Performance Evaluation of Material Decomposition With Rapid-Kilovoltage-Switching Dual-Energy CT and Implications for Assessing Bone Mineral Density

Transcription

1 Medical Physics and Informatics Original Research Wait et al. Assessment of Bone Mineral Density With Rapid-Kilovoltage- Switching DECT Downloaded from by on 5/14/18 from IP address Copyright ARRS. For personal use only; all rights reserved Medical Physics and Informatics Original Research John M. S. Wait 1,2 Dianna Cody 1,2 Aaron K. Jones 1,2 John Rong 1,2 Veerabhadran Baladandayuthapani 3 S. Cheenu Kappadath 1,2 Wait JMS, Cody D, Jones AK, Rong J, Baladandayuthapani V, Kappadath SC Keywords: bone mineral density, dual-energy CT, dual-energy x-ray absorptiometry, osteoporosis DOI:1.2214/AJR Received April 29, 214; accepted after revision September 28, 214. This work was supported in part by a research grant from GE Healthcare. V. Baladandayuthapani is partially supported by the M. D. Anderson Cancer Center (grant P3 CA16672). 1 Department of Imaging Physics, The University of Texas M. D. Anderson Cancer Center, 1155 Pressler, Unit 1352, Houston, TX 773. Address correspondence to S. C. Kappadath (skappadath@mdanderson.org). 2 The University of Texas Graduate School of Biomedical Sciences, Houston, TX. 3 Department of Biostatistics, The University of Texas M. D. Anderson Cancer Center, Houston, TX. AJR 215; 24: X/15/ American Roentgen Ray Society Performance Evaluation of Material Decomposition With Rapid-Kilovoltage-Switching Dual-Energy CT and Implications for Assessing Bone Mineral Density OBJECTIVE. The purpose of this article is to quantitatively investigate the accuracy and performance of dual-energy CT (DECT) material density images and to calculate the areal bone mineral density (abmd) for comparison with dual-energy x-ray absorptiometry (DEXA). MATERIALS AND METHODS. A rapid-kilovoltage-switching DECT scanner was used to create material density images of various two-material phantoms of known concentrations under different experimental conditions. They were subsequently also scanned by single-energy CT and DEXA. The total uncertainty and accuracy of the DECT concentration measurements was quantified by the root-mean-square (RMS) error, and linear regression was performed to evaluate measurement changes under varying scanning conditions. Alterations to accuracy with concentric (anthropomorphic) phantom geometry were explored. The sensitivity of DECT and DEXA to changes in material density was evaluated. Correlations between DEXA and DECT-derived abmd values were assessed. RESULTS. The RMS error of DECT concentration measurements in air ranged from 9% to 244% depending on the materials. Concentration measurements made off-isocenter or with a different DECT protocol were slightly lower ( 5%), whereas measurement in scattering conditions resulted in a reduction of 8 27%. In concentric phantoms, higher-attenuating material in the outer chamber increased measured values of the inner material for all methods. DECT was more sensitive than DEXA to changes in BMD at 2 mg/ml. Measurements of abmd using DECT and DEXA were highly correlated (R 2 =.98). CONCLUSION. DECT material density images were linear in response but prone to poor accuracy and biases. DECT-based abmd could be used to monitor relative change in bone density. F irst implemented in the late 197s [1], dual-energy CT (DECT) scans could, in principle, decompose the signal from each voxel into the density of two user-defined materials, assuming that only those materials are present in the voxel (e.g., iodine and water). One potential application of DECT material decomposition is to accurately assess bone mineral density (BMD) by modeling bone signal in terms of bone mineral (hydroxyapatite) and marrow. Although the composition of bone material in osteoporotic bone is similar that of normal bone, the trabecular bone volume (TBV) is reduced. Reduced TBV results in lower bone strength in mechanical stress tests, which translates into an increased risk of fracture [2 4]. Dual-energy x-ray absorptiometry (DEXA) is the current reference standard for assessing BMD. The DEXA scanner outputs a BMD in units of grams per centimeter squared, which is not the true measure of BMD but an areal estimate of bone mineral content within a given projection. Single-energy CT has also been investigated for BMD measurement, although it is used less frequently than DEXA. As a 3D technique, single-energy CT may not be as susceptible as DEXA to variations in cortical bone size or composition. However, single-energy CT is susceptible to beam-hardening artifacts, patient scatter, and the presence of fatty marrow, which can reduce measured BMD [5 7]. Like single-energy CT, DECT provides a 3D dataset, but the use of two different effective energies offers the theoretic possibility of reducing or eliminating beam-hardening artifacts [1]. More important, DECT has the ability to identify the composition of a voxel instead of only the net attenuation, potentially allowing more-accurate assessment of bone composition. However, DECT previously was limited by two sequential single-energy scans, with the potential for mis AJR:24, June 215

