kV; 1 m FFD; film sensitivity of 200) was calculated to be coefficient of variation (CV) = 0.49% for Agfa Curix film and CV = 0.33% for Kodak T-MAT-Plus film, whereas reproducibility error using digital images ranged from CV = 0.57% (digital printouts; Philips) to CV = 1.50% (original digital images; Siemens). The intraradiograph reproducibility error was observed to be CV = 0.13% (conventional; Kodak film) vs CV = 0.27% (digital printouts; Philips). The BMD calculation was not noticeably affected by changes of FFD, exposure level, or film sensitivity/film brand, but was influenced by tube voltage (CV = 0.99% for Kodak film to CV = 2.05% for Siemens digital printouts). No significant differences were observed between the BMD and MCI data. DXR provides measurements of MCI and BMD with high precision and reproducibility. The measurements are unaffected by all tested image-capturing conditions, with the exception of tube voltage. In addition, different digital image devices clearly have an effect on DXR reproducibility. Key Words: Digital X-ray radiogrammetry; bone mineral density (BMD); Metacarpal Index; tube voltage; X-ray equipment.
Introduction Received 06/24/04; Revised 09/20/04; Accepted 09/2004. * Address correspondence to Joachim Böttcher, Institute of Diagnostic and Interventional Radiology, Friedrich-SchillerUniversity Jena, Erlanger Allee 101, 07747 Jena, Germany. E-mail: [email protected]
Osteoporosis results in low bone mineral density (BMD) and microarchitectural deterioration of cortical and trabecular bone tissue, leading to diminished biomechanical competence of the skeletal parts. It is common knowledge that osteoporosis 87
88 in postmenopausal women is characterized by both a reduction of cortical thickness and a decrease of trabecular bone volume (1). The clinical consequences of osteoporosis essentially comprise the onset of low-trauma or atraumatic fractures affecting primarily the spine, hip, and wrist, contributing to higher morbidity, mortality, and health care costs (2). Regarding the multifactorial pathophysiology of osteoporosis, BMD is the most important component of bone strength and accounts for up to 80% of its variance (3). Therefore, decreased bone mass is a useful predictor of increased risk for fracture (4). Different prospective studies revealed the association between the reduction of BMD and an increased risk for fracture by threefold (5–7). The bone densitometers are essential diagnostic tools for the detection of osteoporosis and for identification of highrisk patients and assessment of efficacy regarding treatment and prevention of bone loss. Considerable advancement has been made in noninvasive methods for evaluating BMD in the past 25 yr (5), with particular focus on the evaluation of dual-energy X-ray absorptiometry (DXA) (8), ultrasoundbased methods (7), and quantitative computed tomography (QCT) (9). Conventional radiographs have been long used to estimate changes in BMD; however, they are known for high inaccuracy (varying image-capturing conditions and individual physiological variations). Osteopenia can be verified by hand X-rays only at a visible BMD reduction of more than 35% (10). In 1960, a more precise technique to assess bone status from radiographs was introduced by Barnett and Nordin (11), followed by Virtama and Mahonen (12). They measured the sum of the two cortical thicknesses of the middle metacarpal, divided by the width (i.e., Barnett and Nordin Index or Metacarpal Index [MCI]). Recently, the clinical application of radiogrammetry has shown significant improvement after refinement, computerization, and the use of algorithms for automatic image analysis (13–15). Our study is performed with the Pronosco X-Posure System (Version 2; Sectra Pronosco A/S, Denmark), which uses a combined computerized radiogrammetric and textural analysis of the three middle metacarpal bones for the measurement of BMD and MCI. The potential of digital X-ray radiogrammetry (DXR) for estimating cortical bone loss seems to be promising. Several studies have reported not only a closed correlation between peripheral measured DXA data and DXR–BMD (R = 0.61) (16,17), but they also have published normative values for DXR (18–22). In addition, Rosholm et al. (15) showed a significant association between DXR–BMD and DXA–BMD with R = 0.86 (distal radius), R = 0.73 (femur), and R = 0.62 (lumbar spine) in healthy women. Although the measurement of DXR–BMD did not take place at a location of high fracture incidence, Bouxsein et al. (23) documented, in a prospective study over an observation period of 5 yr, that DXR showed an equal prediction value of fracture risk compared to single-photon absorptiometry regarding fractures of the wrist, spine, and femur. Journal of Clinical Densitometry
Böttcher et al. Moreover, Hyldstrup et al. (24) verified an increase of DXR–BMD and MCI with combined reduction of cortical porosity in postmenopausal women after treatment with bisphosphonates, whereas no improvement of bone partition could be documented for DXA estimated on the spine, femur, and forearm. In addition, postmenopausal women using hormone replacement therapy showed a significant increase of MCI calculated by DXR (24). Unfortunately, a majority of these studies are presumably based on an extensive variety of different image-capturing conditions. Yet, despite the increasing clinical importance of DXR, there is a lack of reliable data demonstrating the possible effects of different image-capturing parameters when acquiring radiographs for BMD calculation with the DXR technology. The present cadaver study aims to evaluate the influence of different technical parameters on the performance of DXR and provides data about the reproducibility of DXR.
Methods Variation of the Technical Parameters The left hands from two deceased males (authorized agreement given) were radiographed according to a predefined study protocol. The radiographs were acquired under a number of different image-capturing conditions, including the following: • • • •
Film-focus distance (FFD; conventional radiographs and digital printouts) Tube voltage (kV; conventional and original digital images vs digital printouts) Exposure level (mAs; conventional radiographs and digital printouts) Film type and film foil modality (conventional radiographs)
With the exception of tube voltage and exposure level, the parameters were varied while retaining other parameters as fixed. Three repeated radiographs were acquired for each hand, in each of the different capture conditions. The following four conventional film types were tested: Kodak MAT plus 200, Kodak MAT plus 400, Agfa Curix 200, and Agfa Curix 400 with film foils from Agfa. For the conventional radiographs, the study protocol (two subjects) included five variations of FFD (90 cm, 100 cm, 110 cm, 120 cm, 130 cm) keeping all other parameters stable (42 kV, 6 mAs, Kodak T-MAT Plus 200). Similarly for conventional radiographs, tube voltage and exposure level were varied using 40 kV, 42 kV, 44 kV, 46 kV, 48 kV, 50 kV, 52 kV, and accordingly, 3 mAs, 4 mAs, 5 mAs, 6 mAs, 7 mAs and 8 mAs, while keeping all other parameters stable (FFD = 100 cm; other parameters as mentioned earlier). All conventional images were acquired using the same Xray equipment (Philips Super 80 CP) and then the BMD and MCI were calculated. Digital radiographs were performed using the Diagnost Philips Optimus device as well as the Siemens Polydoros SX 80 equipment. The sensitivity of digital printouts regarding Volume 8, 2005
Influence of Image-Capturing Parameters on DXR image-capturing conditions was similarly tested (three repeated radiographs, two subjects) for tube voltage, FFD, and exposure level using the same fixed settings as for conventional radiographs. Because of the solely verified influence of tube voltage regarding digital printouts, the original digital radiographs—acquired with varied tube voltages— were also directly subjected (picture archiving and communication system [PACS]) to a DXR–BMD and MCI analysis (one subject).
Intraradiograph and Interradiograph Reproducibility Contrary to many protocols, our in vitro study differentiates between intraradiograph and interradiograph coefficients of variation (CVs). The interradiograph reproducibility of DXR–BMD was evaluated by acquiring 10 repeated radiographs with repositioning under standard settings: •
•
•
For conventional radiographs (Philips Super 80 CP), our study considered the following standardized parameters: 42 kV, 6 mAs, Kodak T-MAT Plus 200 or Agfa Curix 200, FFD = 100 cm (two subjects). For digital printouts (Diagnost Philips Optimus and Siemens Polydoros SX 80), the following constant imagecapturing conditions were observed for two subjects: 42 kV, 4 mAs, FFD = 100 cm. For original digital images (Diagnost Philips Optimus and Siemens Polydoros SX 80), the following constant image-capturing conditions were also observed for one subject: 42 kV, 4 mAs, FFD = 100 cm.
To verify intraradiograph CVs, a single conventional image and a single printout underwent 10 repeated BMD analyses (two subjects). Results were expressed as mean and standard deviation and precision errors as the CV of repeated measurements. The statistical analysis was performed using the SPSS Version 10.13.
