Incremental Prognostic Value of ADC Histogram Analysis over MGMT Promoter Methylation Status in Patients with Glioblastoma 1

Transcription

1 This copy is for personal use only. To order printed copies, contact Incremental Prognostic Value of ADC Histogram Analysis over MGMT Promoter Methylation Status in Patients with Glioblastoma 1 Original Research n Neuroradiology Yoon Seong Choi, MD, PhD Sung Soo Ahn, MD, PhD Dong Wook Kim, PhD Jong Hee Chang, MD, PhD Seok-Gu Kang, MD, PhD Eui Hyun Kim, MD Se Hoon Kim, MD, PhD Tyler Hyungtaek Rim, MD, MBA Seung-Koo Lee, MD, PhD 1 From the Department of Radiology and Research Institute of Radiological Science (Y.S.C., S.S.A., S.-K.L.), Department of Neurosurgery (J.H.C., S.-G.K., E.H.K.), Department of Pathology (S.H.K.), and Department of Ophthalmology (T.H.R.), Yonsei University College of Medicine, 50 Yonsei-ro, Seodaemun-gu, Seoul 03722, Korea; and Department of Policy Research Affairs, National Health Insurance Service Ilsan Hospital, Goyang, Korea (D.W.K.). Received August 30, 2015; revision requested October 13; revision received December 13; accepted January 5, 2016; final version accepted January 20. Address correspondence to S.-K.L. ( SLEE@yuhs.ac). Purpose: Materials and Methods: Results: Conclusion: To investigate the incremental prognostic value of apparent diffusion coefficient (ADC) histogram analysis over oxygen 6 methylguanine-dna methyltransferase (MGMT) promoter methylation status in patients with glioblastoma and the correlation between ADC parameters and MGMT status. This retrospective study was approved by institutional review board, and informed consent was waived. A total of 112 patients with glioblastoma were divided into training (74 patients) and test (38 patients) sets. Overall survival (OS) and progression-free survival (PFS) was analyzed with ADC parameters, MGMT status, and other clinical factors. Multivariate Cox regression models with and without ADC parameters were constructed. Model performance was assessed with c index and receiver operating characteristic curve analyses for 12- and 16-month OS and 12-month PFS in the training set and validated in the test set. ADC parameters were compared according to MGMT status for the entire cohort. By using ADC parameters, the c indices and diagnostic accuracies for 12- and 16-month OS and 12-month PFS in the models showed significant improvement, with the exception of c indices in the models for PFS (P,.05 for all) in the training set. In the test set, the diagnostic accuracy was improved by using ADC parameters and was significant, with the 25th and 50th percentiles of ADC for 16-month OS (P =.040 and P =.047) and the 25th percentile of ADC for 12-month PFS (P =.026). No significant correlation was found between ADC parameters and MGMT status. ADC histogram analysis had incremental prognostic value over MGMT promoter methylation status in patients with glioblastoma. q RSNA, 2016 Online supplemental material is available for this article. q RSNA, 2016 Radiology: Volume 281: Number 1 October 2016 n radiology.rsna.org 175

2 Glioblastomas are the most common malignant brain tumors (1). They carry a poor prognosis, with a median survival time of months despite surgery, followed by concurrent chemoradiation therapy with temozolomide (2,3). However, a subset of patients exhibit exceptionally long survival and good response to chemotherapy, which has led to a search for prognostic factors, including genetic biomarkers such as oxygen 6 methylguanine-dna methyltransferase (MGMT) promoter methylation and advanced magnetic resonance (MR) imaging techniques such as diffusion tensor (DT) imaging (4,5). Methylation of the MGMT promoter has been reported to be associated with a better response to temozolomide therapy and, subsequently, a better prognosis than unmethylated promoter status (6,7). The reference standard for MGMT status is surgical sampling; thus, advanced MR imaging techniques such Advances in Knowledge nn Preoperative apparent diffusion coefficient (ADC) histogram analysis can have incremental prognostic value over postoperatively obtained oxygen 6 methylguanine-dna