2 Assessment of Bone Mineral Density With Rapid-Kilovoltage-Switching DECT Downloaded from by on 5/14/18 from IP address Copyright ARRS. For personal use only; all rights reserved registration artifacts, and by user-defined (variable) image preprocessing techniques. A commercial DECT scanner is now available that creates a dual-energy image with a single rotation using a cathode that rapidly switches kilovoltage (HD75, GE Healthcare); the manufacturer s software processes the dual-energy data into material density images based on the selected two-material basis pairs. These basis pairs are generated by uploading the mass attenuation coefficients for the appropriate materials to the software. However, the performance of the material decomposition feature of the rapid-kilovoltageswitching DECT scanner and the resulting material density images have not been investigated extensively. In addition, to our knowledge, no studies have yet been published on characterizing BMD using the HD75 DECT scanner material density images. The objective of the current study was to quantitatively investigate the performance of gemstone spectral imaging material decomposition images to known changes in the composition of several two-material samples in a variety of conditions. We also evaluated its potential to measure BMD and areal BMD (abmd) and explored correlation with DEXA abmd. Materials and Methods Determination of Material Density Accuracy, Sources of Variation, and Correlation Seven uniform solutions of iodinated contrast agent (Optiray 32, Mallinckrodt Pharmaceuticals) and water, ethanol and water, dipotassium phosphate ( ) and water, and a 1 g/dl and water solution plus denatured ethanol were prepared in syringes. The materials are ubiquitously used as surrogates in phantom experiments for CT research: Optiray 32 for CT contrast agent, for bone, and ethanol for fat. The phantoms containing both solution and ethanol were intended to model the composition of trabecular bone tissue. All phantoms were used to evaluate the accuracy of the DECT material density images in the simplest case of decomposing various two-material samples into their constituent basis pairs. The specific phantom concentrations prepared are shown in Table 1. Data were acquired with the HD75 using the dual-energy gemstone spectral imaging protocol 6 (8- and 14-kVp beams with effective energies of 4 and 58 kev, respectively; 6 mas total with one third and two thirds time-weighting for the 8- and 14-kVp beams; axial acquisition; medium body bowtie filter; rotation time, 1 second; beam width, 4 mm; and volume CT dose TABLE 1: Summary of Two-Material (Basis Pair) Phantoms Prepared for Assessment of In-Air Accuracy Basis Pairs Investigated index, 33.4 mgy). Each syringe was placed in the central bore of a cm electron density (ED) phantom (model 62, CIRS) with the portion containing the solution suspended in air and aligned with the isocenter, as shown in Figure 1A. Eight 5-mm images were reconstructed into material density maps using the constituent basis pairs. The mass attenuation coefficients for each material in the basis pair were obtained from the National Institute of Standards and Technology database using XCOM [8]. For each phantom, the mean and standard deviation (SD) of solute concentration was measured in a circular ROI ( mm 2 ) drawn in the center of the syringe in the central image of the series. Four sources of error in the DECT material density or concentration measurements were evaluated for determining their accuracy: variation in phantom preparation (σ p ), variation between images in each acquisition (σ i ), variation between acquisitions (σ r ), and random variation in concentrations across the ROI (noise, σ n ). To estimate σ p, four 25% iodinated contrast phantoms and three 2.5% phantoms were independently prepared and scanned in the same manner as the other 3-mL phantoms. To estimate σ i, the SD of the mean measured concentration across the six central images was measured in the image stack of three concentrations (1%, 2%, and 3%) of each solute (or ethanol). To estimate σ r, a single 25% iodinated contrast phantom, a single 2.5% phantom, and the 25% solution phantom were scanned three times each on three separate dates, and the SD of the mean concentration was Percentage Solute by Volume Optiray 32/water /water Ethanol/water g/dl solution/ethanol Note Although 1 g/dl solution was considered the solute, for these phantoms the concentration of ethanol instead increased in 5% by volume increments. Optiray 32 is manufactured by Mallinckrodt Pharmaceuticals. = dipotassium phosphate. Fig. 1 Schematic of electron density (ED) phantom and solution-filled syringes used for assessment of material density accuracy. Plus sign (+) indicates isocenter of CT scanner. A, Diagram shows in-air configuration, in which syringe-phantom is inserted into central bore of ED phantom with portion containing solution suspended in air and aligned with isocenter. Other bores are filled with water or water-equivalent inserts. B, Diagram shows in-scatter configuration, in which portion of syringe containing solution is placed within ED phantom and aligned with isocenter. + A calculated. Finally, the SD of the mean