Digital X-Ray Radiogrammetry Measurements of BMD and MCI were obtained from conventional and printed digital radiographs using automated DXR, implemented in the Pronosco X-posure System (Version V.2; Sectra Pronosco A/S, Denmark). The Pronosco X-posure System was originally calibrated for analysis of conventional radiographs. The computation of DXR–BMD from a conventional radiograph and from a digital printout first involves digitization by a flatbed scanner and, subsequently, an automated analysis of the digitized images. The three middle metacarpal bones are automatically located and regions of interests (ROIs) are placed around the narrowest part of the bones. In detail, the algorithm couples the three ROIs from each metacarpal bone with each other and moves them along the bone shaft to a position identified by the minimum combined bone width. Subsequently, the outer and inner cortical edges of the included cortical bone parts are detected. To locate the metacarpals in a radiograph, the DXR technology uses a model-based algorithm according to the Active Shape Model (25), which is adapted to Journal of Clinical Densitometry
89 differentiate the diaphysis of the three middle metacarpals. The analyzed image and the ROIs were displayed on the monitor. The heights of the ROIs were fixed to 2.0 cm, 1.8 cm, and 1.6 cm for the second, third, and fourth metacarpals, respectively. In each area, the cortical thickness and porosity was analyzed 118 times per centimeter. The combined cortical thickness of a single metacarpal bone was calculated as the sum of the ulnar and radial cortical thickness (measured in millimeters). Assuming that the bone is elliptical (26) and that the volumetric BMD is approximately constant, the bone volume per area and, consecutively, a BMD value are calculated from the cortical thickness and provided together with the average cortical thickness and average MCI of the three metacarpals. DXR–BMD is measured in grams per square centimeter, including a minor correction for the estimated porosity (15). MCI expresses the mean cortical thickness normalized with the mean outer bone diameter. The DXR technology checked the quality of the scanned images and aborted the examination in case of inadequate quality. There is no operator activity required for the location of a ROI nor is it possible for the operator to modify or influence the size or the location of the ROI. It should be noted that the Pronosco X-posure System is not intended to analyze printouts of digital radiographs, because the printing process can influence the image characteristics. In addition, the inherent resolution of the digital radiographs might not be appropriately compensated for in the printing process; thus, the measured size of objects in the printed films could deviate from the correct size. BMD and MCI from the original digital radiographs were obtained by DXR using an in-house software version (PACS-compatible) applying the same analysis, except that the step of digitization by a scanner was superseded and that the software did not fully adapt to the specific resolution of the digital radiographs, because the BMD and MCI values were not calibrated to give the same average values as with the Pronosco X-posure System, which is primarily adjusted for analysis of conventional radiographs. Notice, therefore, that the average DXR–BMD level from digital radiographs (printouts and also original analysis) was not directly comparable to the level for conventional radiographs.
Results No statistically relevant differences could be observed between the two individuals and all results are consequently calculated using the values from both hands (with exception of the analyses regarding the original digital radiographs).
Influence of Technical Parameters The coefficient of variation (CV) with respect to changes of FFD was 0.98% for DXR–BMD and 0.51% for MCI. The influence of different film brands and film sensitivities on precision was calculated to be 0.49% (film brand) and 0.33% (film sensitivity) for DXR–BMD and 0.76% vs 0.51% for MCI (see Table 1). In the case of digital Volume 8, 2005
90
Böttcher et al. Table 3 Influence of Tube Voltage on DXR–BMD and MCI for Conventional Radiographs Analyzed by DXR
Table 1 Influence of Film-Focus Distance, Film Brand, and Film Sensitivity on DXR-BMD and MCI for Conventional Radiographs Analyzed by DXR
Conventional BMD mean (g/cm2) SD CV (%) MCI mean SD CV (%)
Film-focus distance (N = 30)
Film brand (N = 12)
Film foil sensitivity (N = 12)
0.611 0.006 0.98 0.391 0.002 0.51
0.609 0.003 0.49 0.396 0.003 0.76
0.613 0.002 0.33 0.395 0.002 0.51
Conventional
BMD mean (g/cm2) SD CV (%) MCI mean SD CV (%)
Digital (printouts) (N = 36)
0.607 0.003 0.49 0.401 0.002 0.50
0.574 0.002 0.35 0.388 0.003 0.77
printouts, the CV was 0.86% by alterations regarding filmfocus distance. Our results also revealed a minor precision error for variation of the exposure level; for conventional radiographs, the CV scored 0.49% (BMD) vs 0.50% (MCI), and for digital printouts, the CV was 0.35% (BMD) vs 0.77% (MCI). Detailed results are given in Table 2. Altered tube voltages showed a noticeable CV for conventional and apparently for the digital radiographs (see Tables 3 and 4). For the conventional radiographs, the calculated variations of DXR–BMD were 1.32% (Agfa) and 0.99% (Kodak), whereas MCI variations were observed to be 1.01% (Agfa) and 1.00% (Kodak). For digital images with varied tube voltages, calculations of CV were solely subdivided into the modality of calculation via printouts and direct analysis of the original digital radiographs (without a scanning procedure). Printouts showed a precision error for DXR–BMD of 0.57% for the Philips device and a higher CV of 2.05% for the Siemens equipment. The BMD values based on original digital images showed minor precision errors for both devices compared to printouts (see Table 4). The following average correction factors could be calculated for one step of the tube voltage (± 1 kV) based on the tube voltage of 42 kV: Journal of Clinical Densitometry
Kodak T MAT Plus (N = 24)
0.608 0.008 1.32 0.400 0.004 1.01
0.608 0.006 0.99 0.400 0.00 1.00
BMD mean (g/cm2) SD CV (%) MCI mean SD CV (%)
Table 4 Influence of Tube Voltage on DXR–BMD and MCI for Digital Radiographs Analyzed by DXR
Table 2 Influence of Exposure Level on DXR–BMD and MCI for Conventional and Digital Radiographs (Printouts) Analyzed by DXR Conventional (N = 36)
Agfa Curix (N = 36)
Digital Printouts (two subjects)
Original (PACS) (one subject)
Philips Siemens Philips Siemens (N = 42) (N = 32) (N = 19) (N = 15) BMD mean (g/cm2) SD CV (%) MCI mean SD CV (%)
• •
0.524 0.003 0.57 0.383 0.003 0.78
0.585 0.012 2.05 0.386 0.010 2.59
0.597 0.003 0.50 0.470 0.002 0.43
0.594 0.008 1.35 0.464 0.006 1.29
For DXR–BMD: −0.48% (conventional, Kodak), −0.75% (conventional, Agfa), +0.11% (digital printouts, Philips), and −0.92% (digital printouts, Siemens). For MCI: −0.35% (conventional, Kodak), −0.74% (conventional, Agfa), +0.04% (digital printouts, Philips), and −1.28% (digital printouts, Siemens).
Additionally, the Pronosco X-posure System was not able to analyze conventional images with tube voltages of 48 kV (three radiographs for Kodak), 50 kV (three radiographs for Kodak), and 52 kV (three radiographs for Agfa Curix and Kodak, respectively). No results could be calculated for digital printouts made by the Siemens device with tube voltages of 44 kV (one radiograph) and 48 kV vs 50 kV (two radiographs). No results were estimated for original digital images (PACS) by the Siemens device with tube voltages of 40 kV (two radiographs), 44 kV (one radiograph), 46 kV (one radiograph), 48 kV (one radiograph), and 50 kV (one radiograph), whereas missegmented images occurred using the Philips device with tube voltages of 48 kV (one radiograph) and 52 kV (one radiograph). Volume 8, 2005
Influence of Image-Capturing Parameters on DXR
91
Table 5 Interradiograph Reproducibility for Conventional Radiographs Analyzed by DXR Agfa Curix (N = 20) BMD mean (g/cm2) BMD SD CV (%) MCI mean SD CV (%)
Kodak T MAT Plus (N = 20)
0.614 0.003 0.49 0.404 0.002 0.50
0.611 0.002 0.33 0.401 0.002 0.50
Table 6 Interradiograph Reproducibility for Digital (Printouts vs Original) Radiographs Analyzed by DXR
Original (PACS) (one subject)
Philips Siemens Philips Siemens (N = 20) (N = 20) (N = 10) (N = 10) BMD mean (g/cm2) SD CV (%) MCI mean SD CV (%)
0.523 0.003 0.57 0.383 0.003 0.78
0.591 0.007 1.18 0.392 0.005 1.28
0.602 0.002 0.33 0.387 0.001 0.26
0.599 0.009 1.50 0.389 0.006 1.54
Reproducibility Our study differentiates between interradiograph and intraradiograph CVs.
Interradiograph Reproducibility •
•
•
For conventional radiographs, the interradiograph reproducibility for the BMD was CV = 0.49% (Agfa) and CV = 0.33% (Kodak) when analyzing 10 different images obtained from the same object. Similar results were shown for MCI (see Table 5). For digital printouts we found comparatively small precision errors regarding MCI (CV = 0.78% vs CV = 1.28%) and BMD measurements with CV = 0.57% vs 1.18% depending on digital X-ray equipment (see Table 6). For original digital radiographs, the CV ranged between 0.33% vs 1.50% (BMD) and 0.26% vs 1.54% (MCI) using an in-house PACS compatible software version for the DXR measurements (see Table 6).