methyltransferase (MGMT) promoter methylation status in patients with glioblastomas who undergo temozolomide therapy, showing better prediction of 12- and 16-month overall survival (OS) (area under the receiver operating characteristic curve [AUC], ) and 12-month progression-free survival (PFS) (AUC = ) with the use of prognostic models with ADC parameters than without ADC parameters (for 12- and 16-month OS, AUC = , and for 12-month PFS, AUC = 0.751). nn No significant association was found between ADC and MGMT promoter methylation status (P =.694 for mean, and P =.185, P =.256, and P =.875 and 25th, 50th, and 75th percentiles of ADC, respectively). as DT imaging have been used to establish noninvasive imaging biomarkers. DT imaging provides information about the motion of water protons at the cellular level. The apparent diffusion coefficient (ADC) is an index of tumor cellularity that reflects tumor burden and inversely correlates with glioma grade (8 12). The prognostic value of the ADC in patients with glioblastomas who undergo temozolomide therapy and the relationship between ADC and MGMT status were analyzed, although the results are controversial (2,11,13 18). Given that most patients with glioblastomas undergo surgery as part of the standard treatment and that their MGMT status may be known, it is of clinical relevance to know whether ADC values have incremental prognostic value over MGMT status; however, such a finding has remained unmined. The purpose of this study was to investigate the incremental prognostic value of ADC histogram analysis over MGMT promoter methylation status in patients with glioblastomas and to investigate the correlation between ADC histogram parameters and MGMT status. Materials and Methods This retrospective study was approved by our institutional review board. The requirement for informed consent was waived. Participants The records of 139 patients with newly diagnosed glioblastomas who underwent preoperative DT imaging from December 2009 to April 2014 were reviewed. All patients were treated with Implications for Patient Care nn Preoperative ADC histogram analysis can have incremental prognostic value over MGMT promoter methylation status in patients with glioblastomas. nn Preoperative ADC histogram analysis has the potential to improve characterization of glioblastomas and treatment planning and is useful both postoperatively and preoperatively. a standard regimen of surgery and concurrent chemoradiation therapy with temozolomide (19). Patients were included on the basis of the following criteria: (a) a pathologically confirmed glioblastoma; (b) preoperative MR imaging, including DT imaging, with a b value of 600 sec/mm 2 ; (c) no previous biopsy or treatment; (d) MGMT promoter methylation status available in the pathologic report; and (e) follow-up for longer than 12 months in cases of living patients to assess the diagnostic accuracy for 12-month overall survival (OS) and progression-free survival (PFS), as described in a later section of this article. The following exclusion criteria were used: (a) a history of brain biopsy before MR imaging (n = 9), (b) preoperative DT imaging with a b value of 1000 sec/mm 2 (n = 2), (c) no surgery performed (n = 2), (d) follow-up shorter than 12 month in living patients (n = 12), and (e) inadequate image quality (n = 2) due to an intractable artifact on the ADC map. A total of 112 patients were enrolled and semirandomly divided into a training Published online before print /radiol Content code: Radiology 2016; 281: Abbreviations: ADC = apparent diffusion coefficient AUC = area under the receiver operating characteristics curve DT = diffusion tensor KPS = Karnofsky performance status MGMT = oxygen 6-methylguanine-DNA methyltransferase OS = overall survival PFS = progression-free survival ROI = region of interest Author contributions: Guarantors of integrity of entire study, Y.S.C., D.W.K., S.-G.K.; 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 final version of submitted manuscript, all authors; agrees to ensure any questions related to the work are appropriately resolved, all authors; literature research, Y.S.C., E.H.K.; clinical studies, Y.S.C., S.S.A., J.H.C., S.-G.K., E.H.K., S.H.K., S.-K.L.; experimental studies, Y.S.C., E.H.K., S.H.K.; statistical analysis, Y.S.C., D.W.K., E.H.K., T.H.R.; and manuscript editing, Y.S.C., S.S.A., J.H.C., E.H.K., T.H.R., S.-K.L. Conflicts of interest are listed at the end of this article. 176 radiology.rsna.org n Radiology: Volume 281: Number 1 October 2016