concentration measurement was defined as σ n for each iodinated contrast agent, ethanol,, and solution. The total uncertainty of each mean concentration (σ t ) was then calculated by combining these four independent SDs in quadrature. For comparison, single-energy CT images were acquired sequentially at 12 kvp and 15 mas, and at 8 kvp and 2 mas, with the medium body filter immediately after the DECT scans on the same scanner. The corresponding means and SDs of the CT numbers were recorded for all image sets. The phantoms were also scanned with a DEXA unit (Discovery, Hologic) with the phantoms placed horizontally between five 3/8-inchthick (9.53-mm-thick) acrylic slabs on top and three slabs beneath to simulate patient scattering conditions in a lumbar spine measurement. The abmd was measured using the vendor-supplied software by manually drawing a rectangular ROI over the central axis of the phantom to segment the bone region. Total uncertainty (σ t ) was calculated as described previously for single-energy CT and DEXA measurements; however, only σ p and σ r apply to DEXA abmd measurements. The derived solute concentrations were compared with known concentrations using linear regression, and their Pearson correlation coefficient was computed. The probabilities of measured data yielding a slope of 1 and of a y-intercept equal to mg/ml were determined with an F test using statistical analysis software (Prism 6, GraphPad Software). Accuracy was quantified by the total root-mean-square (RMS) error as a percentage of the average concen- + B AJR:24, June

3 Wait et al. Downloaded from by on 5/14/18 from IP address Copyright ARRS. For personal use only; all rights reserved Derived Concentration (mg/ml) Known Concentration (mg/ml) tration of the solute across all relevant phantoms. Potential bias in derived solute concentrations and their limits of agreement with known concentrations were evaluated using Bland-Altman analysis [9]. Effect of Variation of Scan Parameters To investigate the effect of changing scan conditions on the accuracy of material density measurements, images of the 28 phantoms were obtained as previously described using a different gemstone spectral imaging protocol, distance from isocenter, and scattering conditions. To assess the effect of using a different gemstone spectral imaging protocol, we scanned each phantom with the gemstone spectral imaging protocol 5, which uses a different bowtie filter (large vs medium) and has a larger scan FOV (5 vs 36 cm) and a smaller volume CT dose index (32. vs 33.4 mgy) than the gemstone spectral imaging protocol 6. To assess the effect Equivalence Iodinated Contrast Agent Ethanol 1 g/dl Solution Fig. 2 Dual-energy CT material density image concentrations of solutes compared with known concentrations along with linear regression. Error bars (total uncertainty of mean concentration, σ t ) are indicated but are too small to be visible on iodinated contrast agent (Optiray 32, Mallinckrodt Pharmaceuticals) and measurements. Because intercept was statistically significantly different from only for ethanol, slopes for other materials were recalculated with intercept set to. Slope was 1.6 ±.4 for Optiray 32, 1.8 ±.2 for, 1.12 ±.1 for 1 g/dl solution, and 1.3 ±.9 for ethanol, with y-intercept of 252 ± 13 mg/ml for ethanol. Scatter Measurement (mg/ml) Large Phantom Small Phantom Equivalence Air Measurement (mg/ml) Difference (mg/ml) of phantom positioning, each syringe was scanned in air centered 1.5 cm above isocenter, and the results were compared with measurements made with the syringe located at isocenter. To assess the effect of scatter radiation, the portion of the syringe containing the two-material solution was placed within the central bore of the ED phantom (made from water-equivalent material) and aligned with the isocenter, whereas the surrounding bores were filled with either water-filled 6-mL syringes or water-equivalent inserts (Fig. 1B). To assess the effect of different scattering conditions (i.e., a smaller-sized phantom), only the central 18-cm diameter circular insert was scanned with the twomaterial solution and was placed within the central bore and other bores filled with either syringes containing water or water-equivalent inserts. For comparison, axial single-energy CT scans (12 kvp and 15 mas; and 8 kvp and 2 mas) Iodinated Contrast Agent Ethanol 1 g/dl Solution Known Concentration (mg/ml) Fig. 3 Bland-Altman plot of dual-energy CT concentration measurements compared with known concentrations. Graph shows differences (measured known) plotted against known concentrations. Error bars indicate total uncertainty of mean concentration (σ t ) measurements. Fig. 4 Comparison of dualenergy CT concentration measurements under various scattering conditions. Initial regression found y-intercept to be consistent with, so all lines have y-intercept set to. Slope and 95% CI of measurements made with entire electron density phantom (labeled Large with diameter 3 cm) and central insert (labeled Small with diameter 18 cm) compared to in-air measurements are.74 ±.2 and.91 ±.1, respectively. were acquired immediately after each DECT scan using the various scattering conditions. We predicted that