Intraradiograph Reproducibility When analyzing one identical image 10 times, our data revealed the following low CVs (see Table 7): Journal of Clinical Densitometry
Conventional
Digital (Printouts)
Agfa Curix Kodak T Philips Siemens MAT Plus BMD mean (g/cm2) SD CV (%) MCI mean SD CV (%) •
Digital Printouts (two subjects)
Table 7 Intraradiograph Reproducibility (Measuring One Identical Image 10 Times) for Conventional and Digital Radiographs (Printouts) Analyzed by DXR
•
0.608 0.0016 0.26 0.399 0.0013 0.33
0.613 0.0008 0.13 0.402 0.0007 0.17
0.528 0.0014 0.27 0.386 0.0016 0.42
0.578 0.0003 0.05 0.380 0.0005 0.13
For conventional radiographs, the CV was 0.26% (Agfa) vs 0.13% (Kodak) with respect to BMD and 0.33% (Agfa) vs 0.17% (Kodak) regarding MCI. For digital printouts, the intraradiograph reproducibility of BMD was CV = 0.27% (Philips) vs 0.05% (Siemens) and ranged from 0.42% (Philips) to 0.13% (Siemens) regarding MCI.
Discussion DXR allows for detailed measurements of distances in a radiograph, partly as a result of the high resolution and the averaging over a larger number of individual geometrical measurements. Measurements of DXR can also be performed on standard radiographs (with an effective radiation dose of <0.01 mSv for acquisition of an standard forearm radiograph), thereby giving this technique the potential to be widely available (17). Many studies have confirmed the relevance of DXR as an advantageous diagnostic tool for the quantification of osteoporosis regarding age- and menopause-related decreases of BMD (15,16,20–23) and also for the detection of diseaserelated BMD loss (27). In addition, DXR is not influenced by beam hardening and is insensitive to soft tissue variations. In contrast to DXA, DXR works without an aluminum wedge, and because of the fully automated procedure, it is independent of operator-induced variations in the BMD calculations (28). In view of these advantages, in 1999 the US Food and Drug Administration approved DXR as a clinical method for estimating BMD. However, radiogrammetry is principally different from the usual bone densitometry, because it involves measurements of geometrical dimensions, whereas densitometry evaluates mineral densities. One major advantage of radiogrammetry is the higher spatial resolution of the digitized radiographs compared to the resolution of densitometry, allowing for a better separation of cortical and trabecular partitions, together with a more precise edge detection. Various improvements to the initial radiogrammetric method (11) have been suggested, including averaging over Volume 8, 2005
92 multiple measuring sites on the same bone and averaging over several bones (29). Reportedly, the CV has been improved by 11% for one measurement on a single metacarpal to 8.4% for the average over nine measurements made on three metacarpal bones. An optical magnifier has been used to provide a higher visual resolution (30). Aguado et al. (31) have used digitally enlarged radiographs to improve the visual resolution with a CV value of 2.4%. In the past several years, there has been increased interest in radiogrammetry because of the availability of automated or semiautomated algorithms based on digital X-ray equipment. The reduced observer dependence of DXR has improved the CV to 1.5% (repeated measurements on different images) and 0.89% (repeated measurements on the same image) as published by Maggio et al. (32) and Derisquebourg et al. (33), respectively. Only our data and those of Maggio et al. (32) differentiated between intraradiograph (CV for MCI: 1.5%) and interradiograph reproducibility (CV for MCI: 5.0%). Our intraradiograph and interradiograph reproducibilities, addressing both acquisition technique as well as analysis of BMD and MCI, demonstrated high values, suggesting the excellent precision of DXR for conventional as well as digitally acquired radiographs, similar to recently published studies (15,34). In these studies, the short-term CV of DXR is 0.24–0.65 % on a very low level (15,16,34), indicating that estimated bone loss was not based on the precision error of the osteodensitometric method itself. The high precision of DXR might partly be explained by the kind of algorithm and the procedure to detect the inner cortical edges of a bone (35). The inner cortical edges are used for calculation of the cortical thickness of the metacarpals and, consequently, the BMD estimate. In the image analysis, the inner cortical edge is associated with the intensity maximum of the intensity profile across the bone of interest. For a bone with thick cortical tissue, the intensity maximum is located on a broader and less curved top than for a bone with a thin cortical shell. This property implies that the position of the intensity maximum is more precisely defined on the narrow top of a thin bone than on the broader top of a thick bone (17). In the evaluation of CV values from other published studies, it is important to note that most likely the data have been obtained from repeated measurements of a single radiograph for one person, thereby neglecting possible variations in the results arising from different image-capturing conditions. Up to now, only a few studies (15,28,34) have reported on the potential influence of different imaging parameters on the calculation of BMD and MCI. Such information, however, is essential when using normative values of BMD interpretation, particularly with regard to the guarantee of reliable data. To achieve the most realistic study design possible, this study utilized the hands from two deceased males. Between the two hands, our data revealed no interobject differences, which is important when considering the reliability of DXR. Further investigations of the interobject variability using phantoms would be beneficial. Journal of Clinical Densitometry
Böttcher et al. In our study, technical parameters, potentially different between various sites during imaging, were varied. These included FFD, tube voltage, exposure level, film sensitivity, and film brand, as well as image developing devices. Other parameters such as film–object distance, spot size, and displacement of the object were not evaluated because these parameters do not vary in a high-quality imaging. Particularly when taking into account relative values of standard deviations (e.g., up to 10% of calculated mean in childhood [21]), our data document insensitivity of the DXR technology to altered exposure level, film brand, and film sensitivity, as well as varied FFD. Tube voltage appears to significantly influence the calculations of DXR–BMD and MCI, both for conventional radiographs and for digital images (printouts and original). The influence of tube voltage on DXR could be also verified in an in vitro phantom study (36). These results suggest that it is essential to use standard values of tube voltage that are similar between those images to be analyzed and those representing reference data. The fact that the DXR system did not recognize radiographs obtained by 48 kV, 50 kV, and 52 kV (conventional), 44 kV, 48 kV, and 50 kV (digital printouts), and 40 kV, 44 kV, 46 kV, 48 kV, 50 kV, and 52 kV (digital originals) emphasizes the importance of a standardized tube voltage. Most likely, this problem originates from minor contrasting of bone structures in suboptimally performed radiographs, whereby digital printouts seem to be more sensitive. Similarly, the MCI not only performs independently of body size variables (19) but also shows—widely comparable to DXR–BMD—minor CVs to most of the technical parameters altered during imaging, with the exception of tube voltage, as also mentioned by Nielsen (37). For the first time, our data have verified a decrease of precision for DXR–BMD and MCI when using different devices for digital imaging. The Siemens digital X-ray equipment showed not only a significantly reduced interradiograph reproducibility but also higher precision errors with a varied tube voltage (see Table 4); the larger precision error and the higher dependency from the tube voltage for Siemens images were likely the result of standard image processing applied to the images by the Siemens X-ray workstation. The DXR technology is dependent on a robust and consistent detection of edges. Image enhancement procedures that affect gray levels nonlinearly must, therefore, not be applied on images intended to be measured with DXR. This is ensured by defining a standard image processing protocol for each modality and, subsequently, verifying compliance to the protocol through digital imaging and communications in medicine (DICOM) tags before analyzing an image. This referred image processing protocol comprises edge enhancement, noise reduction, and other similar procedures that depend on the image contents. The short-term precision of the Siemens device will be affected by edge enhancement because it introduces a stochastic influence on the edge properties when obtaining repeated images of the same object (i.e., metacarpal). Because of the research purpose of the present study, the involved Siemens Volume 8, 2005
Influence of Image-Capturing Parameters on DXR images have been analyzed off-line without definition and use of a standard protocol excluding image enhancement. In this study, it was also revealed that the usage of printouts of digital images for BMD measurement should be limited. The reproducibility errors were significantly higher compared to conventional radiographs, with the exception of the analysis of the original digital images acquired by the Philips device. This phenomenon is explainable by a smearing effect of the image contents (especially edges) during the printing process, which influences the DXR results. The DXR technology actually provides a calculation of BMD and MCI using original digital images for direct analysis with an internal batch program version of the X-posure System. This software applies its own calibration constants to adapt to the different resolutions. All noise-related interaction from the printing process, unintended image enhancement, variations based on different scanner resolutions, and slight position alterations during the scanning process of radiographs could be avoided; consequently, the precision of DXR is significantly improved.
Summary DXR provides calculations of MCI and BMD with high intraradiograph and interradiograph reproducibility of conventional as well as digital radiographs; this technique has been shown to be insensitive toward alterations of film sensitivity, film brand, FFD, and exposure level. Only variation in the tube voltage demonstrated an influence on the calculation process for BMD and MCI in conventional and, particularly, digital printouts; also, different digital image devices clearly have an effect on DXR reproducibility. Constancy of tube voltage and the assortment of appropriate digital image devices are essential for achieving reliable results. The use of digital printouts for analysis is significantly limited by an increased smearing resulting from the printing process, but might be prevented by the use of an internal batch program version of the X-posure System (PACS-compatible) for DXR analysis of original digital images without a scanning procedure. The DXR technique has the potential to be an inexpensive and advantageous method for the assessment of osteoporosis in patients in different clinical settings, both prospective and retrospective (34).
