Apparent Diffusion Coefficient for Prostate Cancer Imaging: Impact of b Values

Transcription

1 Genitourinary Imaging Original Research Peng et al. Impact of b Value on Apparent Diffusion Coefficient in Prostate Cancer Imaging Genitourinary Imaging Original Research Yahui Peng 1 Yulei Jiang 2 Tatjana Antic 3 Ila Sethi 2 Christine Schmid-Tannwald 2 Scott Eggener 4 Aytekin Oto 2 Peng Y, Jiang Y, Antic T, et al. Keywords: apparent diffusion coefficient, b value, Gleason score, monoexponential decay, prostate cancer DOI: /AJR Received March 18, 2013; accepted after revision July 9, This work was supported in part by the U.S. Army Medical Research and Materiel Command Prostate Cancer Research Program through an Idea Development Award (PC093485). 1 School of Electronic and Information Engineering, Beijing Jiaotong University, No. 3 Shangyuancun, Haidan District, Beijing, China, Address correspondence to Y. Peng (yhpeng@bjtu.edu.cn). 2 Department of Radiology, The University of Chicago, Chicago, IL. 3 Department of Pathology, The University of Chicago, Chicago, IL. 4 Department of Surgery, Section of Urology, The University of Chicago, Chicago, IL. WEB This is a web exclusive article. AJR 2014; 202:W247 W X/14/2023 W247 American Roentgen Ray Society Apparent Diffusion Coefficient for Prostate Cancer Imaging: Impact of b Values OBJECTIVE. The purpose of this article is to investigate the effect of b values on apparent diffusion coefficient (ADC) values estimated from 1.5-T diffusion-weighted MRI (DWI) of the prostate acquired with an endorectal coil in distinguishing prostate cancer from normal-tissue regions of interest (ROIs) and the correlation of ADC values with the tumor Gleason score. MATERIALS AND METHODS. Pretreatment DWI studies were analyzed retrospectively in 51 consecutive patients with prostate cancer with either two (b = 0 and 1000 s/mm 2 ; n = 26 patients) or five (b = 0, 50, 200, 1500, and 2000 s/mm 2 ; n = 25 patients) b values. In 45 normal peripheral-zone ROIs and 65 prostate cancer ROIs (14 in the central gland), ADC values were estimated by use of several combinations of two or five b values and a monoexponential model. We used the area under the receiver operating characteristic curve to characterize the effectiveness of ADC values in distinguishing prostate cancer from normal-tissue ROIs, and we calculated Spearman rank-order correlation between ADC values and the Gleason score. RESULTS. ADC values were often significantly different (p < 0.001) when estimated from different combinations of two or five b values. However, except when both b values were less than or equal to 200 mm 2 /s or greater than or equal to 1500 mm 2 /s, the AUC value for distinguishing prostate cancer from normal-tissue ROIs was similar ( ). The correlation coefficients between ADC values and the Gleason score were between 0.30 and CONCLUSION. The choice of b values can significantly affect ADC estimates. ADC values can produce a similar discriminant performance in distinguishing prostate cancer from normal-tissue ROIs and in correlation with the Gleason score, but an appropriate ADC cutoff value needs to be selected specifically for each b-value combination. D iffusion-weighted MRI (DWI) reveals unique biologic information on tissue related to water diffusion and is an important part of multiparametric MRI techniques for prostate cancer [1, 2]. Prostate cancer is typically associated with restricted water diffusion and with lower apparent diffusion coefficient (ADC) values calculated from DWI, compared with normal prostatic tissue [3 9]. ADC maps and DWI studies, in combination with T2-weighted and other imaging sequences, are used clinically for prostate cancer detection, and ADC values are found to correlate with the Gleason score [7 10]. Estimating the ADC value requires DWI studies acquired with different diffusion weightings, also known as b values [11]. The widely used, and the simplest, monoexponential model of water diffusion requires a minimum of two b values, and more complex models require larger minimum numbers of b values. Currently, there is no clear consensus on either the number or the optimal value of b values for prostate cancer imaging [12]. The literature indicates that there is a wide spectrum of b values in clinical use [3 10]. Our purpose in this study was to investigate the effect of b values on ADC values estimated from 1.5-T prostate MRI with an endorectal coil in distinguishing prostate cancer from normal prostatic tissue regions of interest (ROIs) and the correlation strength of ADC values with the tumor Gleason score. Materials and Methods Patients This retrospective study was HIPAA compliant and was approved by our institutional review board with a waiver of informed patient consent. We searched the radiology image archive at our institution and identified 60 consecutive patients who received a diagnosis of prostate cancer and underwent multiparametric endorectal MRI before prostatectomy between March 2008 and March W247