measurements made in scattering conditions would have higher variances than those made in air, so σ t values of the phantoms in homogeneous scattering conditions in the ED phantom were reassessed using the same methods as before. For all scattering conditions, the total error for measurements was estimated with these σ t values. The derived solute concentration or Hounsfield units for each set of measurements was compared with the original concentration or Hounsfield unit measurement in air. Linear regression and F tests were applied to determine the slope with 95% CI of the correlation. The Pearson correlation coefficient was also derived for each fit. Determination of Concentric Phantom Accuracy We assessed the performance of the DECT material decomposition in imaging phantoms that mimic the concentric geometry of human bones. We prepared 13 concentric phantoms, each consisting of a 5-mL cylindric plastic vial containing solution placed inside a 5-mL plastic centrifuge tube containing iodinated contrast agent solution. The lower-attenuating inner solution of was intended to simulate trabecular bone; the greater-attenuating outer solution of iodinated contrast agent was intended to simulate cortical bone. Two sets of concentric phantoms were created, one with the concentration of the inner solution fixed at 1% by volume and the concentration of the outer solution increasing in increments of 5% per volume, from % to 3%, and the other set with a fixed 1% outer solution 1236 AJR:24, June 215

4 Assessment of Bone Mineral Density With Rapid-Kilovoltage-Switching DECT Downloaded from by on 5/14/18 from IP address Copyright ARRS. For personal use only; all rights reserved Inner Concentration (mg/ml) Concentration (Air) Concentration (Scatter) Expected (Air) Expected (Scatter) Uniform Contrast Agent Concentration (mg/ml) A Inner HU concentration and the inner solution concentration increasing in increments of 5% per volume, from % to 3%. DECT and single-energy CT scans were acquired with the concentric phantom placed in the ED phantom in air and in scattering conditions, as described previously. Average concentration and Hounsfield units were measured using GSI Viewer (GE Healthcare) with circular ROIs drawn within central (ROI diameter, mm 2 ) and annular (ROI diameter, mm 2 ) cross-section of the inner and outer cylindric volumes, respectively. For both the outer and inner solutions, the measured concentration or Hounsfield units was plotted against the measurement made in the uniform phantom with the equivalent concentration. Linear regression was performed, and the slope was compared with the expected value of 1 or. Sensitivity Comparison To compare the sensitivity of DECT and DEXA, 13 additional syringe-phantoms containing 3 ml of solution were prepared with concentrations of 5 / 2 x % by volume, where x = 12. Measurements were acquired on three separate dates with the DECT scanner in air and in scattering conditions and with DEXA. The concentrations for which the average measurement was greater than 2σ (95.5% CI) from the measurement corresponding to a pure water phantom were combined with the syringe-phantom measurements used in assessment of accuracy. We reestimated σ values HU (12 kvp) (Air) HU (12 kvp) (Scatter) HU (8 kvp) (Air) HU (8 kvp) (Scatter) Expected (12 kvp) (Air) Expected (12 kvp) (Scatter) Expected (8 kvp) (Air) Expected (8 kvp) (Scatter) Uniform Contrast Agent HU Fig. 5 Dual-energy CT measured concentration and single-energy CT Hounsfield unit measurements of fixed inner solution, with increasing outer iodinated contrast agent (Optiray 32, Mallinckrodt Pharmaceuticals) solution concentration. A, Slope and 95% CI of regression for in-air measurement is.3 ±.1, whereas those for in-scatter conditions are 1.1 ±.4. Slope of these measured values when compared with measurements in uniform phantoms was expected to be. B, Single-energy CT measured Hounsfield units at 8 and 12 kvp. Slope of in-air measurement regression is.8 ±.3 and.9 ±.6 at 12 and 8 kvp, respectively. Slope and 95% CI of regression for measurements in scattering conditions are.9 ±.3 and.12 ±.4 at 12 and 8 kvp. For both measurement conditions, slope of measured values when compared with measurements in uniform phantoms was expected to be. Measured Concentration (mg/ml) Air Scatter Known Concentration (mg/ml) Measured abmd (g/cm 2 ) for each averaged measurement, where σ i was estimated from a stack of images obtained of a 2.5% phantom in air and scattering conditions using the method described already, σ r was calculated separately for each solution, and σ n was calculated as before. We assumed that σ p was the same value as calculated previously for. Regression analysis was applied to the data, and the bestfit values for the slope and y-intercept with their 95% CIs were determined. We defined sensitivity as the smallest detectable change outside a 95% CI from a concentration associated with the normal BMD measurement of a postmenopausal woman [1, 11]. Sensitivity was determined separately for DECT in air and in scattering conditions and for DEXA Known Concentration (mg/ml) A B Fig. 6 Comparison of sensitivity measurements for dual-energy CT (DECT) and dual-energy x-ray absorptiometry (DEXA); 95% CIs are shown for each regression along with total uncertainty of mean concentration (σ t ) error bars (not visible for some measurements). A, Graph shows that y-intercepts for DECT regressions in-air and in-scatter were consistent with. Equations for DECT regression in air and scatter with 95% CI are y (in g/cm 2 ) = (1.8 ±.2)x (in mg/ml) and y (in g/cm 2 ) = (.73 ±.1)x (in mg/ml). B, DEXA areal bone mineral density (abmd) values plotted as function of concentrations. Equation for DEXA regression with 95% CI is y (in g/cm 2 ) = (2.23 ±.3) 1 3 x (in mg/ml) (8 ± 1) 1 2 (in g/cm 2 ). B AJR:24, June

