Volumetric Quantitative CT of the Spine and Hip Derived from Contrast-Enhanced MDCT: Conversion Factors

Bauer et al. QCT of the Spine and Hip Musculoskeletal Imaging Original Research Jan S. Bauer 1,2 Tobias D. Henning 2 Dirk Müeller 1 Ying Lu 2 Sharmila Majumdar 2 Thomas M. Link 2 Bauer JS, Henning TD, Müeller D, Lu Y, Majumdar S, Link TM Keywords: bone mineral density, dual-energy X-ray absorptiometry, femur, fractures, geriatrics, MDCT, osteoporosis, quantitative CT, spine DOI:10.2214/AJR.06.1006 Received July 31, 2006; accepted after revision December 6, 2006. 1 Department of Radiology, Technische Univerität München, Klinikum rechts der Isar, Institut für Roentgendiagnostik, Ismaninger Str. 22, München, Germany 81675. Address correspondence to J. S. Bauer (jsb@roe.med.tum.de). 2 Department of Radiology, University of California at San Francisco, San Francisco, CA. AJR 2007; 188:1294 1301 0361 803X/07/1885 1294 American Roentgen Ray Society Volumetric Quantitative CT of the Spine and Hip Derived from Contrast-Enhanced MDCT: Conversion Factors OBJECTIVE. The purposes of this study were to perform volumetric quantitative CT (QCT) of the spine and hip using nondedicated contrast-enhanced standard MDCT data sets and to derive a conversion factor for bone mineral density (BMD) assessment based on dedicated volumetric QCT data sets. SUBJECTS AND METHODS. Forty postmenopausal women with a mean ± SD age of 71 ± 9 years underwent routine contrast-enhanced abdominal and pelvic MDCT. Before this imaging examination, standard volumetric QCT of the spine (L1 L3, n = 40) and hip (n = 21) was performed. Relations between QCT and contrast-enhanced MDCT findings were assessed with linear regression analysis. RESULTS. Mean lumbar BMD was 84.1 ± 35.8 mg/ml, and mean femoral BMD was 0.62 ± 0.12 g/cm 2, as determined with QCT. Contrast-enhancement values with MDCT were on average 30.3% higher than those of QCT in the spine and 2.3% higher in the proximal femur (p < 0.05). Based on linear regression, a correlation coefficient of r = 0.98 was calculated for lumbar BMD with the equation BMD QCT =0.96 BMD MDCT 20.9 mg/ml. A coefficient of r =0.99 was calculated for the proximal femur with the equation BMD QCT =0.99 BMD MDCT 12 mg/cm 2 (p < 0.01). In 17 of 40 patients, 33 vertebral fractures were found. The dedicated QCT and enhanced MDCT data sets did not show a significant difference (p > 0.05) between patients with fractures and those without fractures. CONCLUSION. With the conversion factors, reliable volumetric BMD measurements can be calculated for the hip and the spine from routine abdominal and pelvic MDCT data sets. dequate screening methods are A needed to identify patients at high risk of osteoporotic fractures and to initiate appropriate therapy. Although initial assessment of osteoporotic fracture risk may be based on clinical evaluation, the diagnosis of osteoporosis is established by identifying vertebral insufficiency fractures or low bone mineral density (BMD) in the spine or the proximal part of the femur, which is measured with dual-energy X-ray absorptiometry (DEXA) [1]. The advantage of BMD assessment is that osteoporosis can be identified before a fracture occurs. Quantitative CT (QCT) is one of the standard techniques for assessing BMD of the lumbar spine and the proximal femur [2]. The quality and frequency of abdominal CT scans have increased substantially [3, 4]. Elderly patients frequently undergo abdominal CT for reasons other than BMD measurement [5]. In these studies, volumetric data sets are generated that include densitometric information. It may be possible to use this information for quantitative assessment of BMD. Previous studies [6, 7] have shown that routine abdominal single-detector helical CT can be used, with limitations, to determine BMD in the lumbar spine. MDCT has several advantages over single-detector helical CT. It has higher spatial resolution, and routine abdominal CT protocols yield thinner sections without additional radiation dosage. The aims of this prospective clinical study were to use standard contrast-enhanced abdominal MDCT data sets to generate volumetric 3D BMD data on the L1 L3 vertebrae and the proximal part of the femur, to correlate these data with standard 3D QCT measurements, and to determine the value of the calculated BMD data in differentiating patients with from those without vertebral fractures. Subjects and Methods Patients All studies were performed after approval of the study protocol by the institutional review board. 1294 AJR:188, May 2007