2 Peng et al. TABLE 1: Patient Characteristics (n = 51) Characteristic, Group Average Median SD Range Age (y) Two-b-value group (n = 26) a Five-b-value group (n = 25) b Prostate-specific antigen level (ng/ml) Two-b-value group (n = 26) a Five-b-value group (n = 24) b,c a b = 0 and 1000 mm 2 /s. b = 0, 50, 200, 1500, and 2000 mm 2 /s. Data were missing in one case. Patients were excluded if they had received radiation therapy before MRI (n = 1), they were scanned with 3-T MRI scanners (n = 5), DWI studies were missing (n = 1), or tumor-specific Gleason score were missing (n = 2). A total of 51 patients was analyzed in this study. MRI Data Acquisition All MRI studies, including T1-weighted, T2- weighted, DWI, and dynamic contrast-enhanced MRI, were acquired with an endorectal coil (Prostate ecoil, Medrad) and a phased-array surface coil in 1.5-T MRI scanners (Achieva, Philips Healthcare). Because we were interested in only ADC values in this study, we describe here only DWI acquisition. For each patient, axial freebreathing DWI studies were acquired with the single-shot echo-planar imaging technique before dynamic imaging with contrast. An effective sensitivity-encoding (parallel-imaging) factor of 2 was used. The patient images can be naturally separated into two groups in terms of the b values used: a twob-value group with b = 0 and 1000 s/mm 2 (n = 26 patients, March 2008 through May 2009) and a five-b-value group with b = 0, 50, 200, 1500, and 2000 s/mm 2 (n = 25 patients, May 2009 through March 2010). The b values used in these imaging studies were the results of clinical considerations and were not influenced in any way by the current study. Additional DWI parameters are summarized in Appendix 1. Histology-Radiology Correlation Prostatectomy specimens were processed routinely at our institution as follows. After fixation with 10% buffered formalin, specimens were cut serially into 4-mm-thick slices from apex to base in the transverse plane. These were then halved or quartered depending on size and processed into paraffin blocks. The blocks were then microtomed into 3- to 4-µm sections and stained with H and E. The reference standard of prostate cancer foci in MRI was established through a systematic consensus-seeking correlative review of histology and MRI by a genitourinary pathologist (with 8 years of experience in genitourinary pathology) and a radiologist (with 9 years of experience in prostate MRI). The pathologist identified all distinct tumor foci larger than 5 mm in diameter. Simultaneously, the radiologist outlined one ROI for each tumor focus on a single MRI slice in which visualization of the tumor best correlated with histology, after reviewing T2-weighted, ADC, DWI, and dynamic contrast-enhanced MRI. For tumor foci that were not visible on MRI slices, their locations were determined according to the relationship to other identifiable landmarks on MRI (e.g., urethra or ejaculatory ducts), when it was possible, by consensus of the radiologist and the pathologist. The outline of the tumor ROIs on MRI did not necessarily accurately reflect the tumor outline on histology, and tumor ROIs on MRI were sometimes smaller than the tumor seen on histology. The pathologist provided the Gleason score for each tumor ROI. For each patient, the radiologist manually drew ROIs of peripheral-zone (PZ) and/or central-gland tumor foci (when present), as well as a normal PZ focus (unless no histologically normal region was found [n = 6 patients]; benign prostatic hyperplasia and other abnormal benign conditions were not considered as a normal region). All ROIs were outlined on T2-weighted images, except for four ROIs in four cases on DWI studies, in which the