Acknowledgments We would like to thank Pronosco-Sectra and Arewus (i.e., Mrs. M. Arens) for the use of the X- posure equipment. Finally, the authors would also like to thank Mr. D. Felsenberg, PhD (Berlin, Germany), Mr. C. C. Gluer, PhD (Kiel, Germany), Mr. S. Grampp, PhD (Wien, Austria), and Mr. H. Imhof, PhD (Wien, Austria) for their comments regarding our study.
References 1. Barger-Lux MJ, Recker RR. 2002 Bone microstructure in osteoporosis: transilial biopsy and histomorphometry. Topics Magn Reson Imaging 13:297–305. Journal of Clinical Densitometry
93 2. Cooper C, Atkinson EJ, Jacobsen SJ, O`Fallon WM, Melton LJ. 1993 Population-based study of survival after osteoporotic fractures. Am J Epidemiol 137:1001–1005. 3. Rice JC, Cowin SC, Bowman JA. 1988 On the dependence of the elasticity and strength of cancellous bone on apparent density. J Biomech 21:155–168. 4. Ross PD, Davis JW, Vogel JM, Wasnich RD. 1990 A critical review of bone mass and the risk of fracture in osteoporosis. Calcif Tissue Int 46:149–161. 5. Genant HK, Engelke K, Fuerst T, et al. 1996 Noninvasive assessment of bone mineral and structure: state of the art. J Bone Miner Res 11:707–730. 6. Gnudi S, Malavolta N. 2003 Comparison between T-score-based diagnosis of osteoporosis and specific skeletal site measurements: prognostic value for predicting fracture risk. J Clin Densitom 6:267–273. 7. Hans D, Genton L, Allaoua S, et al. 2003 Hip fracture discrimination study: QUS of the radius and the calcaneum. J Clin Densitom 6:163–172. 8. Sweeney AT, Malabanan AO Blake MA, et al. 2002 Bone mineral density assessment: Comparison of dual-energy X-ray absorptiometry measurements at the calcaneus, spine and hip. J Clin Densitom 5:57–62. 9. Braillon PM. 2002 Quantitative computed tomography precision and accuracy for long-term follow-up of bone mineral density measurements: a five yr in vitro assessment. J Clin Densitom 5:259–266. 10. Jergas M, Uffmann M, Escher H, et al. 1994 Interobserver variation in the detection of osteopenia by radiography and comparison with dual X-ray absorptiometry (DXA) of the lumbar spine. Skeletal Radiol 23:195–199. 11. Barnett E, Nordin B. 1960 The radiological diagnosis of osteoporosis: a new approach. Clin Radiol 1:166–174. 12. Virtama P, Mahonen H. 1960 Thickness of the cortical layer as an estimate of mineral content of human finger bones. Br J Radiol 6:60–62. 13. Kalla AA, Meyers OL, Parkyn ND, Kotze TJ. 1989 Osteoporosis screening—radiogrammetry revisited. Br J Rheumatol 28:511–517. 14. Kiel DP, Hannan MT, Broe KE, Felson DT, Cupples LA. 2001 Can metacarpal cortical area predict the occurence of hip fracture in women and men over 3 decades of follow-up? Results from the Framingham Osteoporosis Study. J Bone Miner Res 16:2260–2266. 15. Rosholm A, Hylsdrup L, Baeksgaard L, Grunkin M, Thodberg HH. 2001 Estimation of bone mineral density by digital X-ray radiogrammetry: theoretical background and clinical testing. Osteoporos Int 12:961–969. 16. Black DM, Palermo L, Sorensen T, et al. 2001 A normative reference database study for the Pronosco X-posure radiogrammetry system. J Clin Densitom 4:5–12. 17. Jorgensen JT, Andersen PB, Rosholm A, Bjarnason NH. 2000 Digital X-ray radiogrammetry: a new appendicular bone densitometric method with high precision. Clin Physiol 20: 330–335. 18. Van Rijn RR, Grootfaam DS, Lequin MH, et al. 2004 Digital Radiogrammetry of the hand in a pediatric and adolescent dutch caucasian population: normative data and measurements in children with inflammatory bowel disease and juvenile chronic arthritis. Calcif Tissue Int 74:342–350. 19. Hyldstrup L, Nielsen SP. 2001 The metacarpal index by digital X-ray radiogrammetry: normative reference values and comparison with dual-X-ray absorptiometry. J Clin Densitom 4:299–306. 20. Maffei L, Venarotti H, Treviso J, Sorensen TK, Nissen D. 2000 Digital X-ray radiogrammetry: a Hispanic normative reference Volume 8, 2005
94
21.
22.
23.
24. 25. 26. 27.
Böttcher et al. database for the Pronosco X-posure system. J Bone Miner Res 15:304. Malich A, Freesmeyer MG, Mentzel HJ, et al. 2003 Normative values of bone parameters of children and adolescents using digital computer-assisted radiogrammetry (DXR). J Clin Densitom 6:103–112. Wüster C, Wenzler M, Kappes J, Rehm C, Gühring T, Arnbjerg C. 2000 Digital X-ray radiogrammetry as a clinical method for estimating bone mineral density—a German reference database. J Bone Miner Res 15:298. Bouxsein ML, Palermo L, Yeung C, Black DM. 2002 Digital Xray radiogrammetry predicts hip, wrist and vertebral fracture in elderly women: a prospective analysis from the study of osteoporotic fractures. Osteoporos Int 13:358–365. Hyldstrup L, Jorgensen JT, Sorensen TK, Baeksgaard L. 2001 Response of cortical bone to antiresorptive treatment. Calcif Tissue Int 68:135–139. Cootes TF, Taylor CJ, Cooper D, Graham J. 1995 Active shape models: their training and application. Comp Vision Image Understand 61:38–59. Lazenby R. 1995 Brief communication: Non-circular geometry and radiogrammetry of the second metacarpal. Am J Phys Anthropol 97:323–327. Böttcher J, Malich A, Pfeil A, et al. 2004 Potential clinical relevance of digital radiogrammetry for quantification of periarticular bone demineralization in patients suffering from rheumatoid arthritis dependend on severity and compared with DXA. Eur Radiol 14:631–637.
Journal of Clinical Densitometry
28. Thodberg HH, Jensen JK, Rosholm A. 1999 BMD from digital X-ray radiogrammetry: Sensitivity to details of the image capture. J Bone Miner Res 14:369. 29. Bloom RA. 1980 A comparative estimation of the combined cortical thickness of various bone sites. Skeletal Radiol 5: 167–170. 30. Memma HE. 1991 Improved vertebral fracture threshold in postmenopausal osteoporosis by radiogrammetric measurements: Its usefulness in selection for preventive therapy. J Bone Miner Res 6:9–14. 31. Aguado F, Revilla M, Villa LF, Rico H. 1997 Cortical bone resorption in osteoporosis. Calcif Tissue Int 60:323–326. 32. Maggio D, Pacifici R, Cherubini A, et al. 1997 Age-related cortical bone loss at the metacarpal. Calcif Tissue Int 60:94–97. 33. Derisquebourg T, Dubois P, Devogelaer JP, et al. 1994 Automated computerized radiogrammetry of the second metacarpal and its correlation with absorptiometry of the forearm and spine. Calcif Tissue Int 54:461–465. 34. Ward KA, Cotton J, Adams JE. 2003 A technical and clinical evaluation of digital X-ray radiogrammetry. Osteoporos Int 14:389–395. 35. Helboe AB, Juul K, Rosholm A, Andersen PB, Bjarnason NH. 2001 High short-term in-vivo precision of the Pronosco Xposure System on hand X-rays. J Bone Miner Res 28:335. 36. Böttcher J, Pfeil A, Lehmann G, et al. 2004 Clinical trial for differentiation between corticoid-induced osteoporosis and periarticular demineralization via digital radiogrammetry in patients suffering from rheumatoid arthritis. J Rheumatol 63:473–482. 37. Nielsen SP. 2001 The metacarpal index revisited: A brief overview. J Clin Densitom 4:199–207.