QCT of the Spine and Hip Written consent was obtained from all patients. Forty postmenopausal patients (mean age ± SD, 71 ± 9 years; age range, 60 88 years) who underwent either non-emergency clinically indicated abdominal (n = 19) or abdominal and pelvic (n =21) MDCT were enrolled consecutively in this prospective study. For all patients, previous imaging studies and clinically available data were thoroughly analyzed to exclude recruitment of patients with metastatic bone disease or a history of malignant bone marrow infiltration or radiation therapy to the lumbar spine and pelvis. Patients with known metabolic bone disease other than osteoporosis also were excluded. CT All examinations were performed with a 16- MDCT scanner (Mx8000 IDT, Philips Medical Systems) and a dedicated calibration phantom (osteoporosis phantom, (QCT Pro, Mindways Software). Before clinically indicated contrast-enhanced MDCT, dedicated volumetric QCT examinations of L1 L3 were performed on all patients. In a subset of 21 patients who also underwent MDCT of the pelvis, QCT scans of the proximal femur were obtained as well. A standard QCT protocol with 120 kvp, 90 mas, slice thickness of 3 mm, and 3-mm increment was used with a collimation of 16 1.5 mm (array with 16 detectors each 1.5 mm wide), pitch of 0.9, and rotation time of 0.75 second. After the dedicated QCT examination, the clinically indicated standard MDCT study was performed. Settings of 120 kvp and an absorption-adapted average effective tube current of 225 mas were chosen. Collimation, pitch, and rotation time were similar to those for QCT. Examinations were performed after standardized IV administration of contrast medium (Omnipaque 350, Amersham Health) with a delay of 80 seconds at a flow rate of 3 ml/s and a dose of 1 ml/kg of body weight up to a maximum dose of 150 ml. Patients also received 750 ml oral meglumine diatrizoate (Hypaque, E-Z-EM). In addition to the standard abdominal reconstruction (slice thickness, 5 mm), to obtain volumetric BMD, the MDCT data set was used to reconstruct the regions of the L1 L3 vertebrae and the proximal femur at a slice thickness of 3 mm with a standard kernel. BMD Analysis The images obtained with standard QCT and MDCT were reconstructed in an analogous manner and used to assess volumetric BMD of the L1 L3 vertebrae and the proximal part of the femur with commercially available software (QCT Pro, Mindways Software). Lumbar BMD was determined with oval regions of interest (ROIs) within the anterior three fourths of the trabecular compartment of the L1 L3 vertebrae, approximately 1 cm thick and equidistant to both endplates. Fractured vertebrae were excluded, resulting in analysis of 108 vertebrae. The ROIs were determined automatically, manually reviewed by a radiologist, and repositioned if necessary (Fig. 1). Repositioning of the ROIs was necessary for three patients during QCT and nine patients during MDCT. For the proximal part of the femur, 2D and 3D analyses of the QCT studies were performed. The bone was initially automatically segmented with a threshold-based algorithm. For the 2D analysis, this data set was projected and analyzed as in a standard DEXA examination. The total proximal femur and the neck, trochanter, intertrochanteric region, and Ward s triangle were evaluated separately (Fig. 2). For the 3D analysis, cortical and trabecular volumes of interest were assessed separately in the neck, trochanter, intertrochanteric region, and total proximal femur. Both femurs were analyzed in 13 patients. In eight patients, one side was excluded because of insufficient coverage by the MDCT scan: trochanter minor not included (n = 4), advanced degenerative joint disease (n = 3), and motion artifacts (n =1). All analyses were performed by one radiologist. Fracture Status To assess the fracture status of the spine, a lateral digital radiograph (scout view) of the thoracic and lumbar sections of the spine was obtained as previously performed by Gilsanz et al. [8, 9] and Link et al. [10]. In addition, sagittal reformations of the spine were reviewed because it has been shown that these reconstructions are highly accurate in the detection of vertebral fractures [11]. With this combination of techniques, vertebrae were assessed at least up to T8. Thus, according