tumors were seen better on DWI studies than on T2-weighted images in correlation with histopathology. We transferred the ROIs from T2-weighted images to DWI studies, when necessary, assuming that there was no patient motion. Subsequently, the study radiologist reviewed the transferred ROIs and, when necessary, adjusted the locations of the ROIs manually in DWI studies without modifying the size or shape of the ROI. TABLE 2: Region of Interest (ROI) Characteristics (n = 110) Cancer Characteristic Normal Gleason Score 6 Gleason Score 7 Gleason Score 8 Gleason Score 9 All Tumors No. (%) of ROIs Two-b-value group (n = 55) a 26 6 (20.7) 13 (44.8) 6 (20.7) 4 (13.8) 29 (100) Five-b-value group (n = 55) b 19 9 (25.0) 21 (58.3) 4 (11.1) 2 (5.6) 36 (100) Total (n = 110) (23.1) 34 (52.3) 10 (15.4) 6 (9.2) 65 (100) No. (%) of central-gland ROIs Two-b-value group (n = 7) a 0 1 (14.3) 4 (57.1) 1 (14.3) 1 (14.3) 7 (100) Five-b-value group (n = 7) b 0 1 (14.3) 4 (57.1) 0 (0) 2 (28.6) 7 (100) Total (n = 14) 0 2 (14.3) 8 (57.1) 1 (7.1) 3 (21.4) 14 (100) ROI size (mm 2 ), average ± SD Two-b-value group (n = 55) a 69 ± ± ± ± ± ± 71 Five-b-value group (n = 55) b 67 ± ± ± ± ± ± 93 Total (n = 110) 68 ± ± ± ± ± ± 85 a b = 0 and 1000 mm 2 /s. b = 0, 50, 200, 1500, and 2000 mm 2 /s. W248

3 Impact of b Value on Apparent Diffusion Coefficient in Prostate Cancer Imaging ADC Value Calculation ADC values were first calculated for each pixel and then were averaged over all pixels within each ROI. To estimate pixel-wise ADC values from five b values, we used a linear least-squares fit to the logarithmic form of a monoexponential model: lns b = ADC b + lns 0 (1), where S b is the DWI signal intensity with diffusion weighting of b, S 0 is the signal intensity with diffusion weighting of b = 0 mm 2 /s, and ADC is the slope of this linear equation. With only two b values, the ADC value was estimated by use of a simplified equation: ADC = (lns 1 lns 2 )/ (b 2 b 1 ) (2), Spearman ρ AUC Average ADC Value in an ROI ( 10 3 mm 2 /s) , 50, 200, 1500, 2000 b = 0, 50 0, 200 0, 1000 where S 1 and S 2 are signal intensities with diffusion weightings of b 1 and b 2, respectively. For images with five b values (b = 0, 50, 200, 1500, and 2000 s/mm 2 ), in addition to the pixelwise ADC values estimated from all five b values by use of Equation 1, it was also possible to estimate the ADC from a subset of only two b values (e.g., b = 0 and 50 mm 2 /s, b = 50 and 200 mm 2 /s, and so forth), and we calculated 10 additional pixel-wise ADC values from 10 combinations of two of the five b values by use of Equation 2. Statistical Analysis We evaluated the impact of b values on the calculated ADC values in the context of distinguishing prostate cancer from normal PZ ROIs, by use of the maximum-likelihood estimation of the proper binormal receiver operating characteristic (ROC) curve and the area under the ROC curve (AUC) [13 15]. AUC values were compared statistically by use of the Z-test [15] when two AUC values were estimated from two groups of patients, and by use of the conventional bivariate binormal model [16] when two AUC values were estimated from a single b Values Used to Estimate ADC Value (mm 2 /s) group of patients. The Student t test for paired data was used in the comparison of ADC values calculated from a single group of patients (between various combinations of two of the five b values), and the Student t test for independent samples was used in the comparison of ADC values calculated from two groups of patients (between the two- and fiveb-value calculations and between prostate cancer and normal-tissue ROIs). Furthermore, the sensitivity and specificity were calculated for each set of ADC values by use of an empirically selected ADC cutoff value. Two ADC cutoff values were used in this calculation of the sensitivity and specificity: one was selected specifically for each set of ADC values to produce approximately 90% sensitivity and 90% specificity, and the other was held fixed for all sets of ADC values. Finally, the Spearman rank-order correlation coefficient (ρ) was calculated for assessment of the correlation strength between ADC values and the Gleason score [17], with the 95% CI Significantly different ADC values in Normal ROIs between two and five b values Significantly different ADC values in Cancer ROIs between two and five b values