Gliomas account for almost 80% of primary malignant brain tumors, and they result in more years of life lost than any other tumor type ( 1 ). Because

Transcription

1 Note: This copy is for your personal, non-commercial use only. To order presentation-ready copies for distribution to your colleagues or clients, contact us at ORIGINAL RESEARCH n NEURORADIOLOGY Gliomas: Histogram Analysis of Apparent Diffusion Coeffi cient Maps with Standard- or High- b-value Diffusion-weighted MR Imaging Correlation with Tumor Grade1 Yusuhn Kang, MD Seung Hong Choi, MD Young-Jae Kim, PhD Kwang Gi Kim, PhD Chul-Ho Sohn, MD Ji-Hoon Kim, MD Tae Jin Yun, MD Kee-Hyun Chang, MD 1 From the Department of Radiology, Seoul National University College of Medicine, 28 Yongon-dong, Chongno-gu, Seoul , Korea (Y.K., S.H.C., C.H.S., J.H.K., T.J.Y., K.H.C.); and Biomedical Engineering Branch, Division of Convergence Technology, National Cancer Center, Gyeonggi-do, Korea (Y.J.K., K.G.K.). Received April 3, 2011; revision requested May 5; revision received June 13; accepted June 23; fi nal version accepted July 14. Supported by the National Research and Development Program for Cancer Control, Ministry of Health and Welfare, Republic of Korea (grant ). Address correspondence to S.H.C. ( verocay@snuh.org ). q RSNA, 2011 Purpose: Materials and Methods: Results: Conclusion: To explore the role of histogram analysis of apparent diffusion coefficient (ADC) maps based on entire tumor volume data in determining glioma grade and to evaluate the diagnostic performance of ADC maps at standard (1000 sec/mm 2 ) and high (3000 sec/mm 2 ) b values. This retrospective study was approved by the institutional review board, and informed consent was waived. Twentyseven patients with astrocytic tumors underwent diffusionweighted magnetic resonance imaging with b values of 1000 and 3000 sec/mm 2, and the corresponding ADC maps were calculated ( and, respectively). Regions of interest containing the lesion were drawn on every section of the ADC map containing the tumor and were summated to derive volume-based data of the entire tumor. Histogram parameters were correlated with tumor grade by using repeated measurements analysis of variance, the Tukey- Kramer test for post hoc comparisons, and an unpaired Student t test. Receiver operating characteristic (ROC) curves were constructed to determine the optimum threshold for each histogram parameter, and sensitivity and specificity were assessed. Minimum and both decreased with increasing tumor grade. The 50th and 75th percentiles of cumulative histograms showed significant differences between grades ( P =.015 and.001, respectively), while the fifth and 75th percentiles of cumulative histograms showed such differences ( P =.015 and.014, respectively). Minimum ADC and the fifth percentile for both ( P,.001 and P =.024, respectively) and ( P,.001 and P =.001, respectively) proved to be significant histogram parameters for differentiating highfrom low-grade gliomas. The diagnostic value of the parameters derived from and were compared, and a significant difference (0.202, P =.014) was found between the areas under the ROC curve of the fifth percentiles for and. Histogram analysis of ADC maps based on entire tumor volume can be a useful tool for grading gliomas. The fifth percentile of the cumulative ADC histogram obtained at a high b value was the most promising parameter for differentiating high- from low-grade gliomas. q RSNA, 2011 Supplemental material: /suppl/doi: /radiol /-/dc1 882 radiology.rsna.org n Radiology: Volume 1: Number 3 December 2011

2 Gliomas account for almost 80% of primary malignant brain tumors, and they result in more years of life lost than any other tumor type ( 1 ). Because the therapeutic approaches for gliomas differ considerably according to tumor grade, it is important to accurately assess glioma grade prior to treatment ( 2 ). Various assessments have been used to differentiate high-grade from low-grade gliomas. Whereas early studies were based on contrast enhancement of the tumor ( 3 5 ), more recent studies are based on the perfusion or diffusion characteristics of the lesions. The role of diffusion-weighted (DW) magnetic resonance (MR) imaging with quantitative apparent diffusion coefficients (ADCs) in the pretreatment evaluation of glioma grade has been investigated in various studies ( 2,6 11 ). A study by Murakami et al ( 8 ) showed that regions exhibiting the minimum ADC corresponded to the highest-grade glioma foci and that elevated ADC difference values (the difference between maximum and minimum ADCs) corresponded most strongly