to Davis et al. [12], only approximately 10% of osteoporotic fractures of T1 T7 may have been missed. Fractures were classified by two radiologists in consensus according to the spinal fracture index described by Genant et al. [13]. Statistical Analysis Mean BMD values and SD were calculated from the QCT and MDCT studies of the lumbar spine and proximal part of the femur. A paired, two-tailed Student s t test was used to determine the significance of the differences between dedicated and contrast-enhanced scans. Relations between BMD values, determined separately with dedicated and contrast-enhanced scans, were evaluated with linear regression analysis. A linear fit based on a leastsquares algorithm was used to calculate QCT data from the MDCT values. To evaluate future prediction errors, 10-fold cross-calibration was performed whereby all patients were randomly divided into 10 groups. Nine of the 10 groups were used to develop a regression equation, and prediction errors were calculated on the basis of the group not used in the regression analysis. This procedure was rotated, and final prediction error (root mean square) was calculated as the root mean square of the single prediction errors of all analyzed samples in the 10 subgroups. Finally, a Bland-Altman plot was used to determine whether values of predicted and observed BMD were exchangeable and whether systematic bias was present [14]. Differences in BMD between patients with and those without vertebral fractures were assessed with an unpaired, two-tailed Student s t test. To evaluate potential differences in diagnostic performance, receiver operator characteristic (ROC) analyses were performed. Areas under the curve for dedicated and contrast-enhanced QCT scans were compared by use of 95% CIs [15]. All statistical computations were processed with JMP 5.1 (SAS Institute) and SPSS 11.5 (SPSS) software. Results In 17 of 40 patients, 33 vertebral fractures were found. The highest fracture grade was 3, found in two patients. Grade 2 fractures were found in four patients and grade 1 fractures in 11 patients. The mean volumetric BMD values, obtained from the first three lumbar vertebrae, were 84.1 ± 35.8 mg/ml for the QCT scans and 109.6 ± 36.6 mg/ml for the contrast-enhanced MDCT scans (Table 1, Fig. 3A), corresponding to a percentage difference of 31% (p < 0.001). Average BMD was 89.6 mg/ml for L1, 85.7 mg/ml for L2, and 78.4 mg/ml for L3. A coefficient of correlation of r = 0.98 (p < 0.01) was calculated between the averaged enhanced and unenhanced BMD measurements for the lumbar spine (Table 1). Linear fit was used to calculate the following equation: mean BMD QCT = 0.96 mean BMD MDCT 20.9 mg/ml. The related prediction error was 9% as calculated with 10- fold cross calibration. No significant differences were found for the single linear fits or prediction errors of the different vertebrae. The results of the Bland-Altman plot indicated that results of predicted and observed BMDs were exchangeable and that no systematic bias was present (Fig. 4). For 2D analysis of the proximal femur, the QCT value of the neck region was 0.54 ± 0.10 g/cm 2, and the corresponding MDCT value was 0.56 ± 0.11 g/cm 2. The QCT value was 0.62 ± 0.12 g/cm 2 for the total proximal femur region, and the MDCT value was 0.63 ± 0.12 g/cm 2 (Table 1). For all measurements at the proximal femur, the coefficient of correlation was r =0.99 (p < 0.001) (Fig. 3B). Linear fit was used to calculate the following equation for all 2D ROIs: mean BMD = 0.99 mean BMD MDCT 0.012 g/cm 2. AJR:188, May 2007 1295

Bauer et al. A B C D E F G H I Fig. 1 69-year-old woman with colon cancer. Evaluation of lumbar bone mineral density in 3D MDCT data sets. Although bone mineral density measurements were significantly different visually, no differences are evident. L1 exhibits hemangioma and was excluded from evaluation. A C, Dedicated quantitative CT scans show L1 (A), L2 (B), and L3 (C). D F, CT slice selections in sagittal view corresponding to A C show L1 (D), L2 (E), and L3 (F). G I, Contrast-enhanced MDCT scans show L1 (G), L2 (H), and L3 (I). 1296 AJR:188, May 2007