Significantly different ADC values in Normal and Cancer ROIs 0, , , , , 50, , , 2000 Normal ROIs Cancer ROIs 1500, 2000 Fig. 1 Comparison of apparent diffusion coefficient (ADC) values calculated from various combinations of b values. Plots show Spearman correlation coefficients (ρ with 95% CIs) between ADC value and Gleason score (top), area under receiver operating characteristic curve (AUC) values (± SE) in differentiating prostate cancer from normal peripheral-zone tissue ROIs (middle), and average ADC values within region of interest (ROI) (bottom). Statistically significant differences in average ADC values after Bonferroni correction for multiple comparisons (p < 0.001) are indicated just above horizontal axis by asterisks (*) or plus signs (+). Note that ADC values estimated from b = 0 and 1000 mm 2 /s are based on different group of patients from that of all other b value combinations. Vertical lines denote range of data points, horizontal lines through boxes denote medians, and dots denote outliers. W249

4 Peng et al. estimated by use of the Fisher transformation [18]. Two-sided p values were calculated, and p < 0.05 was considered to indicate statistical significance; after Bonferroni correction for multiple comparisons, the critical value of p < was used. Inhouse computer software was used for all statistical analyses except for the ROC analysis [19]. Results Patient characteristics and characteristics of the cancer and normal-tissue ROIs were similar between the two- and five-b-value groups (Tables 1 and 2). A total of 110 ROIs was drawn for the 51 patients, among which 40.9% (45/110) were from normal PZ tissue (a normal PZ region was not present in six patients) and 59.1% (65/110) from prostate cancer. Among cancer ROIs, 21.5% (14/65) were located in the central gland and 78.5% (51/65) were in the PZ. The size of cancer ROIs had a range of mm 2, with a mean (± SD) of 111 ± 85 mm 2 and a median of 90 mm 2. Effect of b Values on the ADC Value The average ADC values within an ROI calculated from various combinations of two b values are shown in Figure 1 in comparison with the average ADC values within an ROI calculated from the five b values. For both normal PZ and cancer ROIs, there were significant differences in the calculated ADC values based on different combinations of b values. In particular, when calculated from two b values that were both less than or equal to 200 mm 2 /s, the calculated ADC values were substantially greater than ADC values calculated from other b value combinations. Although some two-b-value combinations yielded ADC values similar to that calculated from the five b values, consistent, sometimes small, and systematic differences in the calculated ADC values rendered the differences between the calculated ADC values statistically significant. For example, the difference in the average ADC value calculated in the same normaltissue ROIs from two b values of b = 0 and 2000 mm 2 /s versus from the five b values was 0.03 ± mm 2 /s (p = ). Fig. 2 Receiver operating characteristic (ROC) curves of apparent diffusion coefficient values calculated from various combinations of b values for differentiation between prostate cancer and normaltissue regions of interest. p values for comparison of area under ROC curve (AUC) values between six lowest ROC curves and ROC curve from five b values are shown in key; comparison between all other ROC curves and ROC curve from five b values yielded p > 0.2. Sensitivity (%) 100 Effect of b Values on Distinguishing Prostate Cancer From Normal Tissue Figure 1 also shows AUC values in distinguishing prostate cancer from normal PZ ROIs according to the average ADC value within an ROI, and Figure 2 shows the corresponding ROC curves. Except for b values less than or equal to 200 mm 2 /s and greater than or equal to 1500 mm 2 /s, the AUC value for distinguishing between prostate cancer and normal-tissue ROIs based on the ADC values was similar ( ) among all two-b-value combinations and the five b values (Fig. 2). This discriminant ability was diminished for ADC values calculated on the basis of only b values of less than or equal to 200 mm 2 /s or greater than or equal to 1500 mm 2 /s. Effect of b Values on the ADC Cutoff Value Table 3 lists sensitivity and specificity values for various combinations of b values in AUC and b Values (mm 2 /s) 0.93 ± 0.03; b = 0, ± 0.04; b = 50, ± 0.04; b = 200, ± 0.04; b = 0, 50, 200, 1500, ± 0.04; b = 0, ± 0.05; b = 50, ± 0.05; b = 0, 2000 (p = 0.008) ± 0.05; b = 200, 2000 (p = 0.01) 0.81 ± 0.06; b = 50, 200 (p = 0.06) 0.79 ± 0.05; b = 0, 200 (p = 0.04) 0.65 ± 0.08; b = 1500, 2000 (p < ) 0.58 ± 0.07; b = 0, 50 (p = ) Specificity (%) the task of distinguishing between prostate cancer and normal prostatic tissue ROIs, as well as the empirically selected ADC cutoff values. Reasonably comparable sensitivity and specificity values were obtained from various b value combinations except when both b values were less than or equal to 200 mm 2 /s or greater than or equal to 1500 mm 2 /s. However, the ADC cutoff value varied considerably within a range between 1.00 and 1.44 mm 2 /s, not including those for only b values of less than or equal to 200 mm 2 /s TABLE 3: Sensitivity and Specificity in Distinguishing Between Prostate Cancer and Normal Prostatic Tissue Regions of Interest at an Empirically Selected Apparent Diffusion Coefficient (ADC) Cutoff Value for Each Combination of b Values b Values (mm 2 /s) Sensitivity (%) Specificity (%) ADC Cutoff Value ( 10 3 mm 2 /s) 0, , , , 50, 200, 1500, , , , , , , , , Note Data (rows) are presented in the order of the corresponding receiver operating characteristic curves shown in Figure 2. W250

5 Impact of b Value on Apparent Diffusion Coefficient in Prostate Cancer Imaging TABLE 4: Sensitivity and Specificity in Distinguishing Between Prostate Cancer and Normal Prostatic Tissue Regions of Interest at a Fixed Apparent Diffusion Coefficient Cutoff Value of mm 2 /s b Values (s/mm 2 ) Sensitivity (%) Specificity (%) 0, , , , 50, 200, 1500, , , , , , , , , Note Data (rows) are presented in the order of the corresponding receiver operating characteristic curves shown in Figure 2. or greater than or equal to 1500 mm 2 /s. Table 4 lists sensitivity and specificity values in the same task, but with a fixed ADC cutoff value of mm 2 /s for all b-value combinations. This cutoff value is the same as that listed in Table 3 for b values of 0, 50, 200, 1500, and 2000 mm 2 /s. Large variations in the sensitivity and specificity values were apparent. If a different ADC cutoff value were used for this analysis (Table 4), then the sensitivity and specificity results would change, but the large variation in sensitivity and specificity values would remain apparent. Effect of b Values on Correlation Between ADC Values and Gleason Score The Spearman correlation coefficient between the ADC value and the Gleason score are also shown in Figure 1. The correlation coefficients ranged between 0.30 and Figure 3 shows box plots of the correlation between the average ADC values within an ROI and the Gleason score. Discussion Previous studies have shown that the choice of b values can significantly influence estimated ADC values based on the monoexponential diffusion model at 3 T [6 8]. In this study, we have reproduced these findings at the 1.5-T field strength. In addition, we found that systematic, and sometimes small, differences in ADC values estimated from various b value combinations often do not significantly affect the discriminant ability of the ADC value in differentiating prostate cancer from normaltissue ROIs and in the correlation of ADC values with tumor Gleason score. The AUC value of the average ADC value in differentiating prostate cancer from normal-tissue ROI is approximately 0.9, and the Spearman correlation coefficient between the average ADC value and tumor Gleason score is between 0.30 and 0.68, both of which are consistent with literature reports [8, 20, 21]. However, the selection of an appropriate ADC cutoff value for distinguishing between prostate cancer and normal tissue can be substantially affected by the b-value combinations (Table 3). In our analysis, in an effort to achieve sensitivity and specificity values around 90%, the ADC cutoff value varied between 1.00 and mm 2 /s, depending on the b-value combinations (excluding those combinations of b values that were 200 and 1500 mm 2 /s). Therefore, it was not possible to select a fixed optimal ADC cutoff value for the various b-value combinations, and an appropriate ADC cutoff value had to be selected specifically for each b-value combination. This is further illustrated in Table 4 by the large variations in sensitivity