5 Wait et al. Downloaded from by on 5/14/18 from IP address Copyright ARRS. For personal use only; all rights reserved TABLE 2: Components of the Dual-Energy CT (DECT) and Dual- Energy X-Ray Absorptiometry (DEXA) Variance Expressed as a Coefficient of Variance DECT/DEXA Combined Errors Dual-Energy CT Areal Bone Mineral Density We devised a method for processing and analyzing DECT material density images to estimate abmd (grams per centimeters squared) and evaluated its relationship with DEXA abmd. To generate DECT material density images in hydroxyapatite density, the European Spine Phantom (Quality Assurance in Radiology in Medicine) [12], was scanned using the gemstone spectral imaging protocol 6 and the -water basis pair. We then developed a conversion from measured density to the true hydroxyapatite density (known from the European Spine Phantom specifications) by fitting the measured values for the European Spine Phantom vertebral bodies with a linear function. Axial image stacks of the DECT hydroxyapatite images were reformatted without interpolation into coronal-plane images using ImageJ image analysis software [13]. The air and bone material had sufficiently high contrast so that a visual threshold selected the entire volume of bone material in the stack, and voxels below the threshold in each image were set to mg/ml. The coronal images in each stack were arithmetically summed into a single coronal image, and the summed values were multiplied by the voxel volume (milliliters) to calculate the total mass (N) of the material (milligrams), onto which a final threshold was applied to segment the bone or an ROI corresponding to those used for DEXA measurements. The total area (A) of the selected pixels (centimeters squared) was calculated for each summed image after segmentation. The estimate of abmd in hydroxyapatite density, expressed in grams per centimeters squared, on the basis of DECT material density images, was then calculated as follows: 1 g abmd = ( N )/ A. 1 mg The integration processing was applied to 34 DECT material density image sets consisting of Optiray 32 Ethanol Solution 1 g/dl DECT DEXA DECT DEXA DECT DEXA DECT DEXA Phantom preparation (σ p ) Image to image (σ i ) Scan to scan (σ r ) ROI noise (σ n ) Combined (σ t ) Note Data are percentages. The combined variance σ t is the sum of the individual variances added in quadrature. DECT measurements were acquired in air. = dipotassium phosphate. Optiray 32 is manufactured by Mallinckrodt Pharmaceuticals. three uniform phantoms of each solute composition (iodinated contrast agent,, and 1 g/dl solution), three phantoms from each set of concentric phantoms (including the shared 1% inner concentration and 1% outer iodinated contrast concentration phantom), the DEXA daily quality control phantom, and the European Spine Phantom. For arbitrary anthropomorphic samples, several bovine and porcine bones were scanned in the DECT using the gemstone spectral imaging protocol 6 and in the DEXA scanner between polymethylmethacrylate blocks, as previously described. DECT abmd values were plotted against DEXA abmd measurements to calculate Pearson correlation coefficients. Linear regression and Bland-Altman analysis were performed. Results Determination of Material Density Accuracy, Sources of Variation, and Correlation The magnitude of each source of variation for each uniform two-material phantom imaged with DECT and DEXA is summarized in Table 2. For noise (σ n ) and total variation (σ t ), the values reported are an average across all concentrations of a given set of phantoms except for pure water. The total variance of all solute concentration measurements except for ethanol was less than or equal to 2%. The material decomposition derived solute concentration of each basis pair and the known concentration of each material were strongly correlated (R 2 of each fit >.98; p <.1) (Fig. 2). The F test revealed that the fitted slope was statistically significantly different from 1 for all solutes (p <.5) and the intercept was statistically significantly different from only for ethanol. The RMS errors for contrast agent,, 1 g/dl solution, and ethanol were 9%, 1%, 12%, and 244%, respectively. The very large RMS error for ethanol apparently resulted from the large offset ( 252 mg/ml) seen using the ethanol-water basis pair. As shown in Figure 3 and Table 3, the bias in material decomposition derived solute concentration and the 95% confidence limits of agreement were relatively small for iodinated contrast agent and. However, both sets of phantoms containing ethanol exhibited