to overall highgrade, rather than low-grade, lesions. However, contradictory findings have been reported. Zonari et al ( 12 ) found no significant difference in the minimum ADCs of low- versus high-grade glial tumors, and Catalaa et al ( 13 ) found that ADCs could not be used to separate different tumor grades. Most previous studies ( 2,6,8,10, 11,14 16 ) were based on selected regions of interest (ROIs) placed on a Advances in Knowledge n Histogram analysis of apparent diffusion coefficient (ADC) maps at a b value of 3000 sec/mm 2 based on entire tumor volume can be a useful diagnostic tool for grading gliomas. n Minimum ADC and the fifth per- centile of cumulative ADC histograms at a b value of 3000 sec/ mm 2 accurately distinguish highfrom low-grade gliomas, whereas the 75th percentile of cumulative ADC histograms distinguished grade III from grade IV gliomas. rep resentative section of the tumor for analysis. It is widely known that a single glioma, especially if high grade, may demonstrate a wide spectrum of histologic features ranging from grade II to grade IV; in turn, the ADCs within a given glioma can vary widely between different regions of that tumor. Therefore, an ADC derived from regional ROIs will likely underestimate the heterogeneity of glioma cellularity ( 17 ). Moreover, the selection of a localized area within a tumor can be subjective and prone to sampling bias ( 9 ). This sampling bias may explain the discrepant results among previous articles ( 8,12,13 ) for the ability of DW MR imaging to be used to distinguish high- from low-grade gliomas. A more objective approach than placement of regional ROIs would be to analyze the ADC of the entire volume of the tumor, which should provide quantitative information about the tissue characteristics and heterogeneity of the whole tumor. By encompassing the whole tumor, potential sampling bias would be largely eliminated, and the tumor areas with different diffusion characteristics would be reflected in the histogram, meaning that all elements of the tumor that could contribute to group differences would be analyzed ( 9 ). The purposes of our study were to explore the role of histogram analysis of ADC maps based on entire tumor volume in determining the grade of gliomas and to evaluate the diagnostic performance of ADC maps at standard (1000 sec/mm 2 ) and high (3000 sec/mm 2 ) b values. Materials and Methods This retrospective study was approved by the institutional review board of Seoul National University Hospital, and Implication for Patient Care n Histogram analysis of ADC maps at a b value of 3000 sec/mm 2 based on entire tumor volumes may facilitate accurate noninvasive tumor grading, potentially improving the care of patients with glioma. informed consent was waived. This study was supported by a grant from the National Research and Development Program for Cancer Control, Ministry of Health and Welfare, Republic of Korea (grant ). Patient Selection Fifty-seven patients with astrocytic tumors who had undergone initial MR imaging at Seoul National University Hospital between October 2008 and May 2010 were selected from the radiology report database. Inclusion criteria were as follows: (a) a histopathologic diagnosis of astrocytic tumors according to the World Health Organization criteria, without oligodendroglial components, and (b) MR imaging performed with DW at standard and high b values. We excluded 30 patients owing to the following conditions: (a) MR imaging performed at 3 T ( n = ), (b) DW images obtained with inadequate MR imaging ( n = 3), and (c) prior treatment ( n = 1). The patients who had undergone MR imaging with a 3-T imager were excluded to eliminate any bias owing to the combination of data from MR systems with different field strengths. As a result, a total of 27 patients were included. Diagnoses included World Health Organization grade II astrocytomas ( n = 6, 22%), grade III astrocytomas ( n = 6, 22%), Published online before print /radiol Content codes: Radiology 2011; 1: Abbreviations: ADC = apparent diffusion coeffi cient = ADC at b value of 1000 sec/mm 2 = ADC at b value of 3000 sec/mm 2 DW = diffusion weighted ROC = receiver operating characteristic ROI = region of interest Author contributions: Guarantor of integrity of entire study, S.H.C.; study concepts/ study design or data acquisition or data analysis/interpretation, all authors; manuscript drafting or manuscript revision for important intellectual content, all authors; approval of fi nal version of submitted manuscript, all authors; literature research, Y.K., S.H.C.; clinical studies, all authors; statistical analysis, Y.K., S.H.C.; and manuscript editing, Y.K., S.H.C. Potential confl icts of interest are listed at the end of this article. Radiology: Volume 1: Number 3 December 2011 n radiology.rsna.org 883