and specificity that resulted from a fixed ADC cutoff value for all b-value combinations. If another fixed ADC cutoff value were selected, the sensitivity and specificity values in Table 4 would change, but the large variation in sensitivity and specificity would remain. Our results indicate that estimating the ADC value with low b values that are less than or equal to 200 mm 2 /s produces substantially greater ADC values, accompanied by larger variations, compared with ADC values estimated from other b-value combinations. Furthermore, ADC values estimated from low b values are associated with diminished discriminant ability in distinguishing between prostate cancer and normal tissue. This is likely because low-b-value DWI signal intensity can be heavily contaminated by perfusion. Similarly, estimating the ADC value with high b values that are greater than or equal to 1500 mm 2 /s produces smaller ADC values compared with those estimated from other b-value combinations and leads to diminished discriminant ability in distinguishing between prostate cancer and normal tissue (Fig. 1). This is probably due to the effect of noise in MRI signals. Therefore, one should avoid estimating ADC values from only b values that are either extremely small ( 200 mm 2 /s) or extremely large ( 1500 mm 2 /s) when 1.5-T MRI scanners are used [8]. Our results suggest that ADC values estimated by use of the monoexponential model from two versus five b values can be similarly effective in distinguishing between prostate cancer and normal tissue, even when systematic differences in the estimated ADC values are statistically significant. Estimating ADC value from five b values is expected, statistically, to yield more reliable curvefit results than from two b values and, thus, more reliable ADC-value estimates. However, this was not observed in this study. In addition to limitations of the monoexponential model, other sources of variability (e.g., patient motion) can also affect the ADC-value estimate, potentially masking the curve-fit effect. Further analysis is needed to address this question. For the five-b-value images, it was possible to estimate ADC values by use of a biexponential or intravoxel incoherent motion model [22]. The ADC values estimated from the intravoxel incoherent motion model are likely to be similar to those estimated from b values of 200 and 1500 mm 2 /s or from b values of 200 and 2000 mm 2 /s, because the perfusion component diminishes at high b values. These ADC values were similar to other ADC values estimated by use of the monoexponential model. It has been reported that more complex models could improve curve fitting but not necessarily improve the effectiveness of the ADC value in discriminating prostate cancer from normal tissue [23, 24]. Because the purpose of this study was to investigate the impact of b values on ADC estimates, we used the monoexponential model W251

6 Peng et al. Average ADC Value in an ROI ( 10 3 mm 2 /s) Normal on all b-value combinations and did not also use the biexponential model, which could be used only for the five-b-value combination. Others have investigated the effect of b values on DWI quality [25]. For example, Metens et al. [25] found that, at 3 T, prostate cancer lesions are best depicted on images with b values of 1500 and 2000 mm 2 /s because of the high contrast of prostate cancer in these images. This approach is clinically relevant and likely to be important. We focused in this study on the effect of b values on the estimated ADC values and did not visually analyze the quality of DWI studies. Both of these analyses might be important for understanding how to choose the b values to optimize DWI of prostate cancer. Most other studies on the effect of b values on the ADC estimate were conducted on 3-T scanners [6 8], because 3-T prostate MRI has become more commonplace, whereas our study was conducted on 1.5-T scanners. However, we expect the findings of this study to be applicable to 3-T scanners as well as to 1.5- T scanners [2, 26]. Images acquired at higher field strength have a better signal-to-noise ratio, but there is no reason to expect the dependence of the ADC value estimation on b values to differ fundamentally at higher field strength. Furthermore, our findings in this study are consistent with those of previous studies of 3-T scanners