Volume 8, 2005
Original Article
Influence of Image-Capturing Parameters on Digital X-Ray Radiogrammetry Joachim Böttcher,*,1 Alexander Pfeil,1 Anders Rosholm,2 Ansgar Malich,1 Alexander Petrovitch,1 Bianka Heinrich,1 Gabriele Lehmann,3 Hans-Joachim Mentzel, MD,1 Gert Hein,3 Werner Linss,3 and Werner A. Kaiser,1 1Institute
of Diagnostic and Interventional Radiology Friedrich-Schiller-University Jena, Erlanger Allee 101, 07747 Jena, Germany; and 2Pronosco A/S, Kohavevej 5, 2950 Vedbaek, Denmark, 3Department of Rheumatology and Osteology, Clinic of Internal Medicine, Friedrich-Schiller-University Jena, Erlanger Allee 101, 07747 Jena, Germany, and 4Institute of Anatomy I, Friedrich-Schiller-University Jena, Bachstrasse 18, 07740 Jena, Germany
Abstract The purpose of this study was to evaluate the importance of different image-capturing conditions, which might influence the characteristics of radiographs and, consequently, impact calculations of bone mineral density (BMD) and Metacarpal Index (MCI) using digital X-ray radiogrammetry (DXR). Radiographs of the left hand of deceased males were acquired three times using systematically varied parameters: 4–8 miliamp seconds (mA); 40–52 kV; film-focus distance (FFD); 90–130 cm; film sensitivity, 200/400; and different image modalities (conventional vs original digital radiographs as well as digital printouts). Furthermore, the interradiograph reproducibility using both conventional equipment and printouts vs originals of digital images and the intraradiograph reproducibility (either conventional or digital printouts) were evaluated. All BMD and MCI measurements were obtained with the DXR technology. The interradiograph reproducibility of DXR–BMD using conventional images under standardized conditions (6 mAs; 42 kV; 1 m FFD; film sensitivity of 200) was calculated to be coefficient of variation (CV) = 0.49% for Agfa Curix film and CV = 0.33% for Kodak T-MAT-Plus film, whereas reproducibility error using digital images ranged from CV = 0.57% (digital printouts; Philips) to CV = 1.50% (original digital images; Siemens). The intraradiograph reproducibility error was observed to be CV = 0.13% (conventional; Kodak film) vs CV = 0.27% (digital printouts; Philips). The BMD calculation was not noticeably affected by changes of FFD, exposure level, or film sensitivity/film brand, but was influenced by tube voltage (CV = 0.99% for Kodak film to CV = 2.05% for Siemens digital printouts). No significant differences were observed between the BMD and MCI data. DXR provides measurements of MCI and BMD with high precision and reproducibility. The measurements are unaffected by all tested image-capturing conditions, with the exception of tube voltage. In addition, different digital image devices clearly have an effect on DXR reproducibility. Key Words: Digital X-ray radiogrammetry; bone mineral density (BMD); Metacarpal Index; tube voltage; X-ray equipment.
Introduction Received 06/24/04; Revised 09/20/04; Accepted 09/2004. * Address correspondence to Joachim Böttcher, Institute of Diagnostic and Interventional Radiology, Friedrich-SchillerUniversity Jena, Erlanger Allee 101, 07747 Jena, Germany. E-mail: [email protected]
Osteoporosis results in low bone mineral density (BMD) and microarchitectural deterioration of cortical and trabecular bone tissue, leading to diminished biomechanical competence of the skeletal parts. It is common knowledge that osteoporosis 87
88 in postmenopausal women is characterized by both a reduction of cortical thickness and a decrease of trabecular bone volume (1). The clinical consequences of osteoporosis essentially comprise the onset of low-trauma or atraumatic fractures affecting primarily the spine, hip, and wrist, contributing to higher morbidity, mortality, and health care costs (2). Regarding the multifactorial pathophysiology of osteoporosis, BMD is the most important component of bone strength and accounts for up to 80% of its variance (3). Therefore, decreased bone mass is a useful predictor of increased risk for fracture (4). Different prospective studies revealed the association between the reduction of BMD and an increased risk for fracture by threefold (5–7). The bone densitometers are essential diagnostic tools for the detection of osteoporosis and for identification of highrisk patients and assessment of efficacy regarding treatment and prevention of bone loss. Considerable advancement has been made in noninvasive methods for evaluating BMD in the past 25 yr (5), with particular focus on the evaluation of dual-energy X-ray absorptiometry (DXA) (8), ultrasoundbased methods (7), and quantitative computed tomography (QCT) (9). Conventional radiographs have been long used to estimate changes in BMD; however, they are known for high inaccuracy (varying image-capturing conditions and individual physiological variations). Osteopenia can be verified by hand X-rays only at a visible BMD reduction of more than 35% (10). In 1960, a more precise technique to assess bone status from radiographs was introduced by Barnett and Nordin (11), followed by Virtama and Mahonen (12). They measured the sum of the two cortical thicknesses of the middle metacarpal, divided by the width (i.e., Barnett and Nordin Index or Metacarpal Index [MCI]). Recently, the clinical application of radiogrammetry has shown significant improvement after refinement, computerization, and the use of algorithms for automatic image analysis (13–15). Our study is performed with the Pronosco X-Posure System (Version 2; Sectra Pronosco A/S, Denmark), which uses a combined computerized radiogrammetric and textural analysis of the three middle metacarpal bones for the measurement of BMD and MCI. The potential of digital X-ray radiogrammetry (DXR) for estimating cortical bone loss seems to be promising. Several studies have reported not only a closed correlation between peripheral measured DXA data and DXR–BMD (R = 0.61) (16,17), but they also have published normative values for DXR (18–22). In addition, Rosholm et al. (15) showed a significant association between DXR–BMD and DXA–BMD with R = 0.86 (distal radius), R = 0.73 (femur), and R = 0.62 (lumbar spine) in healthy women. Although the measurement of DXR–BMD did not take place at a location of high fracture incidence, Bouxsein et al. (23) documented, in a prospective study over an observation period of 5 yr, that DXR showed an equal prediction value of fracture risk compared to single-photon absorptiometry regarding fractures of the wrist, spine, and femur. Journal of Clinical Densitometry
Böttcher et al. Moreover, Hyldstrup et al. (24) verified an increase of DXR–BMD and MCI with combined reduction of cortical porosity in postmenopausal women after treatment with bisphosphonates, whereas no improvement of bone partition could be documented for DXA estimated on the spine, femur, and forearm. In addition, postmenopausal women using hormone replacement therapy showed a significant increase of MCI calculated by DXR (24). Unfortunately, a majority of these studies are presumably based on an extensive variety of different image-capturing conditions. Yet, despite the increasing clinical importance of DXR, there is a lack of reliable data demonstrating the possible effects of different image-capturing parameters when acquiring radiographs for BMD calculation with the DXR technology. The present cadaver study aims to evaluate the influence of different technical parameters on the performance of DXR and provides data about the reproducibility of DXR.
Methods Variation of the Technical Parameters The left hands from two deceased males (authorized agreement given) were radiographed according to a predefined study protocol. The radiographs were acquired under a number of different image-capturing conditions, including the following: • • • •
Film-focus distance (FFD; conventional radiographs and digital printouts) Tube voltage (kV; conventional and original digital images vs digital printouts) Exposure level (mAs; conventional radiographs and digital printouts) Film type and film foil modality (conventional radiographs)
With the exception of tube voltage and exposure level, the parameters were varied while retaining other parameters as fixed. Three repeated radiographs were acquired for each hand, in each of the different capture conditions. The following four conventional film types were tested: Kodak MAT plus 200, Kodak MAT plus 400, Agfa Curix 200, and Agfa Curix 400 with film foils from Agfa. For the conventional radiographs, the study protocol (two subjects) included five variations of FFD (90 cm, 100 cm, 110 cm, 120 cm, 130 cm) keeping all other parameters stable (42 kV, 6 mAs, Kodak T-MAT Plus 200). Similarly for conventional radiographs, tube voltage and exposure level were varied using 40 kV, 42 kV, 44 kV, 46 kV, 48 kV, 50 kV, 52 kV, and accordingly, 3 mAs, 4 mAs, 5 mAs, 6 mAs, 7 mAs and 8 mAs, while keeping all other parameters stable (FFD = 100 cm; other parameters as mentioned earlier). All conventional images were acquired using the same Xray equipment (Philips Super 80 CP) and then the BMD and MCI were calculated. Digital radiographs were performed using the Diagnost Philips Optimus device as well as the Siemens Polydoros SX 80 equipment. The sensitivity of digital printouts regarding Volume 8, 2005
Influence of Image-Capturing Parameters on DXR image-capturing conditions was similarly tested (three repeated radiographs, two subjects) for tube voltage, FFD, and exposure level using the same fixed settings as for conventional radiographs. Because of the solely verified influence of tube voltage regarding digital printouts, the original digital radiographs—acquired with varied tube voltages— were also directly subjected (picture archiving and communication system [PACS]) to a DXR–BMD and MCI analysis (one subject).
Intraradiograph and Interradiograph Reproducibility Contrary to many protocols, our in vitro study differentiates between intraradiograph and interradiograph coefficients of variation (CVs). The interradiograph reproducibility of DXR–BMD was evaluated by acquiring 10 repeated radiographs with repositioning under standard settings: •
•
•
For conventional radiographs (Philips Super 80 CP), our study considered the following standardized parameters: 42 kV, 6 mAs, Kodak T-MAT Plus 200 or Agfa Curix 200, FFD = 100 cm (two subjects). For digital printouts (Diagnost Philips Optimus and Siemens Polydoros SX 80), the following constant imagecapturing conditions were observed for two subjects: 42 kV, 4 mAs, FFD = 100 cm. For original digital images (Diagnost Philips Optimus and Siemens Polydoros SX 80), the following constant image-capturing conditions were also observed for one subject: 42 kV, 4 mAs, FFD = 100 cm.