3 and grade IV glioblastomas ( n = 15, 56%) ( Fig 1 ). All 15 cases of grade IV glioblastomas were primary types. For statistical analysis, grade III astrocytomas and grade IV glioblastomas were collectively classified as high-grade gliomas, and grade I and II astrocytomas were grouped as low-grade gliomas. Image Acquisition All MR images were obtained with a 1.5-T MR imager (Signa HDx or HDxt; GE Medical Systems, Milwaukee, Wis) with an eight-channel head coil. The imaging protocol included axial T2-weighted fast spin-echo (repetition time msec/ echo time msec, 5000/131; 25 sections; section thickness, 5 mm; intersection gap, 1 mm; field of view, mm; matrix, ; one acquired signal; echo train length, 16; voxel resolution, mm) and axial T1-weighted spin-echo (466/11; section thickness, 5 mm; intersection gap, 1 mm; field of view, mm; matrix, ; voxel resolution, mm) sequences. Echo-planar DW MR imaging (6000/ 63: b = 0 and 1000 sec/mm 2 ; 8000/84: b = 0 and 3000 sec/mm 2 ; 25 sections; bandwidth, 1953 Hz per voxel; section thickness, 5 mm; intersection gap, 1 mm; field of view, mm; matrix, ; two acquired signals; voxel resolution, mm) was performed in the axial plane before the injection of contrast material. DW MR images were acquired in three orthogonal directions and combined into a trace image. By using these data, ADC maps at standard ( ) and high ( ) b values were calculated on a voxel-by-voxel basis with the software incorporated into the MR imaging unit. Axial T1-weighted sequences were repeated after the intravenous administration of a single dose of 0.1 mmol per kilogram of body weight of gadopentetate dimeglumine (Magnevist; Bayer Schering Pharma, Berlin, Germany). A fat-suppression pulse was added to the axial T1-weighted spin-echo sequence after the administration of the contrast agent. Depending on the location of the primary tumor, coronal and/or sagittal Figure 1 Figure 1: Flow diagram of patient selection and inclusion and exclusion criteria. DWI = DW MR imaging. T1-weighted spin-echo sequences without fat suppression were performed with identical imaging parameters after contrast agent administration. Qualitative Image Analysis Two neuroradiologists (S.H.C. and J.H.K., with 8 and 10 years of brain MR imaging experience, respectively) independently reviewed the DW MR images along with the conventional MR images to evaluate the probability of highgrade glioma. The observers analyzed the conventional MR images for the presence of contrast material enhancement, border definition, mass effect, signal intensity heterogeneity, hemorrhage, necrosis, degree of edema, and involvement of the corpus callosum or crossing of the midline. On DW MR images, they looked at the signal intensity of the lesions ( 18,19 ). A five-point scale was used to grade the probability that a lesion was a high-grade glioma (1 = definitely low-grade, 2 = probably low-grade, 3 = possibly high-grade, 4 = probably high-grade, 5 = definitely high-grade). For cases in which the two radiologists findings were discrepant, a consensus grade was allocated. Quantitative Image Analysis The MR data for the and maps were digitally transferred from the picture archiving and communication system workstation to a personal computer and processed with ImageJ (available at gov/ij/ ) ( 20 ) and a software program developed in-house (Y.J.K. and K.G.K.) by using Visual C++ (Microsoft, Redmond, Wash). ROIs that contained the entire tumor were drawn in each section of the maps and copied to the maps ( Fig 2 ). Tumor boundaries were defined with reference to the high-signalintensity areas thought to represent tumor tissue on the T2-weighted images by one author (Y.K.) ( 21 ). The data acquired from each section were summated to derive voxel-by-voxel ADCs for the entire tumor by using the software developed in-house. ADC histograms were plotted with ADC on the x-axis, with a bin size of mm 2 /sec, and the y-axis expressed as a percentage of the total lesion volume by dividing the frequency in each bin by the total number of voxels analyzed. For further quantitative analysis, 884 radiology.rsna.org n Radiology: Volume 1: Number 3 December 2011