QCT of the Spine and Hip A B C Fig. 2 69-year-old woman with colon cancer. Use of CT data sets for evaluation of bone mineral density (BMD) in hip. A, Dual-energy X-ray absorptiometry like projection image for 2D evaluation. B, Dedicated quantitative CT scan shows 3D BMD evaluation. C, Contrast-enhanced CT scan shows 3D BMD evaluation. Enhanced blood vessels are evident. On average, the MDCT values were 0.014 g/cm 2 (2%) higher than the QCT values (p < 0.01) (Table 1). The specific coefficients of correlation for the single regions were r = 0.98 for the neck region, r = 0.99 for the trochanteric region, r = 0.99 for the intertro- TABLE 1: Comparison of Bone Mineral Density Values Calculated on Dedicated Quantitative CT (QCT) Scans and Contrast-Enhanced MDCT Scans Root Mean Square Location, Bone Type QCT a MDCT a Difference a Error (%) r Lumbar spine 3D, trabecular L1 89.6 ± 36.1 117.2 ± 37.8 28.1 ± 21.3 (31.4) 10.8 0.97 L2 85.7 ± 36.6 112.1 ± 36.9 24.1 ± 07.8 (28.1) 8.1 0.98 L3 78.4 ± 36.3 103.4 ± 36.7 25.1 ± 14.7 (32.0) 9.5 0.98 L1 L3 (average) 84.1 ± 35.8 109.6 ± 36.7 25.5 ± 07.1 (30.3) 9.0 0.98 Femur 2D, integrated Neck 0.54 ± 0.10 0.56 ± 0.11 0.02 ± 0.02 (4.1) 6.8 0.98 Trochanter 0.47 ± 0.10 0.48 ± 0.10 0.01 ± 0.02 (2.4) 6.5 0.99 Intertrochanteric region 0.75 ± 0.16 0.77 ± 0.16 0.02 ± 0.03 (3.0) 5.3 0.99 Total proximal femur 0.62 ± 0.12 0.63 ± 0.12 0.01 ± 0.02 (2.3) 13.2 0.98 Ward s triangle 0.40 ± 0.12 0.43 ± 0.13 0.02 ± 0.02 (6.1) 6.0 0.99 Femur 3D Trabecular Neck 110.9 ± 19.1 116.9 ± 22.7 06.0 ± 09.2 (5.4) 13.2 0.92 Trochanter 104.8 ± 22.6 109.6 ± 17.3 04.8 ± 17.4 (4.6) 5.1 0.97 Intertrochanteric region 106.2 ± 18.7 112.4 ± 21.0 06.2 ± 05.4 (5.8) 12.3 0.97 Total proximal femur 107.5 ± 16.6 111.8 ± 18.6 04.4 ± 04.0 (4.1) 8.9 0.98 Cortical Neck 1127.1 ± 275.2 1039.0 ± 398.4 47.2 ± 397.5 ( 4.2) 20.3 0.60 Trochanter 1078.7 ± 301.4 1052.0 ± 737.3 74.8 ± 903.9 (6.9) 34.3 0.75 Intertrochanteric region 947.3 ± 82.6 880.3 ± 57.0 67.0 ± 49.1 ( 7.1) 12.7 0.81 Total proximal femur 1076.5 ± 260.0 935.8 ± 136.0 140.8 ± 146.6 ( 13.1) 21.2 0.91 a Average ± SD in milligrams per milliliter (3D measurements) and in grams per square centimeter (2D measurements). Values in parentheses are percentages. chanteric region and r = 0.98 for the total proximal femur region (all p < 0.01) (Table 1). The related prediction error calculated with 10-fold cross calibration ranged from 5.3% for the intertrochanteric region to 13.2% for the total proximal femur ROI. Three-dimensional volumetric BMD of the proximal femur was calculated separately for the cortical and trabecular compartments in the specific regions. QCT values for the trabecular compartment were 110.9 mg/ml in the neck and 107.5 mg/ml in the entire proximal femur, the MDCT values being 5.4% higher in the neck (116.9 mg/ml) and 4.1% higher in the entire proximal femur (111.8 mg/ml) (Table 1). Although high correlations between MDCT and QCT with small prediction errors were found for the trabecular compartment (r = 0.92 to 0.98; root mean square, 5.1 13.2%), low correlations with high prediction errors were found for the cortical compartment because of segmentation problems, surrounding enhancing soft tissue being included in the region analyzed (r = 0.60 to 0.91; root mean square, 12.7 34.3%) (Table 1). BMD values of the spine were compared in patients with and those without fractures. These values were significantly lower in patients with osteoporotic fractures, both for the quantitative and the enhanced MDCT scans (p < 0.05) (Table 2). The diagnostic power was similar for the two techniques (p > 0.05). The specific areas under the curve were 0.71 for the contrast-enhanced and 0.70 for the unenhanced scans. The degree of change in density after contrast ad- AJR:188, May 2007 1297

Bauer et al. 1.2 BMD (QCT, mg/cm 3 ) 150.0 100.0 50 50 50.0 150 BMD (MDCT, mg/cm 3 ) A Fig. 3 Plots of bone mineral density (BMD) without (quantitative CT [QCT]) and with (MDCT) contrast enhancement. A, Graph shows linear fit for L1 L3 vertebral bodies of 40 patients (n = 108, r = 0.98, p < 0.001). Squares indicate L1; triangles, L2; crosses, L3. B, Graph shows linear fit for different regions of hips (n =170, r =0.99, p < 0.001) of 21 patients (n = 34, 13 bilateral, eight single hip examinations). Diamonds indicate total proximal femur; triangles, trochanter; crosses, intertrochanteric region; squares, neck; circles, Ward s triangle. Predicted Error (SD) 0.5 0.0 0.5 0 A Fig. 4 Bland-Altman plots. Squares indicate L1; triangles, L2; crosses, L3. A, Graph shows bone mineral density (BMD) measured with quantitative CT and predicted with MDCT findings are exchangeable because no significant difference is present (p > 0.05). B, Graph shows no systematic error is present. ministration in patients with fractures was not significantly different from that in patients without fractures (p > 0.05). However, a trend of greater contrast enhancement was found for L3 in patients without fractures (p = 0.10). 