3 Figure 1 Figure 1: Use of a semiautomatic method with thresholds to segment entire enhancing tumors as ROIs in a 60-year-old man with glioblastomas. A, Contrast-enhanced T1-weighted MR image, B, contrastenhanced T1-weighted MR image with ROIs to segment the entire enhancing tumor, C, ADC map with the ROI of the entire enhancing tumor overlaid, and, D, ADC histogram extracted from the ROI show the use of a semiautomatic method with thresholds to segment entire enhancing tumors as ROIs. set of 74 patients and a test set of 38 patents, with stratification for age and MGMT status (20). Imaging Preoperative MR imaging was performed with a 3.0-T MR imaging machine (Achieva; Philips, Best, the Netherlands) and an eight-channel sensitivity-encoding head coil. The preoperative MR imaging protocol included T1-weighted (repetition time msec/echo time msec, 2000/10; field of view, 240 mm; section thickness, 5 mm; matrix, ), T2-weighted (3000/80; field of view, 240 mm; section thickness, 5 mm; matrix, ), and fluid-attenuated inversion recovery (10 000/125; field of view, 240 mm; section thickness, 5 mm; matrix, ) images. Three-dimensional contrast material enhanced T1-weighted images (6.3/3.1; field of view, 240 mm; section thickness, 1 mm; matrix, ) were acquired after administering 0.1 ml/kg gadolinium-based contrast material (Gadovist; Bayer, Toronto, Ontario, Canada). DT imaging was performed with b values of 600 sec/mm 2 and 0 sec/mm 2, 32 directions, and the following parameters: /77; field of view, 220 mm; section thickness, 2 mm; and matrix, Image Postprocessing and Analysis To contain each entirely enhancing tumor and avoid cystic or necrotic portions, regions of interest (ROIs) were drawn on every tumor section on contrast-enhanced T1-weighted images by using a semiautomatic method with signal intensity thresholds that are described elsewhere (21). ROIs were drawn by a single neuroradiologist (Y.S.C.) and coregistered to ADC maps by using affine transformation with normalized mutual information as a cost function (22,23), and ADC histograms were generated with a bin of mm 2 /sec and a range of mm 2 /sec. We considered ADC values less than mm 2 / sec to be artifacts and values greater than mm 2 /sec to be cystic portions. ADC histogram parameters that consisted of the mean and the 25th, 50th, and 75th percentile values were calculated from ROIs overlaid on ADC maps. A representative case is presented in Figure 1. Image analysis Radiology: Volume 281: Number 1 October 2016 n radiology.rsna.org 177

4 was performed with the Medical Image Processing, Analysis, and Visualization software package version 7.0 (National Institutes of Health; mipav.cit.nih.gov). MGMT and Other Prognostic Parameters MGMT promoter methylation status was assessed on the basis of methylation-specific polymerase chain reaction and retrospectively recorded from pathologic reports. Age; sex; preoperative Karnofsky performance status (KPS); extent of tumor resection; and conventional MR imaging findings, including the volume of each enhancing tumor, degree of edema, presence of a nonenhancing tumor, eloquent brain involvement, and deep white matter invasion, were also recorded. The extent of tumor resection was classified as total, subtotal (,100% and 75% of gross total removal), or partial (,75% of gross tumor removal) resection or biopsy only on the basis of the surgeon s (J.H.C., S.G.K., and E.H.K., with 20, 16, and 10 years of experience in neurosurgery, respectively) intraoperative impressions in conjunction with postoperative MR imaging findings. Conventional MR imaging findings were assessed by the consensus of two neuroradiologists (Y.S.C. and S.K.L, with 3 and 20 years of experience in brain MR imaging, respectively). Enhancing tumor volume was automatically calculated from ROIs as part of DT imaging analysis. Edema was scored according to the maximum distance of edema from the tumor margin on the basis of high signal intensity on axial T2-weighted or fluid-attenuated inversion recovery images not associated with mass effect or architectural distortion, with a score of 0 indicating not apparent ( 1 cm), a score of 1 indicating mild to moderate (.1 cm and 2 cm), and a score of 2 indicating severe (. 