4 Figure 2 Figure 2: Images of a 72-year-old woman with grade IV glioma show how ROIs were drawn on (b) ADC (b = 0 and 1000 sec/mm 2 ) and (c) ADC3000 (b = 0 and sec/mm 2 ) maps with reference to the (a) axial fast spin-echo T2-weighted MR image (5000/131). = border of ROI. cumulative ADC histograms were obtained from the ADC histograms, in which the cumulative number of observations in all of the bins up to the specified bin was mapped on the y-axis as a percentage. The following parameters were derived from the ADC histograms: (a) mean; (b) standard deviation; (c) mode, which equals the peak height position; (d) minimum; (e) kurtosis, which is the degree of peakedness of a distribution; and (f) skewness, which is a measure of the degree of asymmetry of a distribution. For the cumulative ADC histograms, the fifth, 50th, and 75th percentiles were derived (the n th percentile is the point at which n % of the voxel values that form the histogram are found to the left) ( 9 ). Subgroup Analysis of Nonenhancing Gliomas Five cases of nonenhancing gliomas were selected from the patient group on the basis of a review of pre- and postcontrast T1-weighted images by one author (Y.K.). Three cases of World Health Organization grade II gliomas and two cases of World Health Organization grade III gliomas were included. The histogram parameters derived from the ADC histograms and the cumulative ADC histograms were applied to the nonenhancing gliomas, and the diagnostic accuracy was evaluated. Statistical Analysis All statistical analyses were performed with MedCalc software (version for Microsoft Windows 2000/XP/Vista/7; MedCalc Software, Mariakerke, Belgium) and GraphPad InStat (version 3.05, 32 bits for Win 95/NT; GraphPad Software, San Diego, Calif). Results with P values less than.05 were considered to be significant. The histogram parameters were correlated with the tumor grades by using repeated measurements analysis of variance with Tukey-Kramer post hoc comparisons for normally distributed data. The data for each histogram parameter were assessed for normality with the Kolmogorov-Smirnov test. To compare the histogram parameters of high- and low-grade gliomas, the unpaired Student t test was applied. Receiver operating characteristic (ROC) curves were constructed to determine the optimum threshold for each histogram parameter and visual scale to differentiate low- and high-grade gliomas. To assess the superiority of histogram parameters compared with those of, the areas under the ROC curves were compared by using the method of DeLong et al ( 22 ). The cutoff values obtained from the ROC analysis were applied to the nonenhancing gliomas, and the sensitivities and specificities were calculated for each parameter. The McNemar test was performed to compare the sensitivities and specificities. Results Both minimum and minimum decreased with increasing tumor grade. The minimum ADCs for grade II, III, and IV gliomas were ( [standard deviation]) mm 2 / sec, ( ) mm 2 /sec, and ( ) mm 2 /sec, respectively, at and ( ) mm 2 /sec, ( ) mm 2 / sec, and ( ) mm 2 / sec, respectively, at. A significant difference was found between the three grades with regard to the minimum ( P =.010) and minimum ( P =.003). The mean differed significantly between grades II and IV ( P,.05) and between grades III and IV ( P,.05) Radiology: Volume 1: Number 3 December 2011 n radiology.rsna.org 885

5 Figure 3 Figure 3: (a, c) ADC histograms and (b, d) cumulative histograms for grade II, III, and IV gliomas obtained at b values of (a, b) 0 and 1000 sec/mm 2 and (c, d) 0 and 3000 sec/mm 2. ADC histograms of high-grade gliomas showed a higher relative frequency at low ADCs compared with grade II gliomas, resulting in signifi cant divergence between low- and high-grade gliomas at the low end of the cumulative histograms. This suggests that high-grade gliomas contained more pixels with low ADCs, which indicates high cellularity. The left shift of the cumulative histograms of high-grade gliomas was more obvious at than at. Table 1 ADC Histogram Parameters in Terms of Low- and High-Grade Gliomas Parameter Low-Grade Glioma ( n = 6) High-Grade Glioma ( n = 21) P Value * Mean ( 3 10 mm2 /sec) Minimum ( 3 10 mm 2 /sec) ,.001 Mode ( 3 10 mm2 /sec) Kurtosis Skewness Mean ( 3 10 mm2 /sec) Minimum ( 3 10 mm 2 /sec) ,.001 Mode ( 3 10 mm2 /sec) Kurtosis Skewness Note. Unless otherwise specifi ed, data are means 6 standard deviations. * Difference between low- and high-grade gliomas was evaluated by using the unpaired Student t test. but not between grades II and III. Other histogram parameters, including mode, kurtosis, and skewness, did not show significant differences between grades (Table E1 [online]). The results of the cumulative histograms revealed that the 50th and 75th percentiles for ( P =.015 and.001, respectively) and fifth and 75th percentiles