100 200 BMD (mg/cm 3 ) BMD (QCT, g/cm 2 ) Predicted Error (mg/cm 3 ) 0.8 0.4 0.0 0.0 0.4 20.0 0.0 20.0 0 BMD (MDCT, g/cm 2 ) 0.8 1.2 100 200 BMD (mg/cm 3 ) Discussion The results of this study showed that BMD data on the proximal part of the femur and the lumbar spine obtained from routine abdominal MDCT scans yield highly significant correlations with BMD values obtained from dedicated QCT examinations. Thus the findings suggest that contrast-enhanced MDCT can be used to obtain reliable BMD data. Because CT is one of the most frequently used radiographic techniques, particularly among elderly patients, additional BMD quantification, which so far is B B 1298 AJR:188, May 2007

QCT of the Spine and Hip TABLE 2: Vertebral Bone Mineral Density (BMD) Measurements in Patients With (n =17) and Without (n = 23) Vertebral Fractures with Results of Receiver Operating Characteristic Analysis Fracture No Fracture Area Under the Curve BMD (mg/ml) Enhancement BMD (mg/ml) Enhancement Vertebral Level QCT MDCT (H) QCT MDCT (H) QCT MDCT L1 72.4 ± 38.1 101.5 ± 44.7 28.3 ± 30.7 99.3 ± 31.9 127.2 ± 29.5 28.0 ± 10.9 0.68 (0.49 0.87) 0.72 (0.53 0.90) L2 69.1 ± 39.1 96.9 ± 41.5 23.0 ± 10.0 97.2 ± 30.6 122.1 ± 30.5 24.9 ± 05.9 0.68 (0.50 0.87) 0.70 (0.52 0.88) L3 64.2 ± 34.8 86.8 ± 34.7 25.4 ± 15.2 89.1 ± 34.5 116.5 ± 33.4 24.9 ± 14.6 0.71 (0.52 0.90) 0.68 (0.48 0.87) L1 L3 (average) 68.4 ± 36.4 93.8 ± 39.7 25.4 ± 08.6 95.8 ± 31.3 121.3 ± 30.1 25.5 ± 05.9 0.71 (0.54 0.88) 0.70 (0.52 0.87 Note Values in parentheses indicate 95% CI. QCT = quantitative CT. not standard, may have great clinical significance for individual patients. Hip and spinal fractures are the most serious manifestations of osteoporosis, being associated with a 20% mortality rate and substantial loss of quality of life [16, 17]. In a clinical setting, osteoporosis can be diagnosed on the basis of BMD measurements and the presence of fragility fractures. QCT has several advantages over DEXA because true volumetric BMD data are obtained, as opposed to the areal BMD data obtained with DEXA. QCT therefore can be used to measure bone compartments separately. Volumetric BMD is not influenced by the size of the vertebral bodies or falsified by degenerative changes in the spine, including the presence of osteophytes and degenerative changes in the facet joints [18]. QCT has been shown to be highly sensitive for monitoring changes in the metabolically more active trabecular compartment [19]. In a 1-year trial [20] with 238 postmenopausal women, twofold to threefold greater change in trabecular BMD was shown with QCT compared with integrated BMD measured with DEXA. Regarding differentiation of patients with and without osteoporotic vertebral fractures, both longitudinal and cross-sectional studies [21 29] have shown superior diagnostic performance of QCT in comparison with DEXA. Clinical MDCT generates densitometric data that are used to assess mass lesions but so far have not been used in a standard procedure for assessment of bone density. Researchers have begun to use nondedicated CT to measure BMD in patients. Preliminary studies [6, 7, 30] have been performed, and high correlations between BMD measured with helical CT and that measured with QCT have been found. Technical limitations, however, have to be considered in the use of clinical scans for evaluation of BMD. The scans were obtained with IV contrast material, a higher effective tube current (200 300 mas) was applied, and the slice thicknesses and gantry tilts were different from those used in dedicated QCT examinations. All these factors decrease accuracy and increase precision error compared with the optimized dedicated QCT protocol. With the advent of MDCT scanners, it has become possible to obtain thin-section volumetric data sets from the original raw data without additional radiation dosage. Therefore, BMD can be determined directly from the 3D data set. In our study, gantry tilt was not needed, and slice thickness and evaluation software were similar in dedicated QCT and clinical MDCT. The evaluation software can be used to correct for differences in effective tube current with a conversion factor