2 cm) edema (24,25). A nonenhancing tumor was defined as having an intermediate signal intensity on T2-weighted images that is lower than that of cerebrospinal fluid, associated with mass effect and architectural distortion, and classified as positive in patients with more than 25% enhancing tumor volume. The presence of eloquent brain involvement and deep white matter invasion was determined according to the Visually Accessible Repository for Molecular Brain Neoplasia Data Images lexicon, which is described elsewhere (26,27). Definition of Survival Time OS was defined as the time from MR imaging-based diagnosis to death or the date of the last follow-up examination. PFS was defined as the time from MR imaging-based diagnosis to tumor progression, recurrence, death, or the date of the last follow-up examination if the patient showed no disease progression or did not expire. The date of tumor progression was defined on the basis of the following response assessment in neuro-oncology criteria: (a) the date of the first follow-up examination in which an increase of 25% or more of enhancing or nonenhancing tumor within the radiation field was seen to increase in size between consecutive follow-up MR imaging examinations or was pathologically proved to be tumor recurrence, (b) the date of the initial follow-up examination in which a newly appearing enhancing lesion was seen outside the radiation field, (c) the date at which clinical deterioration secondary to disease was observed, or (d) the date of death in patients who did not meet the above criteria and expired. Image analysis that was used to determine the date of tumor progression was performed on the basis of the consensus of two neuroradiologists (Y.S.C. and S.K.L.). Statistical Analysis Interrater reliability for ADC parameters was determined by calculating the intraclass correlation coefficient with the two-way random model from data from 25 randomly selected patients in whom ROIs were drawn by two neuroradiologists (Y.S.C. and S.S.A). This intraclass correlation coefficient was interpreted as follows: less than 0 indicated no reproducibility, indicated slight reproducibility, indicated fair reproducibility, indicated moderate reproducibility, indicated substantial reproducibility, and indicated almost perfect reproducibility (28). Interrater reliability showed almost perfect reproducibility for ADC parameters, with intraclass correlation coefficients of 0.97 (95% confidence interval [CI]: 0.93, 0.99), 0.96 (95% CI: 0.91, 0.98), 0.97 (95% CI: 0.93, 0.99), and 0.98 (95% CI: 0.95, 0.99) for the mean and the 25th, 50th, and 75th percentiles of ADC, respectively. By using the training set, univariate analysis was performed with the Kaplan- Meier method and the log-rank test. Analysis included the mean; 25th, 50th, and 75th percentiles of ADC; MGMT promoter methylation status; and clinical factors, such as age, sex, KPS, extent of tumor resection, and conventional MR imaging findings. Continuous variables, such as ADC histogram parameters, and categoric variables of more than three categories were dichotomized for OS and PFS before the log-rank test by using optimal cutoff values that were based on the score test from the Cox regression model and calculated from the cutp function of the survmisc R package (R Foundation for Statistical Computing, Vienna, Austria) (29). Factors with P,.100 at univariate analysis were used in multivariate Cox regression analysis with the training set to build prognostic models for OS and PFS, in which ADC parameters, age, KPS, and enhancing tumor volume were entered as continuous variables, and