for ( P =.015 and.014, respectively) showed significant differences between grades. A significant difference was found between grades II and IV for the fifth percentile for ( P,.01), with a higher fifth percentile for grade II than grade IV. The difference between grades III and IV was significant for the 50th and 75th percentiles for and 75th percentile for ( P,.01), and the cumulative histograms showed divergence toward the higher end (Table E2 [online]; Fig 3 ). Minimum ( P,.001), minimum ( P,.001), fifth and 75th percentiles for ( P =.024 and.010, respectively), and fifth percentile for ( P =.001) proved to be significant histogram parameters for differentiating high-from low-grade gliomas ( Tables 1, 2 ). The ADC maps, histograms, and cumulative histograms of representative cases of low- and high-grade gliomas are shown in Figure 4. Table 3 summarizes the results of the ROC analyses of the qualitative grade and histogram parameters used to distinguish high- from low-grade gliomas. showed higher areas under the ROC curve than did the fifth percentile for and minimum ( P =.014, and.040, respectively). We noted no significant differences between the areas under the ROC curves of the fifth percentile for and minimum. Compared with the qualitative grade, only the fifth percentile for resulted in a significant difference in the area under the ROC curve ( P =.049). The fifth percentile for showed the highest sensitivity and specificity (85.7% and 100%, respectively) among the histogram parameters, but the difference with the sensitivities and specificities of other histogram parameters was not significant ( Table 3 ). Figure radiology.rsna.org n Radiology: Volume 1: Number 3 December 2011

6 shows that the ADC histograms of highgrade gliomas showed a higher relative frequency at the low ADCs compared with grade II gliomas, resulting in substantial divergence between low- and high-grade gliomas at the low end of the cumulative histograms. This suggests that the high-grade gliomas contained more pixels with low ADCs, which indicates high cellularity. The left shift of the cumulative histogram of high-grade gliomas was more obvious at than at, which correlates well with the significant difference in the fifth percentile for between the highand low-grade gliomas. Figure 4 Subgroup Analysis of Nonenhancing Gliomas With the cutoff value derived from the aforementioned ROC analysis, minimum could be used to differentiate nonenhancing grade III gliomas ( n = 2) from nonenhancing grade II gliomas ( n = 3) with a sensitivity of 0% and a specificity of 100%. The fifth percentile for showed a sensitivity of 50% and a specificity of 100%, which resulted in an insignificant difference compared with the values for minimum. Discussion The results of our study suggest that the low percentile values (ie, minimum or fifth percentile) of cumulative ADC histograms based on entire tumor volumes could be used to differentiate high- from low-grade gliomas, whereas the high percentile values (ie, 75th percentile) may be useful for the discrimination between grade III and grade IV gliomas. The histogram parameters derived from high b values showed better diagnostic performance than those from standard b values. However, the wholetumor summary statistics (ie, mean and median ADC) obtained from volume data appear to have limited value in distinguishing tumor grade. Previous studies ( 2,6,8,23 ) have shown that the minimum ADC measured within a tumor correlates well with areas of high cellularity and, thus, high grade. High cellularity is associated Table 2 Cumulative ADC Histogram Parameters in Terms of Low- and High-Grade Gliomas Parameter Low-Grade Glioma ( n = 6) High-Grade Glioma ( n = 15) P Value * (3 10 mm 2 /sec) Fifth percentile th percentile th percentile (3 10 mm 2 /sec) Fifth percentile th percentile th percentile Note. Unless otherwise specifi ed, data are means 6 standard deviations. * Difference between low- and high-grade gliomas was evaluated by using the unpaired Student t test. with relative reductions in extracellular space as compared with low cellularity, resulting in a decreased diffusivity of water molecules. In our study, minimum ADC and the fifth percentile at both standard and high b values were found to be discriminant histogram parameters, and these results are in concordance with the findings of previous reports. Our study has shown that the correlation of Radiology: Volume 1: Number 3 December 2011 n radiology.rsna.org 887