determined in a phantom calibration. In our study, this reduction in error substantially improved correlation between scans compared with correlations determined in previous studies. In 2000, Hopper et al. [6] obtained r 2 values of up to 0.82 between BMD derived from QCT and clinical CT scans. In 2004, Link et al. [7] improved the results to r 2 = 0.91. In our study, a coefficient of determination of r 2 =0.96 was obtained for the spine and r 2 =0.98 for the proximal femur. Prediction error determined with 10-fold cross calibration was 9% for the spine and 5% for the proximal femur. This remaining error was mainly caused by differences in contrast enhancement. In the hip, only a 2% increase in signal intensity due to contrast enhancement was found, compared with 31% in the spine. This finding may explain the differences in prediction error. Although use of IV contrast material was standardized in this study, variations in the blood supply of the vertebral bodies, and thus contrast enhancement, were evident, particularly in elderly patients. The variations in blood flow and contrast enhancement may be related to osteoporosis. A 2005 study [31] showed reduced blood flow in the third lumbar vertebral bodies of patients with osteoporosis compared with healthy persons. Thus differences in perfusion-related contrast enhancement of osteoporotic vertebrae may add to differences in BMD. The difference, however, may vary among specific patient populations. In particular, in patients with increased BMD caused by medication (e.g., bisphosphonates), the relation between blood flow and BMD may change. In our study, no significant difference was found between contrast enhancement in patients with fractures and that in patients without fractures. A trend was shown only for L3. Thus potential differences in contrast enhancement may be a limitation rather than an advantage. Our study had a small number of patients, and contrast-enhanced CT and dedicated QCT showed no significant differences (p > 0.05) in the diagnostic ability to differentiate the presence from the absence of osteoporotic vertebral fractures. In the hip, the trabecular and cortical compartments were analyzed separately. Contrast enhancement was substantially higher in the trabecular compartment (4%) than in the integrated cortical and trabecular bone of the hip (2%). This finding was expected because metabolism and blood flow of trabecular bone are approximately eight times greater than those of cortical bone [32]. The ROIs of the cortical compartment showed low correlation between MDCT and QCT and high prediction error. In addition, the average cortical BMD was lower for the contrast-enhanced MDCT scans. This finding was due to segmentation error. The surrounding soft tissue was enhanced on the MDCT scan and partially included in the ROI with use of the thresholdbased segmentation algorithm. Because the ROIs included tissue other than cortical bone, the BMD values were lower for some MDCT scans. Although this difference had no effect on the 3D trabecular and 2D integrated regions, BMD values for the 3D cortical mea- AJR:188, May 2007 1299

Bauer et al. surements could not be determined satisfactorily from contrast-enhanced MDCT scans. In the spine, the prediction error was highest in L1. No significant difference, however, was found for prediction errors and calibration formulas for different vertebral levels (p > 0.05). In earlier studies, the prediction error was higher in the lower lumbar vertebrae, where the spine is more angulated and partial volume effects may have greater influence [6, 7]. This error can be eliminated with 3D evaluation of the MDCT data set. A substantial percentage of the patient population undergoing MDCT examinations is at higher risk of osteoporosis than is a healthy population. In addition, many therapeutic regimens in cancer treatment carry the risk of osteoporosis [33]. This association will be considered and managed more frequently because advances in chemotherapy, radiation therapy, and surgery have led to substantial improvements in survival times. In most of these cases, MDCT is used to assess therapeutic success and disease progression; however, MDCT is not currently used to monitor the progression of osteoporotic changes. Other patient populations at high risk of osteoporosis are those undergoing organ and bone marrow transplantation, postmenopausal women, and men older than 