To verify intraradiograph CVs, a single conventional image and a single printout underwent 10 repeated BMD analyses (two subjects). Results were expressed as mean and standard deviation and precision errors as the CV of repeated measurements. The statistical analysis was performed using the SPSS Version 10.13.
Digital X-Ray Radiogrammetry Measurements of BMD and MCI were obtained from conventional and printed digital radiographs using automated DXR, implemented in the Pronosco X-posure System (Version V.2; Sectra Pronosco A/S, Denmark). The Pronosco X-posure System was originally calibrated for analysis of conventional radiographs. The computation of DXR–BMD from a conventional radiograph and from a digital printout first involves digitization by a flatbed scanner and, subsequently, an automated analysis of the digitized images. The three middle metacarpal bones are automatically located and regions of interests (ROIs) are placed around the narrowest part of the bones. In detail, the algorithm couples the three ROIs from each metacarpal bone with each other and moves them along the bone shaft to a position identified by the minimum combined bone width. Subsequently, the outer and inner cortical edges of the included cortical bone parts are detected. To locate the metacarpals in a radiograph, the DXR technology uses a model-based algorithm according to the Active Shape Model (25), which is adapted to Journal of Clinical Densitometry
89 differentiate the diaphysis of the three middle metacarpals. The analyzed image and the ROIs were displayed on the monitor. The heights of the ROIs were fixed to 2.0 cm, 1.8 cm, and 1.6 cm for the second, third, and fourth metacarpals, respectively. In each area, the cortical thickness and porosity was analyzed 118 times per centimeter. The combined cortical thickness of a single metacarpal bone was calculated as the sum of the ulnar and radial cortical thickness (measured in millimeters). Assuming that the bone is elliptical (26) and that the volumetric BMD is approximately constant, the bone volume per area and, consecutively, a BMD value are calculated from the cortical thickness and provided together with the average cortical thickness and average MCI of the three metacarpals. DXR–BMD is measured in grams per square centimeter, including a minor correction for the estimated porosity (15). MCI expresses the mean cortical thickness normalized with the mean outer bone diameter. The DXR technology checked the quality of the scanned images and aborted the examination in case of inadequate quality. There is no operator activity required for the location of a ROI nor is it possible for the operator to modify or influence the size or the location of the ROI. It should be noted that the Pronosco X-posure System is not intended to analyze printouts of digital radiographs, because the printing process can influence the image characteristics. In addition, the inherent resolution of the digital radiographs might not be appropriately compensated for in the printing process; thus, the measured size of objects in the printed films could deviate from the correct size. BMD and MCI from the original digital radiographs were obtained by DXR using an in-house software version (PACS-compatible) applying the same analysis, except that the step of digitization by a scanner was superseded and that the software did not fully adapt to the specific resolution of the digital radiographs, because the BMD and MCI values were not calibrated to give the same average values as with the Pronosco X-posure System, which is primarily adjusted for analysis of conventional radiographs. Notice, therefore, that the average DXR–BMD level from digital radiographs (printouts and also original analysis) was not directly comparable to the level for conventional radiographs.
Results No statistically relevant differences could be observed between the two individuals and all results are consequently calculated using the values from both hands (with exception of the analyses regarding the original digital radiographs).
Influence of Technical Parameters The coefficient of variation (CV) with respect to changes of FFD was 0.98% for DXR–BMD and 0.51% for MCI. The influence of different film brands and film sensitivities on precision was calculated to be 0.49% (film brand) and 0.33% (film sensitivity) for DXR–BMD and 0.76% vs 0.51% for MCI (see Table 1). In the case of digital Volume 8, 2005
90
Böttcher et al. Table 3 Influence of Tube Voltage on DXR–BMD and MCI for Conventional Radiographs Analyzed by DXR
Table 1 Influence of Film-Focus Distance, Film Brand, and Film Sensitivity on DXR-BMD and MCI for Conventional Radiographs Analyzed by DXR
Conventional BMD mean (g/cm2) SD CV (%) MCI mean SD CV (%)
Film-focus distance (N = 30)
Film brand (N = 12)
Film foil sensitivity (N = 12)
0.611 0.006 0.98 0.391 0.002 0.51
0.609 0.003 0.49 0.396 0.003 0.76
0.613 0.002 0.33 0.395 0.002 0.51
Conventional
BMD mean (g/cm2) SD CV (%) MCI mean SD CV (%)
Digital (printouts) (N = 36)
0.607 0.003 0.49 0.401 0.002 0.50
0.574 0.002 0.35 0.388 0.003 0.77
printouts, the CV was 0.86% by alterations regarding filmfocus distance. Our results also revealed a minor precision error for variation of the exposure level; for conventional radiographs, the CV scored 0.49% (BMD) vs 0.50% (MCI), and for digital printouts, the CV was 0.35% (BMD) vs 0.77% (MCI). Detailed results are given in Table 2. Altered tube voltages showed a noticeable CV for conventional and apparently for the digital radiographs (see Tables 3 and 4). For the conventional radiographs, the calculated variations of DXR–BMD were 1.32% (Agfa) and 0.99% (Kodak), whereas MCI variations were observed to be 1.01% (Agfa) and 1.00% (Kodak). For digital images with varied tube voltages, calculations of CV were solely subdivided into the modality of calculation via printouts and direct analysis of the original digital radiographs (without a scanning procedure). Printouts showed a precision error for DXR–BMD of 0.57% for the Philips device and a higher CV of 2.05% for the Siemens equipment. The BMD values based on original digital images showed minor precision errors for both devices compared to printouts (see Table 4). The following average correction factors could be calculated for one step of the tube voltage (± 1 kV) based on the tube voltage of 42 kV: Journal of Clinical Densitometry
Kodak T MAT Plus (N = 24)
0.608 0.008 1.32 0.400 0.004 1.01
0.608 0.006 0.99 0.400 0.00 1.00
BMD mean (g/cm2) SD CV (%) MCI mean SD CV (%)
Table 4 Influence of Tube Voltage on DXR–BMD and MCI for Digital Radiographs Analyzed by DXR
Table 2 Influence of Exposure Level on DXR–BMD and MCI for Conventional and Digital Radiographs (Printouts) Analyzed by DXR Conventional (N = 36)
Agfa Curix (N = 36)
Digital Printouts (two subjects)
Original (PACS) (one subject)
Philips Siemens Philips Siemens (N = 42) (N = 32) (N = 19) (N = 15) BMD mean (g/cm2) SD CV (%) MCI mean SD CV (%)
• •
0.524 0.003 0.57 0.383 0.003 0.78
0.585 0.012 2.05 0.386 0.010 2.59
0.597 0.003 0.50 0.470 0.002 0.43
0.594 0.008 1.35 0.464 0.006 1.29
For DXR–BMD: −0.48% (conventional, Kodak), −0.75% (conventional, Agfa), +0.11% (digital printouts, Philips), and −0.92% (digital printouts, Siemens). For MCI: −0.35% (conventional, Kodak), −0.74% (conventional, Agfa), +0.04% (digital printouts, Philips), and −1.28% (digital printouts, Siemens).
Additionally, the Pronosco X-posure System was not able to analyze conventional images with tube voltages of 48 kV (three radiographs for Kodak), 50 kV (three radiographs for Kodak), and 52 kV (three radiographs for Agfa Curix and Kodak, respectively). No results could be calculated for digital printouts made by the Siemens device with tube voltages of 44 kV (one radiograph) and 48 kV vs 50 kV (two radiographs). No results were estimated for original digital images (PACS) by the Siemens device with tube voltages of 40 kV (two radiographs), 44 kV (one radiograph), 46 kV (one radiograph), 48 kV (one radiograph), and 50 kV (one radiograph), whereas missegmented images occurred using the Philips device with tube voltages of 48 kV (one radiograph) and 52 kV (one radiograph). Volume 8, 2005
Influence of Image-Capturing Parameters on DXR
91
Table 5 Interradiograph Reproducibility for Conventional Radiographs Analyzed by DXR Agfa Curix (N = 20) BMD mean (g/cm2) BMD SD CV (%) MCI mean SD CV (%)
Kodak T MAT Plus (N = 20)
0.614 0.003 0.49 0.404 0.002 0.50
0.611 0.002 0.33 0.401 0.002 0.50
Table 6 Interradiograph Reproducibility for Digital (Printouts vs Original) Radiographs Analyzed by DXR
Original (PACS) (one subject)
Philips Siemens Philips Siemens (N = 20) (N = 20) (N = 10) (N = 10) BMD mean (g/cm2) SD CV (%) MCI mean SD CV (%)
0.523 0.003 0.57 0.383 0.003 0.78
0.591 0.007 1.18 0.392 0.005 1.28
0.602 0.002 0.33 0.387 0.001 0.26
0.599 0.009 1.50 0.389 0.006 1.54
Reproducibility Our study differentiates between interradiograph and intraradiograph CVs.