7 Figure 4 (continued) Figure 4: (a, e) ADC1000 maps and (b, f) ADC maps with corresponding (c, g) histograms and 3000 (d, h) cumulative histograms for representative (a d) low- grade glioma in a 67-year-old woman and (e h) high-grade glioma in a 48-year-old man. ADCs of the high-grade glioma were dispersed over a wider range than were those of the low-grade glioma, indicating the heterogeneous spectrum of cellularity within the entire tumor volume. ADCs decreased when b value was increased from 1000 to 3000 sec/mm 2. Table 3 ROC Results of Qualitative Analysis and ADCs for Tumor Grading Parameter Area under ROC Curve * Sensitivity (%) Specifi city (%) Cutoff Value P Value Qualitative grade (0.429, 0.810) 90.5 (19/21) 50.0 (3/6) Minimum (0.684, 0.967) 85.7 (18/21) 83.3 (5/6) 702,.001 Fifth percentile (0.505, 0.867) 57.1 (12/21) 100 (6/6) Minimum (0.745, 0.988) 76.2 (16/21) 100 (6/6) 440,.001 Fifth percentile (0.739, 0.986) 85.7 (18/21) 100 (6/6) 660,.001 * Data in parentheses are 95% confi dence intervals. Sensitivity and specifi city for identifying high-grade tumors. Data in parentheses are numbers used to calculate percentages. Except for qualitative grade, data are in units of 3 10 square millimeters per second. minimum ADC with high-grade glioma may apply to measurements based on entire tumor volume. In a study of malignant astrocytic tumors, Higano et al ( 24 ) reported a significant difference in the minimum ADC between grade III and grade IV astrocytic tumors ([ ] mm 2 /sec vs [ ] mm 2 / sec, respectively) at a b value of 1000 sec/mm 2. In our study, the minimum ADC did not differ significantly between grade III and grade IV gliomas. However, the higher-end values of cumulative ADC histograms were significantly different for the two grades, and the histograms showed divergence toward the higher end, which suggests that the difference between the two grades is mostly seen at higher ADCs. We speculate that the necrotic component frequently found in grade IV gliomas resulted in the high frequency of high-adc voxels in these tumors when compared with grade III gliomas, making discrimination between the two grades possible. Although macroscopically cystic or necrotic areas found within tumors were excluded from analysis in most studies, our study shows that the necrotic components may facilitate the differentiation between grade III and grade IV gliomas based on cumulative ADC histograms of the entire tumor volume. The simple summary statistics (eg, median and mean ADC) derived from the entire tumor volume did not show a significant correlation with tumor grade. Whereas grade III gliomas showed a lower mean ADC than did grade II gliomas, grade IV gliomas showed a higher mean ADC than either grade II or grade III gliomas, which contradicts previous reports. We speculate that the inclusion of necrotic portions, frequently found in grade IV gliomas, may have contributed to the high mean ADC. By deriving mean or median ADCs from selected volume elements only, the confounding effect of the high ADCs of necrosis may be reduced. However, microscopic areas of necrosis cannot be completely excluded from the analysis, and the partial volume-averaging effect of adjacent areas of necrosis cannot be completely eliminated. This may partly contribute 888 radiology.rsna.org n Radiology: Volume 1: Number 3 December 2011

8 Disclosures of Potential Conflicts of Interest: Y.K. No potential conflicts of interest to disclose. S.H.C. Financial activities related to the present article: institution has a grant from the National Research & Development Program for Cancer Control, Ministry of Health & Welfare, Republic of Korea (grant ). Financial activities not related to the present article: none to disclose. Other relationships: none to disclose. Y.J.K. No potential conflicts of interest to disclose. K.G.K. No potential conflicts of interest to disclose. C.H.S. No potential conflicts of interest to disclose. J.H.K. No potential conflicts of into the substantial overlap in the regional ADCs between high- and low-grade gliomas. Because the overlap of single ADC measurements between different grades has been pointed out as a limitation of many studies, we decided to focus more on the distribution of ADCs, rather than simple summary statistics. On this basis, we included the area of necrosis in the analysis to assess the heterogeneity of the tumor in its entirety. In our study, ADCs decreased when the b value was increased from 1000 to 3000 sec/mm 2, and a greater decrease was noted with higher tumor grades. It has been shown in previous reports ( ) that ADCs decrease when b values are increased beyond 1000 sec/ mm 2, which may be explained by a biexponential diffusion signal decay in the brain. The