in that the ADC values can be strongly dependent on b values. We expect that our additional findings that the discriminant capability of the ADC value between prostate cancer and normal tissue remains not strongly affected by b values, and that the appropriate ADC cutoff value must be selected specifically for each set of b values, to be also true for 3-T images. Further studies are needed to confirm that finding. ρ = 0.62; b = 0, 50, 200, 1500, 2000 (mm 2 /s) ρ = 0.63; b = 0, 50, 1500 (mm 2 /s) ρ = 0.42; b = 0, 1000 (mm 2 /s) Gleason Score 9 Fig. 3 Correlation between Gleason score and average apparent diffusion coefficient (ADC) values within cancer region of interest (ROI). Box plots show correlation estimated from five versus two b values. Box plots of average ADC values of normal ROIs are also shown for comparison only. Spearman correlation coefficients (ρ) are listed in key. Vertical lines denote range of data points, horizontal lines through boxes denote medians, and dots denote outliers. A moderate and negative correlation between ADC values and Gleason score has been reported fairly consistently in the literature, with Spearman correlation coefficients ranging from approximately 0.3 to 0.65 [7 9]. The variation in this correlation strength that was found in this study due to different b values (Fig. 1) is consistent with that range. Therefore, it is possible that the variation in the reported correlation strength between ADC values and Gleason score is due, in part, to different b values. If this is true, then agreeing on a set of clinically appropriate b values may help strengthen the apparent correlation between ADC values and Gleason score and may potentially enhance the ability of the ADC value to predict the aggressiveness of prostate cancer. Further studies are needed to better understand this relationship. Limitations of this study include the retrospective study design and the limited number of patients. Because all patients had biopsyproven prostate cancer and subsequently underwent prostatectomy, patient selection bias was likely to be present, and high-grade prostate cancer could be underrepresented in these patients. Not all histologically identifiable cancer foci were analyzed on MRI. Because of the limited number of ROIs, central-gland tumors were not analyzed separately from PZ tumors. Other benign abnormalities, such as benign prostatic hyperplasia, were not included in the analysis of this study. More complex diffusion models that also consider perfusion and multiple exponential diffusion components may produce different ADC estimates but were not investigated in this study. Finally, much of the two-b-value combination data were extracted from five-b-value image data rather than from independent image data. In conclusion, ADC values estimated from 1.5-T endorectal prostate DWI based on the monoexponential diffusion model can differ significantly when calculated from various combinations of two or five b values; however, despite these differences, ADC values estimated from two b values can produce a similar discriminant performance as do those estimated from five b values in distinguishing prostate cancer from normal-tissue ROIs and in their correlation strengths with the tumor Gleason score. An appropriate ADC cutoff value for distinguishing between prostate cancer and normal tissue needs to be selected specifically for each b-value combination. References 1. Hoeks CM, Barentsz JO, Hambrock T, et al. Prostate cancer: multiparametric MR imaging for detection, localization, and staging. Radiology 2011; 261: Qayyum A. Diffusion-weighted imaging in the abdomen and pelvis: concepts and applications. RadioGraphics 2009; 29: Issa B. In vivo measurement of the apparent diffusion coefficient in normal and malignant prostatic tissues using echo-planar imaging. J Magn Reson Imaging 2002; 16: Gibbs P, Pickles MD, Turnbull LW. Diffusion imaging of the prostate at 3.0 tesla. Invest Radiol 2006; 41: Kim JH, Kim JK, Park BW, Kim N, Cho KS. Apparent diffusion coefficient: prostate cancer versus noncancerous tissue according to anatomical region. J Magn Reson Imaging 2008; 28: Kim CK, Park BK, Kim B. High-b-value diffusion-weighted imaging at 3 T to detect prostate cancer: comparisons between b values of 1,000 and 2,000 s/mm 2. AJR 2010; 194:[web]W33 W37 7. Vargas HA, Akin O, Franiel T, et