large bias in measured concentrations. The mean bias from Bland-Altman analysis was consistent with mg/ml only for iodinated contrast agent. In comparison, the single-energy CT Hounsfield unit measurement was strongly correlated with the concentration of all solutes at 12 kvp (R 2 >.98) and 8 kvp (R 2 >.91), and DEXA abmd was also highly correlated with the concentration of each material (R 2 >.88). Effect of Variation of Scan Parameters DECT material density images using the different scan protocols were correlated (R 2 >.99), as were measurements with the phantom placed on and offset from isocenter (R 2 >.99). Fitting with a straight line yielded an offset consistent with and a slope near unity (slope ± 95% CI =.96 ±.5 and.954 ±.4, respectively). Scattering conditions underestimated solute concentration (by 26%) compared with TABLE 3: Bland-Altman Analysis of Dual-Energy CT (DECT) Material Decomposition Concentration Measurements Solute Average Bias (Measured Known) 95% Confidence Limits of Standard Error in Bias 95% Confidence Limits of Agreement Iodinated contrast agent 9 5, 24 23, 41 Ethanol , , , 43 12, 64 1 g/dl solution , , 143 Note The DECT material decomposition measurements are compared with known concentrations, measured in milligrams per milliliter. = dipotassium phosphate AJR:24, June 215

6 Assessment of Bone Mineral Density With Rapid-Kilovoltage-Switching DECT Downloaded from by on 5/14/18 from IP address Copyright ARRS. For personal use only; all rights reserved DEXA abmd (g/cm 2 ) Uniform Iodinated Contrast Agent Uniform 1 g/dl Solution Uniform Animal Bones European Spine Phantom DEXA Phantom DECT abmd (g/cm 2 ) measurements made in air (Fig. 4). The effect of scatter in the smaller phantom was less pronounced but still relatively large ( 9%). CT numbers measured in scattering material were also reduced relative to those in air at 12 kvp (slope ± 95% CI =.7 ±.1) and 8 kvp (slope ± 95% CI =.76 ±.4). Measurements made in the head insert were less disparate than those in the entire phantom when compared with air (slope ± 95% CI =.834 ±.7 at 12 kvp and.93 ±.4 at 8 kvp). Determination of Concentric Phantom Accuracy The concentration of iodinated contrast agent in the outer chamber measured while a fixed concentration of was present in the inner chamber was approximately 12% lower than that measured in the uniform iodinated contrast phantom in air. The concentration of in the inner chamber measured while a fixed concentration of iodinated contrast agent was present in the outer chamber was approximately 13% higher than that measured for uniform phantoms in air. Of particular interest, the concentration measurements increased in concentric phantoms in which this concentration was fixed but the surrounding iodinated contrast agent concentration was increased (Fig. 5A). Similar trends were observed in singleenergy Hounsfield unit measurements (Fig. 5B) (the lower than expected iodinated contrast agent measurements in phantoms for which the concentration was fixed were observed at 12 kvp only). Equivalence Regression Concentric ( ) Concentric (Iodinated Contrast Agent).4 A Fig. 7 Dual-energy CT (DECT) and dual-energy x-ray absorptiometry (DEXA) areal bone mineral density (abmd) values. A, Linear regression of DECT abmd with DEXA. Equation for regression is y (in g/cm 2 ) = (.95 ±.4)x (in g/cm 2 ) (.28 ±.9) (in g/cm 2 ). B, Bland-Altman plot of DECT and DEXA abmd values shows difference (DECT DEXA) plotted against average of two abmd values. Difference (g/cm 2 ) Sensitivity Comparison The concentration of solutions measured using DECT material density images (milligrams per milliliter) and DEXA (grams per centimeters squared) are shown as a function of the known concentration (from phantom preparation) in Figure 6. The normal BMD of postmenopausal women as assessed with quantitative CT is approximately 126 mg/ml [1]. DECT material density images were able to detect a change at the 95% CI as small as 2 mg/ ml from this value in both air and scattering conditions. The normal BMD of postmenopausal women as assessed with DEXA is approximately.99 g/cm 2 [11], which corresponds to a concentration of 48 mg/ ml from the regression analysis. DEXA could detect a change at the 95% CI as small as 1 mg/ml from this value. Dual-Energy CT Areal Bone Mineral Density A linear relationship between hydroxyapatite density of the three bone-equivalent inserts in the European Spine Phantom and DECT concentration was observed (R 2 >.98). The DECT concentration measurements of the animal bones and the subset of homogeneous and concentric phantoms were converted to hydroxyapatite concentrations using this relationship. DECT abmds were then calculated and plotted against the DEXA abmds (Fig. 7A). The abmd measurements of these two modalities were linearly correlated (R 2 =.98; p <.1). Bland-Altman analysis (Fig. 7B) estimated a mean difference of.25 g/cm 2 with a standard error of.5 g/cm 2, indicating consistent underestimation of DEXA abmd by DECT abmd. The 95% confidence limits of agreement between DEXA and DECT abmd were estimated as.16 and.57 g/cm Average (g/cm 2 ) Discussion The quantitative pixel values of DECT material density images were well correlated