Interradiograph Reproducibility •
•
•
For conventional radiographs, the interradiograph reproducibility for the BMD was CV = 0.49% (Agfa) and CV = 0.33% (Kodak) when analyzing 10 different images obtained from the same object. Similar results were shown for MCI (see Table 5). For digital printouts we found comparatively small precision errors regarding MCI (CV = 0.78% vs CV = 1.28%) and BMD measurements with CV = 0.57% vs 1.18% depending on digital X-ray equipment (see Table 6). For original digital radiographs, the CV ranged between 0.33% vs 1.50% (BMD) and 0.26% vs 1.54% (MCI) using an in-house PACS compatible software version for the DXR measurements (see Table 6).
Intraradiograph Reproducibility When analyzing one identical image 10 times, our data revealed the following low CVs (see Table 7): Journal of Clinical Densitometry
Conventional
Digital (Printouts)
Agfa Curix Kodak T Philips Siemens MAT Plus BMD mean (g/cm2) SD CV (%) MCI mean SD CV (%) •
Digital Printouts (two subjects)
Table 7 Intraradiograph Reproducibility (Measuring One Identical Image 10 Times) for Conventional and Digital Radiographs (Printouts) Analyzed by DXR
•
0.608 0.0016 0.26 0.399 0.0013 0.33
0.613 0.0008 0.13 0.402 0.0007 0.17
0.528 0.0014 0.27 0.386 0.0016 0.42
0.578 0.0003 0.05 0.380 0.0005 0.13
For conventional radiographs, the CV was 0.26% (Agfa) vs 0.13% (Kodak) with respect to BMD and 0.33% (Agfa) vs 0.17% (Kodak) regarding MCI. For digital printouts, the intraradiograph reproducibility of BMD was CV = 0.27% (Philips) vs 0.05% (Siemens) and ranged from 0.42% (Philips) to 0.13% (Siemens) regarding MCI.
Discussion DXR allows for detailed measurements of distances in a radiograph, partly as a result of the high resolution and the averaging over a larger number of individual geometrical measurements. Measurements of DXR can also be performed on standard radiographs (with an effective radiation dose of <0.01 mSv for acquisition of an standard forearm radiograph), thereby giving this technique the potential to be widely available (17). Many studies have confirmed the relevance of DXR as an advantageous diagnostic tool for the quantification of osteoporosis regarding age- and menopause-related decreases of BMD (15,16,20–23) and also for the detection of diseaserelated BMD loss (27). In addition, DXR is not influenced by beam hardening and is insensitive to soft tissue variations. In contrast to DXA, DXR works without an aluminum wedge, and because of the fully automated procedure, it is independent of operator-induced variations in the BMD calculations (28). In view of these advantages, in 1999 the US Food and Drug Administration approved DXR as a clinical method for estimating BMD. However, radiogrammetry is principally different from the usual bone densitometry, because it involves measurements of geometrical dimensions, whereas densitometry evaluates mineral densities. One major advantage of radiogrammetry is the higher spatial resolution of the digitized radiographs compared to the resolution of densitometry, allowing for a better separation of cortical and trabecular partitions, together with a more precise edge detection. Various improvements to the initial radiogrammetric method (11) have been suggested, including averaging over Volume 8, 2005
92 multiple measuring sites on the same bone and averaging over several bones (29). Reportedly, the CV has been improved by 11% for one measurement on a single metacarpal to 8.4% for the average over nine measurements made on three metacarpal bones. An optical magnifier has been used to provide a higher visual resolution (30). Aguado et al. (31) have used digitally enlarged radiographs to improve the visual resolution with a CV value of 2.4%. In the past several years, there has been increased interest in radiogrammetry because of the availability of automated or semiautomated algorithms based on digital X-ray equipment. The reduced observer dependence of DXR has improved the CV to 1.5% (repeated measurements on different images) and 0.89% (repeated measurements on the same image) as published by Maggio et al. (32) and Derisquebourg et al. (33), respectively. Only our data and those of Maggio et al. (32) differentiated between intraradiograph (CV for MCI: 1.5%) and interradiograph reproducibility (CV for MCI: 5.0%). Our intraradiograph and interradiograph reproducibilities, addressing both acquisition technique as well as analysis of BMD and MCI, demonstrated high values, suggesting the excellent precision of DXR for conventional as well as digitally acquired radiographs, similar to recently published studies (15,34). In these studies, the short-term CV of DXR is 0.24–0.65 % on a very low level (15,16,34), indicating that estimated bone loss was not based on the precision error of the osteodensitometric method itself. The high precision of DXR might partly be explained by the kind of algorithm and the procedure to detect the inner cortical edges of a bone (35). The inner cortical edges are used for calculation of the cortical thickness of the metacarpals and, consequently, the BMD estimate. In the image analysis, the inner cortical edge is associated with the intensity maximum of the intensity profile across the bone of interest. For a bone with thick cortical tissue, the intensity maximum is located on a broader and less curved top than for a bone with a thin cortical shell. This property implies that the position of the intensity maximum is more precisely defined on the narrow top of a thin bone than on the broader top of a thick bone (17). In the evaluation of CV values from other published studies, it is important to note that most likely the data have been obtained from repeated measurements of a single radiograph for one person, thereby neglecting possible variations in the results arising from different image-capturing conditions. Up to now, only a few studies (15,28,34) have reported on the potential influence of different imaging parameters on the calculation of BMD and MCI. Such information, however, is essential when using normative values of BMD interpretation, particularly with regard to the guarantee of reliable data. To achieve the most realistic study design possible, this study utilized the hands from two deceased males. Between the two hands, our data revealed no interobject differences, which is important when considering the reliability of DXR. Further investigations of the interobject variability using phantoms would be beneficial. Journal of Clinical Densitometry
Böttcher et al. In our study, technical parameters, potentially different between various sites during imaging, were varied. These included FFD, tube voltage, exposure level, film sensitivity, and film brand, as well as image developing devices. Other parameters such as film–object distance, spot size, and displacement of the object were not evaluated because these parameters do not vary in a high-quality imaging. Particularly when taking into account relative values of standard deviations (e.g., up to 10% of calculated mean in childhood [21]), our data document insensitivity of the DXR technology to altered exposure level, film brand, and film sensitivity, as well as varied FFD. Tube voltage appears to significantly influence the calculations of DXR–BMD and MCI, both for conventional radiographs and for digital images (printouts and original). The influence of tube voltage on DXR could be also verified in an in vitro phantom study (36). These results suggest that it is essential to use standard values of tube voltage that are similar between those images to be analyzed and those representing reference data. The fact that the DXR system did not recognize radiographs obtained by 48 kV, 50 kV, and 52 kV (conventional), 44 kV, 48 kV, and 50 kV (digital printouts), and 40 kV, 44 kV, 46 kV, 48 kV, 50 kV, and 52 kV (digital originals) emphasizes the importance of a standardized tube voltage. Most likely, this problem originates from minor contrasting of bone structures in suboptimally performed radiographs, whereby digital printouts seem to be more sensitive. Similarly, the MCI not only performs independently of body size variables (19) but also shows—widely comparable to DXR–BMD—minor CVs to most of the technical parameters altered during imaging, with the exception of tube voltage, as also mentioned by Nielsen (37). For the first time, our data have verified a decrease of precision for DXR–BMD and MCI when using different devices for digital imaging. The Siemens digital X-ray equipment showed not only a significantly reduced interradiograph reproducibility but also higher precision errors with a varied tube voltage (see Table 4); the larger precision error and the higher dependency from the tube voltage for Siemens images were likely the result of standard image processing applied to the images by the Siemens X-ray workstation. The DXR technology is dependent on a robust and consistent detection of edges. Image enhancement procedures that affect gray levels nonlinearly must, therefore, not be applied on images intended to be measured with DXR. This is ensured by defining a standard image processing protocol for each modality and, subsequently, verifying compliance to the protocol through digital imaging and communications in medicine (DICOM) tags before analyzing an image. This referred image processing protocol comprises edge enhancement, noise reduction, and other similar procedures that depend on the image contents. The short-term precision of the Siemens device will be affected by edge enhancement because it introduces a stochastic influence on the edge properties when obtaining repeated images of the same object (i.e., metacarpal). Because of the research purpose of the present study, the involved Siemens Volume 8, 2005
Influence of Image-Capturing Parameters on DXR images have been analyzed off-line without definition and use of a standard protocol excluding image enhancement. In this study, it was also revealed that the usage of printouts of digital images for BMD measurement should be limited. The reproducibility errors were significantly higher compared to conventional radiographs, with the exception of the analysis of the original digital images acquired by the Philips device. This phenomenon is explainable by a smearing effect of the image contents (especially edges) during the printing process, which influences the DXR results. The DXR technology actually provides a calculation of BMD and MCI using original digital images for direct analysis with an internal batch program version of the X-posure System. This software applies its own calibration constants to adapt to the different resolutions. All noise-related interaction from the printing process, unintended image enhancement, variations based on different scanner resolutions, and slight position alterations during the scanning process of radiographs could be avoided; consequently, the precision of DXR is significantly improved.