diffusion signal intensity is presumed to be dominated by a fast diffusion component at a relatively low b value, whereas at a high b value, signal intensity is contributed to largely by the slow diffusion component ( 25 ). Seo et al ( 29 ) demonstrated that the degree of ADC decrease was greater in tumors compared with normal brain tissue. Similarly, the degree of ADC decrease tended to increase with increasing tumor grade in our study, possibly suggesting that the slow diffusion component fraction may be larger in highgrade glioma. Apart from the intrinsic limits of any retrospective study, several other limitations of our study should be mentioned. First, the patient population was rather small and did not include patients with World Health Organization grade I astrocytoma. Further investigation that includes a larger population is warranted to strengthen the statistical power. Second, there was a methodologic challenge in defining the tumor boundary. The optimal definition of an entire tumor volume is complicated because gliomas are infiltrating tumors with indistinct borders beyond the radiologic margins ( 30,31 ). A conservative definition of the ROI includes only the solid tumor component with gadolinium enhancement on T1-weighted MR images. A more generous tumor extent would include contrast-enhanced and hyper- intense tissues on T2-weighted MR images ( 32 ). In our study, the tumor boundary was defined with reference to T2-weighted MR images. The highsignal-intensity area of T2-weighted MR images may represent either tumor infiltration, peritumoral edema, or even a combination of the two. An image-based differentiation between these two components is impossible; consequently, defining the exact tumor boundary is impractical. However, results from previous studies ( 21,33 ) show that variations among observers caused by imperfect tumor delineation are relatively unimportant given the large number of data points included in the histogram of an entire tumor volume. Third ( 34 ), despite the exclusion of any visible foci of suspected artifacts on the ADC map from the ROI measurements, the possibility of including extreme ADCs resulting from DW MR imaging and ADC map misregistration artifacts remains. The fifth percentile obtained from cumulative histograms can be presumed to be less affected by these artifacts and, therefore, seems to be a more reliable histogram parameter than minimum, although no significant difference in diagnostic performance was found in our study. In conclusion, our results suggest that histogram analysis based on the ADC of the entire tumor volume can be a useful objective diagnostic tool for grading gliomas. The fifth percentile of the cumulative ADC histogram obtained at a high b value is the most promising parameter for differentiating high- from low-grade gliomas. Acknowledgment: The authors thank So Young Yun, BS, for her continuous support with updating and organizing the clinical data. terest to disclose. T.J.Y. No potential conflicts of interest to disclose. K.H.C. No potential conflicts of interest to disclose. References 1. Schwartzbaum JA, Fisher JL, Aldape KD, Wrensch M. Epidemiology and molecular pathology of glioma. Nat Clin Pract Neurol 2006 ; 2 ( 9 ): ; quiz 1 p following Sugahara T, Korogi Y, Kochi M, et al. Usefulness of diffusion-weighted MRI with echoplanar technique in the evaluation of cellularity in gliomas. J Magn Reson Imaging 1999 ; 9 ( 1 ): Watanabe M, Tanaka R, Takeda N. Magnetic resonance imaging and histopathology of cerebral gliomas. Neuroradiology 1992 ; 34 ( 6 ): Brant-Zawadzki M. Pitfalls of contrastenhanced imaging in the nervous system. Magn Reson Med 1991 ; 22 ( 2 ): Brant-Zawadzki M, Berry I, Osaki L, Brasch R, Murovic J, Norman D. Gd-DTPA in clinical MR of the brain. I. Intraaxial lesions. AJR Am J Roentgenol 1986 ; 147 ( 6 ): Murakami R, Hirai T, Kitajima M, et al. Magnetic resonance imaging of pilocytic astrocytomas: usefulness of the minimum apparent diffusion coefficient (ADC) value for differentiation from high-grade gliomas. Acta Radiol 2008 ; 49 ( 4 ): Provenzale JM, Mukundan S, Barboriak DP. Diffusion-weighted and perfusion MR imaging for brain tumor characterization and assessment of treatment response. Radiology 2006 ; 239 ( 3 ): Murakami R, Hirai T, Sugahara T, et al. Grading astrocytic tumors by using apparent diffusion coefficient parameters: superiority of a one- versus two-parameter pilot method. Radiology 2009 ; 251 ( 3 ): Tozer DJ, Jäger HR, Danchaivijitr N, et al. Apparent diffusion coefficient histograms may predict low-grade glioma subtype. NMR Biomed 2007 ; 20 ( 1 ): Arvinda HR, Kesavadas C, Sarma PS, et al. Glioma grading: sensitivity, specificity, positive and negative predictive values of diffusion and perfusion imaging. J Neurooncol 2009 ; 94 ( 1 ): Lee EJ, Lee SK, Agid R, Bae JM, Keller A, Terbrugge K. Preoperative grading of presumptive low-grade astrocytomas on MR imaging: diagnostic value of minimum apparent diffusion coefficient. AJNR Am J Neuroradiol 2008 ; 29 ( 10 ): Radiology: Volume 1: Number 3 December 2011 n radiology.rsna.org 889