and increased linearly with known solute concentration. Material density image measurements of phantoms containing only water and iodinated contrast agent or coarsely followed the line of equivalence (intercept consistent with and slope only marginally different than unity). Although, for most materials, σ t values for individual DECT concentration measurements in air were small (< 2%), the mean RMS error for two-material phantoms of known concentrations was nonetheless greater than 9%, corresponding to the expected accuracy for the best case scenario where two-material decomposition would be performed using known-material basis pairs. In addition, large bias and wide limits of agreement in material density image concentration measurements were shown by the Bland-Altman analysis (Table 3). The limits of agreement for concentration (9 43 mg/ml) were found to be large relative to the mean change in concentration that signified osteoporosis in postmenopausal women when assessed with quantitative CT ( 48 mg/ml) [14]. This low level of accuracy could be problematic for direct clinical measurements of material concentrations. Measurements of ethanol concentration in ethanol-water phantoms, though linear, had a large RMS error (289 mg/ml) and offset; indeed, a large concentration of ethanol (26 ± 4 mg/ml) was measured for the % concentration phantom, when only distilled water was present. Distinguishing between ethanol and water is evidently problematic for the material density decomposition software. This may be explained by the very similar B AJR:24, June

7 Wait et al. Downloaded from by on 5/14/18 from IP address Copyright ARRS. For personal use only; all rights reserved mass attenuation coefficients of water and ethanol within the energy range of DECT x-rays. The effective energy of the 8- and 14-kVp beams used in the DECT acquisitions were estimated, according to the measured half-value layers of the single-energy beams, to be approximately 4 and 5 kev, respectively. The mass attenuation coefficients of water and ethanol are.268 and.243 g/cm 2 at 4 kev and.227 and.216 g/cm 2 at 5 kev, respectively. In comparison, the mass attenuation coefficient of is.933 g/cm 2 at 4 kev and.557 at 5 kev. Although only small variations ( 5%) were caused by a different gemstone spectral imaging protocol or phantom position relative to isocenter, larger variations ( 9 26%) in material density image measurements were observed between in-air and in-scattering conditions. The apparent sensitivity of DECT material density values to scatter conditions suggests the need for additional corrections before clinical implementation. For example, DECT scans of an individual patient may yield altered BMD values unrelated to changes in bone if the patient experienced weight loss or gain at follow-up. Dual-energy x-ray modeling would ideally compensate for effects such as the amount of nonbone tissue and scatter, but its similarity in behavior to single-energy CT points to limitations in the modeling algorithm used to construct the material decomposition images. Previous works on dual-energy imaging have also reported on limitations of material decomposition images according to the suitability or nature of the algorithm used to generate them [15]. The observed modulation of a fixed inner solution s concentration or Hounsfield unit measurements with the outer iodinated contrast agent concentration was apparent in both material density images and single-energy CT, indicating once again that the effect was present in single-energy CT and not adequately compensated for in the DECT material decomposition algorithm. The effect may be due, in part, to the filtered back projection used to reconstruct the DECT and single-energy CT images; because of the central location of, every line integral contains the high attenuation from the surrounding iodinated contrast agent. We postulate that some high-attenuation signal from iodinated contrast agent may have smeared across the central region containing during reconstruction. In a second set of concentric phantoms, where the iodinated contrast agent solutions of increasing concentration were placed inside the inner chamber and an outer solution concentration remained fixed, no modulation in measured concentration and Hounsfield units with iodinated contrast agent concentration was observed. DECT material density images could detect a smaller change in normal BMD in air and in scattering conditions ( 2 mg/ml) relative to DEXA ( 1 mg/ml), indicating higher theoretic sensitivity. Normal and osteo porotic bone densities correspond to concentrations of approximately 126 ± 24 and 79 ± 24 mg/ml, respectively [1]. The smallest detectable change with DECT in-air and in-scattering conditions ( 2 mg/ml) is well within the SD of the mean value for each population. The smallest detectable change of DEXA ( 1 mg/ml) is likewise sufficient for distinguishing between patients with and without osteoporosis. abmd estimates (grams per centimeters squared) from material density images correlated well with abmd (grams per centimeters squared) measured with DEXA. The slope (near unity,.95 ±.4) of the linear regression for DEXA abmd to the DECT abmd lends credibility