Summary DXR provides calculations of MCI and BMD with high intraradiograph and interradiograph reproducibility of conventional as well as digital radiographs; this technique has been shown to be insensitive toward alterations of film sensitivity, film brand, FFD, and exposure level. Only variation in the tube voltage demonstrated an influence on the calculation process for BMD and MCI in conventional and, particularly, digital printouts; also, different digital image devices clearly have an effect on DXR reproducibility. Constancy of tube voltage and the assortment of appropriate digital image devices are essential for achieving reliable results. The use of digital printouts for analysis is significantly limited by an increased smearing resulting from the printing process, but might be prevented by the use of an internal batch program version of the X-posure System (PACS-compatible) for DXR analysis of original digital images without a scanning procedure. The DXR technique has the potential to be an inexpensive and advantageous method for the assessment of osteoporosis in patients in different clinical settings, both prospective and retrospective (34).
Acknowledgments We would like to thank Pronosco-Sectra and Arewus (i.e., Mrs. M. Arens) for the use of the X- posure equipment. Finally, the authors would also like to thank Mr. D. Felsenberg, PhD (Berlin, Germany), Mr. C. C. Gluer, PhD (Kiel, Germany), Mr. S. Grampp, PhD (Wien, Austria), and Mr. H. Imhof, PhD (Wien, Austria) for their comments regarding our study.
References 1. Barger-Lux MJ, Recker RR. 2002 Bone microstructure in osteoporosis: transilial biopsy and histomorphometry. Topics Magn Reson Imaging 13:297–305. Journal of Clinical Densitometry
93 2. Cooper C, Atkinson EJ, Jacobsen SJ, O`Fallon WM, Melton LJ. 1993 Population-based study of survival after osteoporotic fractures. Am J Epidemiol 137:1001–1005. 3. Rice JC, Cowin SC, Bowman JA. 1988 On the dependence of the elasticity and strength of cancellous bone on apparent density. J Biomech 21:155–168. 4. Ross PD, Davis JW, Vogel JM, Wasnich RD. 1990 A critical review of bone mass and the risk of fracture in osteoporosis. Calcif Tissue Int 46:149–161. 5. Genant HK, Engelke K, Fuerst T, et al. 1996 Noninvasive assessment of bone mineral and structure: state of the art. J Bone Miner Res 11:707–730. 6. Gnudi S, Malavolta N. 2003 Comparison between T-score-based diagnosis of osteoporosis and specific skeletal site measurements: prognostic value for predicting fracture risk. J Clin Densitom 6:267–273. 7. Hans D, Genton L, Allaoua S, et al. 2003 Hip fracture discrimination study: QUS of the radius and the calcaneum. J Clin Densitom 6:163–172. 8. Sweeney AT, Malabanan AO Blake MA, et al. 2002 Bone mineral density assessment: Comparison of dual-energy X-ray absorptiometry measurements at the calcaneus, spine and hip. J Clin Densitom 5:57–62. 9. Braillon PM. 2002 Quantitative computed tomography precision and accuracy for long-term follow-up of bone mineral density measurements: a five yr in vitro assessment. J Clin Densitom 5:259–266. 10. Jergas M, Uffmann M, Escher H, et al. 1994 Interobserver variation in the detection of osteopenia by radiography and comparison with dual X-ray absorptiometry (DXA) of the lumbar spine. Skeletal Radiol 23:195–199. 11. Barnett E, Nordin B. 1960 The radiological diagnosis of osteoporosis: a new approach. Clin Radiol 1:166–174. 12. Virtama P, Mahonen H. 1960 Thickness of the cortical layer as an estimate of mineral content of human finger bones. Br J Radiol 6:60–62. 13. Kalla AA, Meyers OL, Parkyn ND, Kotze TJ. 1989 Osteoporosis screening—radiogrammetry revisited. Br J Rheumatol 28:511–517. 14. Kiel DP, Hannan MT, Broe KE, Felson DT, Cupples LA. 2001 Can metacarpal cortical area predict the occurence of hip fracture in women and men over 3 decades of follow-up? Results from the Framingham Osteoporosis Study. J Bone Miner Res 16:2260–2266. 15. Rosholm A, Hylsdrup L, Baeksgaard L, Grunkin M, Thodberg HH. 2001 Estimation of bone mineral density by digital X-ray radiogrammetry: theoretical background and clinical testing. Osteoporos Int 12:961–969. 16. Black DM, Palermo L, Sorensen T, et al. 2001 A normative reference database study for the Pronosco X-posure radiogrammetry system. J Clin Densitom 4:5–12. 17. Jorgensen JT, Andersen PB, Rosholm A, Bjarnason NH. 2000 Digital X-ray radiogrammetry: a new appendicular bone densitometric method with high precision. Clin Physiol 20: 330–335. 18. Van Rijn RR, Grootfaam DS, Lequin MH, et al. 2004 Digital Radiogrammetry of the hand in a pediatric and adolescent dutch caucasian population: normative data and measurements in children with inflammatory bowel disease and juvenile chronic arthritis. Calcif Tissue Int 74:342–350. 19. Hyldstrup L, Nielsen SP. 2001 The metacarpal index by digital X-ray radiogrammetry: normative reference values and comparison with dual-X-ray absorptiometry. J Clin Densitom 4:299–306. 20. Maffei L, Venarotti H, Treviso J, Sorensen TK, Nissen D. 2000 Digital X-ray radiogrammetry: a Hispanic normative reference Volume 8, 2005
94
21.
22.
23.
24. 25. 26. 27.
Böttcher et al. database for the Pronosco X-posure system. J Bone Miner Res 15:304. Malich A, Freesmeyer MG, Mentzel HJ, et al. 2003 Normative values of bone parameters of children and adolescents using digital computer-assisted radiogrammetry (DXR). J Clin Densitom 6:103–112. Wüster C, Wenzler M, Kappes J, Rehm C, Gühring T, Arnbjerg C. 2000 Digital X-ray radiogrammetry as a clinical method for estimating bone mineral density—a German reference database. J Bone Miner Res 15:298. Bouxsein ML, Palermo L, Yeung C, Black DM. 2002 Digital Xray radiogrammetry predicts hip, wrist and vertebral fracture in elderly women: a prospective analysis from the study of osteoporotic fractures. Osteoporos Int 13:358–365. Hyldstrup L, Jorgensen JT, Sorensen TK, Baeksgaard L. 2001 Response of cortical bone to antiresorptive treatment. Calcif Tissue Int 68:135–139. Cootes TF, Taylor CJ, Cooper D, Graham J. 1995 Active shape models: their training and application. Comp Vision Image Understand 61:38–59. Lazenby R. 1995 Brief communication: Non-circular geometry and radiogrammetry of the second metacarpal. Am J Phys Anthropol 97:323–327. Böttcher J, Malich A, Pfeil A, et al. 2004 Potential clinical relevance of digital radiogrammetry for quantification of periarticular bone demineralization in patients suffering from rheumatoid arthritis dependend on severity and compared with DXA. Eur Radiol 14:631–637.
Journal of Clinical Densitometry
28. Thodberg HH, Jensen JK, Rosholm A. 1999 BMD from digital X-ray radiogrammetry: Sensitivity to details of the image capture. J Bone Miner Res 14:369. 29. Bloom RA. 1980 A comparative estimation of the combined cortical thickness of various bone sites. Skeletal Radiol 5: 167–170. 30. Memma HE. 1991 Improved vertebral fracture threshold in postmenopausal osteoporosis by radiogrammetric measurements: Its usefulness in selection for preventive therapy. J Bone Miner Res 6:9–14. 31. Aguado F, Revilla M, Villa LF, Rico H. 1997 Cortical bone resorption in osteoporosis. Calcif Tissue Int 60:323–326. 32. Maggio D, Pacifici R, Cherubini A, et al. 1997 Age-related cortical bone loss at the metacarpal. Calcif Tissue Int 60:94–97. 33. Derisquebourg T, Dubois P, Devogelaer JP, et al. 1994 Automated computerized radiogrammetry of the second metacarpal and its correlation with absorptiometry of the forearm and spine. Calcif Tissue Int 54:461–465. 34. Ward KA, Cotton J, Adams JE. 2003 A technical and clinical evaluation of digital X-ray radiogrammetry. Osteoporos Int 14:389–395. 35. Helboe AB, Juul K, Rosholm A, Andersen PB, Bjarnason NH. 2001 High short-term in-vivo precision of the Pronosco Xposure System on hand X-rays. J Bone Miner Res 28:335. 36. Böttcher J, Pfeil A, Lehmann G, et al. 2004 Clinical trial for differentiation between corticoid-induced osteoporosis and periarticular demineralization via digital radiogrammetry in patients suffering from rheumatoid arthritis. J Rheumatol 63:473–482. 37. Nielsen SP. 2001 The metacarpal index revisited: A brief overview. J Clin Densitom 4:199–207.
Volume 8, 2005