60 years [34]. All of these patient populations would benefit from early assessment for osteoporosis because they undergo MDCT examinations more frequently than the general population. A combination of BMD measurement and routine clinical CT, as in this study, would be very useful to these patients [33]. This study had several limitations. Only a small patient population participated, and results were limited to the protocols for contrast administration and MDCT. Blood flow was not measured directly; thus this influence was not determined independently. Additional studies are needed to investigate the effect of differences in iodine doses, flow rates, delays, and scanners. We did not evaluate long-term reproducibility; thus this method is not yet recommended for follow-up scans. Although standard evaluation software was used, the ROIs were determined semiautomatically, and registration was not applied for exact matching of dedicated and contrast-enhanced images. As one study [35] has shown, reproducibility is expected to increase with image registration, particularly for 3D regions in the hip. In particular, analysis of the hip would benefit from improvements in software. In this study, only one femur could be analyzed in eight patients. In four cases, registration problems occurred because the lesser trochanter was not included in the MDCT scans. In summary, this study showed that volumetric BMD data on the lumbar spine can be reliably obtained from routine abdominal MDCT scans and may be used to quantify vertebral fracture risk. Although 3D cortical BMD of the proximal femur cannot be determined satisfactorily from contrast-enhanced images with currently available evaluation software, 2D integrated and 3D trabecular BMD can be predicted from MDCT scans. This algorithm may be clinically useful in evaluation of patients at risk of osteoporosis, such as men and postmenopausal women older than 60 years; long-term survivors of chemotherapy for cancer who are undergoing regular follow-up examinations; hypogonadal patients, such as those with a history of testicular or prostate cancer; and patients undergoing organ transplantation. Acknowledgments We thank Steve Blackspur of Mindways Software for help with data transfer of the reconstructions. We also thank Mary McPolin and Feye Wong for help with the CT scans. References 1. Link TM, Guglielmi G, van Kuijk C, Adams JE. Radiologic assessment of osteoporotic vertebral fractures: diagnostic and prognostic implications. Eur Radiol 2005; 15:1521 1532 2. Guglielmi G, Lang TF. Quantitative computed tomography. Semin Musculoskelet Radiol 2002; 6:219 227 3. United Nations Scientific Committee on the Effects of Atomic Radiation. Sources and effects of ionising radiation. In: Medical radiation exposures: sources. New York, NY: United Nations, 2000 4. Flohr T, Stierstorfer K, Raupach R, Ulzheimer S, Bruder H. Performance evaluation of a 64-slice CT system with z-flying focal spot. Rofo 2004; 176:1803 1810 5. Kopka L, Rogalla P, Hamm B. Multislice CT of the abdomen: current indications and future trends [in German]. Rofo 2002; 174:273 282 6. Hopper K, Ming-Peng W, Kunselman A. The use of clinical CT for baseline bone density assessment. J Comput Assist Tomogr 2000; 24:896 899 7. Link TM, Koppers BB, Licht T, Bauer J, Lu Y, Rummeny EJ. In vitro and in vivo spiral CT to determine bone mineral density: initial experience in patients at risk for osteoporosis. Radiology 2004; 231:805 811 8. Gilsanz V, Boechat M, Roe T, Loro M, Sayre J, Goodman W. Gender differences in vertebral body sizes in adults: biomechanical implications. Radiology 1994; 190:678 682 9. Gilsanz V, Loro M, Roe T, Sayre J, Gilsanz R, Schulz E. Vertebral size of elderly women with osteoporosis: mechanical implications and relationship to fractures. J Clin Invest 1995; 95:2332 2337 10. Link TM, Doren M, Lewing G, Meier N, Heinecke A, Rummeny E. Cross-sectional area of lumbar vertebrae in peri- and postmenopausal patients with and without osteoporosis. Osteoporos Int 2000; 11:304 309 11. Bauer JS, Muller D, Ambekar A, et al. Detection of osteoporotic vertebral fractures using multidetector CT. Osteoporos Int 2006; 17:608 615 12. Davis JW, Grove JS, Wasnich RD, Ross PD. Spatial relationships between prevalent and incident spine fractures. Bone 1999; 24:261 264 13. Genant HK, Wu CY, van Kuijk C, Nevitt MC. Vertebral fracture assessment using a semiquantitative technique. J