to the DECT integration method and suggests that the two modalities indeed assess the same fundamental quantity: the total volumetric density of bone mineral. However, the limits of agreement of the DEXA abmd (.16 and.57 g/cm 2 ) were of the same magnitude as a change in DEXA abmd from normal to healthy bone (.3 g/ cm 2 ) [16], which limits the utility of a conversion between these two modalities for abmd measurements. Our approach to measure abmd with DECT was also limited in part by the regression function used to calibrate the hydroxyapatite density measurements from measurements. Hydroxyapatite is not soluble in water and was therefore not used in construction of phantoms used in this study; this precluded us from directly measuring the hydroxyapatite accuracy. By leveraging the volumetric nature of DECT it is possible to measure the BMD specifically for trabecular bone. This may increase its appeal over DEXA in detecting a change in trabecular BMD for postmenopausal women. Coupled with higher sensitivity, relative change of this DECT trabecular BMD may serve as a moresensitive predictor of fracture risk than the DEXA-based abmd metric. DECT analysis of BMD has yet to be extended to human patients. However, the correlation of DECT abmd with DEXA abmd with the potentially higher sensitivity of DECT is intriguing, and, therefore, additional investigation on improvements and corrections to further promote the clinical utility of this method are encouraged. Conclusion DECT material density measurements are linearly correlated with true concentrations in two-material models, but the accuracy is at best approximately 9%. The observed variations of DECT-derived concentrations under scatter and with phantom geometry suggest that DECT material density or concentration measurements are not suitable for clinical implementation without the application of additional corrections. Also, the material decomposition appears to be effective only for materials with mass attenuation coefficients that are well separated in the range of CT x-ray energies. Nonetheless, the DECT material density measurements were found to be theoretically more sensitive than DEXA abmd. DECT material density image derived abmd was also well correlated with DEXA abmd measurements. However, the 95% confidence limits of agreement between material density image derived abmd and DEXA abmd were too wide for the two measurements to be used interchangeably in a clinical setting. References 1. Alvarez RE, Macovski A. Energy-selective reconstructions in X-ray computerised tomography. Phys Med Biol 1976; 21: Parfitt AM, Mathews CH, Villanueva AR, Kleerekoper M, Frame B, Rao DS. Relationships between surface, volume, and thickness of iliac trabecular bone in aging and in osteoporosis: implications for the microanatomic and cellular mechanisms of bone loss. J Clin Invest 1983; 72: Li B, Aspden RM. Composition and mechanical properties of cancellous bone from the femoral head of patients with osteoporosis or osteoarthritis. J Bone Miner Res 1997; 12: Ebbesen EN, Thomsen JS, Beck-Nielsen H, Nepper-Rasmussen HJ, Mosekilde L. Lumbar vertebral body compressive strength evaluated by dual-energy X-ray absorptiometry, quantitative computed tomography, and ashing. Bone 1999; 25: Cann CE, Genant HK. Precise measurement of vertebral mineral content using computed tomography. J Comput Assist Tomogr 198; 4: Merritt RB, Chenery SG. Quantitative CT measurements: the effect of scatter acceptance and filter characteristics on the EMI 77. Phys Med 124 AJR:24, June 215

8 Assessment of Bone Mineral Density With Rapid-Kilovoltage-Switching DECT Biol 1986; 31:55 63 Smith R. Quantitative computed tomography in imagej.nih.gov/ij/. Published 213. Accessed Feb- 7. Glüer CC, Genant HK. Impact of marrow fat on assessment of osteoporosis. Semin Nucl Med ruary 17, 215 accuracy of quantitative CT. J Comput Assist To- 1987; 17: Cann CE, Genant HK, Kolb FO, Ettinger B. mogr 1989; 13: Hall ML, Heavens J, Cullum ID, Ell PJ. The range Quantitative computed tomography for prediction 8. Berger MJ, Hubbell JH, Seltzer SM, et al. XCOM: of bone density in normal British women. Br J Ra- of vertebral fracture risk. Bone 1985; 6:1 7 Downloaded from by on 5/14/18 from IP address Copyright ARRS. For personal use only; all rights reserved photon cross section database. National Institute of Standards and Technology website. physics.nist.gov/ xcom. Published 21. Accessed February 17, Bland JM, Altman DG. Statistical methods for assessing agreement between two methods of clinical measurement. Lancet 1986; 327: Genant HK, Block JE, Steiger P, Glueer CC, diol 199; 63: Kalender WA. A phantom for standardization and quality control in spinal bone mineral measurements by QCT and DXA: design considerations and specifications. Med Phys 1992; 19: Rasband WS. ImageJ: image processing and analysis in Java. National Institutes of Health website. 15. Kappadath SC, Shaw CC. Dual-energy digital mammography: calibration and inverse-mapping techniques to estimate calcification thickness and glandular-tissue ratio. Med Phys 23; 3: Mazess RB, Barden H. Bone density of the spine and femur in adult white females. Calcif Tissue Int 1999; 65:91 99 AJR:24, June