al. Diffusionweighted endorectal MR imaging at 3 T for prostate cancer: tumor detection and assessment of aggressiveness. Radiology 2011; 259: Thörmer G, Otto J, Reiss-Zimmermann M, et al. Diagnostic value of ADC in patients with prostate cancer: influence of the choice of b values. Eur Radiol 2012; 22: Peng Y, Jiang Y, Yang C, et al. Quantitative analysis of multi-parametric prostate MR images: differentiation between prostate cancer and normal tissue and correlation with Gleason score a computeraided diagnosis development study. Radiology 2013; 267: Hambrock T, Somford DM, Huisman HJ, et al. Relationship between apparent diffusion coefficients at 3.0-T MR imaging and Gleason grade in peripheral zone prostate cancer. Radiology 2011; 259: Hagmann P, Jonasson L, Maeder P, Thiran JP, W252

7 Impact of b Value on Apparent Diffusion Coefficient in Prostate Cancer Imaging Wedeen VJ, Meuli R. Understanding diffusion MR imaging techniques: from scalar diffusionweighted imaging to diffusion tensor imaging and beyond. RadioGraphics 2006; 26:S205 S Dickinson L, Ahmed HU, Allen C, et al. Magnetic resonance imaging for the detection, localisation, and characterisation of prostate cancer: recommendations from a European consensus meeting. Eur Urol 2011; 59: Metz CE. ROC methodology in radiologic imaging. Invest Radiol 1986; 21: Metz CE, Pan X. Proper binormal ROC curves: theory and maximum-likelihood estimation. J Math Psychol 1999; 43: Hanley JA, McNeil BJ. The meaning and use of the area under a receiver operating characteristic (ROC) curve. Radiology 1982; 143: Metz CE, Wang P-L, Kronman H. A new approach for testing the significance of differences between ROC curves measured from correlated data. In: Deconinck F, ed. Information processing in medical imaging. The Hague, The Netherlands: Martinus Nijhoff, 1984: Riffenburgh RH. Statistics in medicine, 2nd ed. Amsterdam, The Netherlands: Elsevier Academic Press, Bonett DG, Wright TA. Sample size requirements for estimating Pearson, Kendall and Spearman correlations. Psychometrika 2000; 65: Metz CE. ROC software. metz-roc.uchicago.edu. Accessed June 25, Turkbey B, Shah VP, Pang Y, et al. Is apparent diffusion coefficient associated with clinical risk scores for prostate cancers that are visible on 3-T MR images? Radiology 2011; 258: Tamada T, Sone T, Jo Y, et al. Apparent diffusion coefficient values in peripheral and transition zones of the prostate: comparison between normal and malignant prostatic tissues and correlation with histologic grade. J Magn Reson Imaging APPENDIX 1: Additional Details on Diffusion-Weighted MRI Parameters 2008; 28: Le Bihan D, Breton E, Lallemand D, Aubin ML, Vignaud J, Laval-Jeantet M. Separation of diffusion and perfusion in intravoxel incoherent motion MR imaging. Radiology 1988; 168: Döpfert J, Lemke A, Weidner A, Schad LR. Investigation of prostate cancer using diffusion-weighted intravoxel incoherent motion imaging. Magn Reson Imaging 2011; 29: Quentin M, Blondin D, Klasen J, et al. Comparison of different mathematical models of diffusion-weighted prostate MR imaging. Magn Reson Imaging 2012; 30: Metens T, Miranda D, Absil J, Matos C. What is the optimal b value in diffusion-weighted MR imaging to depict prostate cancer at 3T? Eur Radiol 2012; 22: Koh DM, Collins DJ. Diffusion-weighted MRI in the body: applications and challenges in oncology. AJR 2007; 188: Additional details on the number of regions of interest (ROIs) drawn for each patient are provided as follows. Most patients had exactly one prostate cancer ROI and one normal-tissue ROI drawn, but six patients in the two-b-value group had two prostate cancer ROIs drawn, and eight patients in the five-b-value group had two or more prostate cancer ROIs drawn; one patient each in the two- and five-bvalue groups had two normal-tissue ROIs drawn; one patient in the two-b-value group and seven patients in the five-b-value group had no normal-tissue ROI drawn; and three patients in the two-b-value group and one patient in the five-b-value group had no prostate cancer ROI drawn. Table 5 shows the DWI acquisition parameters. TABLE 5: Typical Diffusion-Weighted MRI Acquisition Parameters No. of Patients b Values (mm 2 /s) TR/TE FOV (mm) a Matrix In-plane Resolution (mm 2 ) Slice Thickness (mm) No. of Averages 26 0, / , 50, 200, 1500, / a One-dimensional length of a square FOV. W253