Bone Miner Res 1993; 8:1137 1148 14. Bland JM, Altman DG. Statistical methods for assessing agreement between two methods of clinical measurement. Lancet 1986; 1:307 310 15. Hanley JA, McNeil BJ. The meaning and use of the area under a receiver operating characteristic (ROC) curve. Radiology 1982; 143:29 36 16. Center JR, Nguyen TV, Schneider D, Sambrook NP, Eisman JA. Mortality after all major types of osteoporotic fracture in men and women: an observational study. Lancet 1999; 353:878 882 17. Chrischilles E, Shireman T, Wallace R. Costs and health effects of osteoporotic fractures. Bone 1994; 15:377 386 18. Cann C. Quantitative CT for determination of bone mineral density: a review. Radiology 1988; 166:509 522 19. Genant H, Cann C, Ettinger B, Gordan G. Quantitative computed tomography of vertebral spongiosa: a sensitive method for detecting early bone loss after oophorectomy. Ann Intern Med 1982; 97:699 704 20. Black DM, Greenspan SL, Ensrud KE, et al. The effects of parathyroid hormone and alendronate alone or in combination in postmenopausal osteoporosis. N Engl J Med 2003; 349:1207 1215 21. Bergot C, Laval-Jeantet A, Hutchinson K, Dautraix I, Caulin F, Genant H. A comparison of spinal quantitative computed tomography with dual energy X- ray absorptiometry in European women with vertebral and nonvertebral fractures. Calcif Tissue Int 2001; 68:74 82 22. Duboeuf F, Jergas M, Schott A, Wu C, Gluer C, Genant H. A comparison of bone densitometry measurements of the central skeleton in post-menopausal women with and without vertebral fracture. Br J Radiol 1995; 68:747 753 23. Grampp S, Genant H, Mathur A, et al. Compari- 1300 AJR:188, May 2007

QCT of the Spine and Hip sons of noninvasive bone mineral measurements in assessing age-related loss, fracture discrimination, and diagnostic classification. J Bone Miner Res 1997; 12:697 711 24. Guglielmi G, Grimston S, Fischer K, Pacifici R. Osteoporosis: diagnosis with lateral and posteroanterior dual x-ray absorptiometry compared with quantitative CT. Radiology 1994; 192:845 850 25. Lafferty F, Rowland D. Correlations of dual-energy X-ray absorptiometry, quantitative computed tomography, and single photon absorptiometry with spinal and non-spinal fractures. Osteoporos Int 1996; 6:407 415 26. Ott S, Kilcoyne R, Chestnut C. Ability of four different techniques of measuring bone mass to diagnose vertebral fractures in postmenopausal women. J Bone Miner Res 1987; 2:201 210 27. Pacifici R, Rupich R, Griffin M, Chines A, Susman N, Avioli L. Dual energy radiography versus quantitative computer tomography for the diagnosis of osteoporosis. J Clin Endocrinol Metab 1990; 70:705 710 28. Rehman Q, Lang T, Modin G, Lane NE. Quantitative computed tomography of the lumbar spine, not dual x-ray absorptiometry, is an independent predictor of prevalent vertebral fractures in postmenopausal women with osteopenia receiving long-term glucocorticoid and hormone-replacement therapy. Arthritis Rheum 2002; 46:1292 1297 29. Yamada M, Ito M, Hayashi K, Ohki M, Nakamura T. Dual energy X-ray absorptiometry of the calcaneus: comparison with other techniques to assess bone density and value in predicting risk of spine fracture. AJR 1994; 163:1435 1440 30. Carr J, Shi R, Lenchik L, Langefeld C, Lange L, Bowden D. Validation of quantitative computed tomography for measurement of bone mineral density in the thoracic spine during a cardiac gated protocol for coronary vascular calcium (abstract). Radiology 2001; 221[program]:380 31. Griffith JF, Yeung DK, Antonio GE, et al. Vertebral bone mineral density, marrow perfusion, and fat content in healthy men and men with osteoporosis: dynamic contrast-enhanced MR imaging and MR spectroscopy. Radiology 2005; 236:945 951 32. Thomsen JS, Ebbesen EN, Mosekilde L. Zone-dependent changes in human vertebral trabecular bone: clinical implications. Bone 2002; 30:664 669 33. Pfeilschifter J, Diel I. Osteoporosis due to cancer treatment: pathogenesis and management. J Clin Oncol 2000; 18:1570 1593 34. Keilholz U, Max R, Scheibenbogen C, Wuster C, Korbling M, Haas R. Endocrine function and bone metabolism 5 years after autologous bone marrow/blood-derived progenitor cell transplantation. Cancer 1997; 79:1617 1622 35. Li W, Sode M, Saeed I, Lang T. Automated registration of hip and spine for longitudinal QCT studies: integration with 3D densitometric and structural analysis. Bone 2006; 38:273 279 AJR:188, May 2007 1301