9 12. Zonari P, Baraldi P, Crisi G. Multimodal MRI in the characterization of glial neoplasms: the combined role of single-voxel MR spectroscopy, diffusion imaging and echoplanar perfusion imaging. Neuroradiology 2007 ; 49 ( 10 ): Catalaa I, Henry R, Dillon WP, et al. Perfusion, diffusion and spectroscopy values in newly diagnosed cerebral gliomas. NMR Biomed 2006 ; 19 ( 4 ): Kono K, Inoue Y, Nakayama K, et al. The role of diffusion-weighted imaging in patients with brain tumors. AJNR Am J Neuroradiol 2001 ; 22 ( 6 ): Yang D, Korogi Y, Sugahara T, et al. Cerebral gliomas: prospective comparison of multivoxel 2D chemical-shift imaging proton MR spectroscopy, echoplanar perfusion and diffusion-weighted MRI. Neuroradiology 2002 ; 44 ( 8 ): Lam WW, Poon WS, Metreweli C. Diffusion MR imaging in glioma: does it have any role in the pre-operation determination of grading of glioma? Clin Radiol 2002 ; 57 ( 3 ): Cha S. Update on brain tumor imaging: from anatomy to physiology. AJNR Am J Neuroradiol 2006 ; 27 ( 3 ): Kondziolka D, Lunsford LD, Martinez AJ. Unreliability of contemporary neurodiagnostic imaging in evaluating suspected adult supratentorial (low-grade) astrocytoma. J Neurosurg 1993 ; 79 ( 4 ): Knopp EA, Cha S, Johnson G, et al. Glial neoplasms: dynamic contrast-enhanced T2*- weighted MR imaging. Radiology 1999 ; 211 ( 3 ): Abramoff M, Magelhães P, Ram S. Image processing with ImageJ. Biophoton Int 2004 ; 11 ( 7 ): Emblem KE, Nedregaard B, Nome T, et al. Glioma grading by using histogram analysis of blood volume heterogeneity from MRderived cerebral blood volume maps. Radiology 2008 ; 247 ( 3 ): DeLong ER, DeLong DM, Clarke-Pearson DL. Comparing the areas under two or more correlated receiver operating characteristic curves: a nonparametric approach. Biometrics 1988 ; 44 ( 3 ): Kitis O, Altay H, Calli C, Yunten N, Akalin T, Yurtseven T. Minimum apparent diffusion coefficients in the evaluation of brain tumors. Eur J Radiol 2005 ; 55 ( 3 ): Higano S, Yun X, Kumabe T, et al. Malignant astrocytic tumors: clinical importance of apparent diffusion coefficient in prediction of grade and prognosis. Radiology 2006 ; 241 ( 3 ): DeLano MC, Cooper TG, Siebert JE, Potchen MJ, Kuppusamy K. High-b-value diffusionweighted MR imaging of adult brain: image contrast and apparent diffusion coefficient map features. AJNR Am J Neuroradiol 2000 ; 21 ( 10 ): Niendorf T, Dijkhuizen RM, Norris DG, van Lookeren Campagne M, Nicolay K. Biexponential diffusion attenuation in various states of brain tissue: implications for diffusion-weighted imaging. Magn Reson Med 1996 ; 36 ( 6 ): Mulkern RV, Gudbjartsson H, Westin CF, et al. Multi-component apparent diffusion coefficients in human brain. NMR Biomed 1999 ; 12 ( 1 ): Maier SE, Bogner P, Bajzik G, et al. Normal brain and brain tumor: multicomponent apparent diffusion coefficient line scan imaging. Radiology 2001 ; 219 ( 3 ): Seo HS, Chang KH, Na DG, Kwon BJ, Lee DH. High b-value diffusion (b = 3000 s/mm2) MR imaging in cerebral gliomas at 3T: visual and quantitative comparisons with b = 1000 s/mm2. AJNR Am J Neuroradiol 2008 ; 29 ( 3 ): Price SJ, Jena R, Burnet NG, et al. Improved delineation of glioma margins and regions of infiltration with the use of diffusion tensor imaging: an image-guided biopsy study. AJNR Am J Neuroradiol 2006 ; 27 ( 9 ): Grier JT, Batchelor T. Low-grade gliomas in adults. Oncologist 2006 ; 11 ( 6 ): Padhani AR, Liu G, Koh DM, et al. Diffusion-weighted magnetic resonance imaging as a cancer biomarker: consensus and recommendations. Neoplasia 2009 ; 11 ( 2 ): Kim HS, Kim JH, Kim SH, Cho KG, Kim SY. Posttreatment high-grade glioma: usefulness of peak height position with semiquantitative MR perfusion histogram analysis in an entire contrast-enhanced lesion for predicting volume fraction of recurrence. Radiology 2010 ; 256 ( 3 ): Law M, Young R, Babb J, Pollack E, Johnson G. Histogram analysis versus region of interest analysis of dynamic susceptibility contrast perfusion MR imaging data in the grading of cerebral gliomas. AJNR Am J Neuroradiol 2007 ; 28 ( 4 ): radiology.rsna.org n Radiology: Volume 1: Number 3 December 2011