the extent of tumor resection was entered after dichotomization into total resection versus otherwise. Prognostic models were built in two ways, each for OS and PFS: one without ADC parameters, and the other with ADC parameters as covariates. The variables in the final models, with the exception of age and KPS, were determined by backward stepwise variable selection on the basis of P =.100 indicating a significant difference. Because of their clinical importance, age and KPS were included in the final models regardless of their P values at multivariate analyses. Model performance was assessed by calculating Harrell concordance index (c index). C indices were compared between models with and without ADC parameters by using the method by Newson (20) to assess the incremental 178 radiology.rsna.org n Radiology: Volume 281: Number 1 October 2016

5 prognostic value of the ADC parameters. Additionally, diagnostic accuracy for 12- and 16-month OS and 12-month PFS was determined with receiver operating characteristic curve analysis, and the incremental values of ADC parameters were assessed with the integrated discrimination improvement index, which was described by Pencina et al (30). After building and evaluating the prognostic models on the training set as was previously described, diagnostic accuracy and incremental values of ADC parameters for 12- and 16-month OS and 12-month PFS were validated by using the test set. By using the entire cohort of 112 patients, the correlations between ADC histogram parameters and MGMT status were assessed with Student t test according to the results of the Shapiro-Wilk test for normality. Statistical analyses were performed with Stata software package version 12.1 (StataCorp, College Station, Tex) and R for Windows version P,.05 was considered to indicate a significant difference. Results Patients clinical characteristics are summarized in Table 1. No significant differences were found in clinical characteristics between the training and test sets. Among the 74 patients in the training set, the median OS was 564 days, and the median PFS was 352 days. Among the 38 patients in the test set, the median OS was 467 days, and the median PFS was 316 days. OS and PFS did not significantly differ between the training and test sets (P =.984 and P =.983 for OS and PFS, respectively). Univariate and Multivariate Survival Analysis At univariate analysis, lower ADC histogram parameters were significant predictors of poor OS and PFS (P,.05 for all) (Table E1 [online]). Unmethylated MGMT was also a significant predictor of poor OS (P =.013) and PFS (P =.005). Kaplan-Meier curves for OS according to ADC mean, Table 1 Clinical Characteristics of Patients Clinical Characteristic Training Set Test Set P Value Median OS (d) * No. deaths 48 (64.9) 24 (63.2) Median PFS (d) * No. tumor progression 57 (77.0) 30 (78.9) Age (y) 57.9 (12.2) 57.8 (12.0).966 Sex.428 Male 37 (50.0) 22 (57.9) Female 37 (50.0) 16 (42.1) KPS 72.0 (16.0) 75.8 (13.3).215 Extent of resection.337 Total 35 (47.3) 15 (39.5) Subtotal 26 (35.1) 14 (36.8) Partial 12 (16.2) 6 (15.8) Biopsy 1 (1.4) 3 (7.9) Conventional MR imaging findings Enhancing tumor volume (cm 3 ) 26.9 (21.2) 28.9 (22.7).641 Edema.098 None 22 (29.7) 6 (15.8) Mild to moderate 15 (20.3) 14 (36.8) Severe 37 (50.0) 18 (47.4) Nonenhancing tumor.601 Negative 39 (52.7) 22 (57.9) Positive 35 (47.3) 16 (42.1) Eloquent brain involvement.513 Yes 28 (37.8) 12 (31.6) No 46 (62.2) 26 (68.4) Deep white matter invasion.291 Yes 39 (52.7) 24 (63.2) No 35 (47.3) 14 (36.8) MGMT.429 Unmethylated 45 (60.8) 26 (68.4) Methylated 29 (39.2) 12 (31.6) Note. Unless otherwise indicated, data are numbers of patients, and data in parentheses are percentages. For all patients, n = 112; for the training set, n = 74; and for the test set, n = 38. Unless otherwise indicated, P values were calculated from Student t test for continuous variables and x 2 test for categoric variables. * P values were calculated from the log-rank test. MGMT status, and both ADC mean and MGMT status are shown in Figure 2. For clinical factors and tumor descriptors, an age of 69 years or more (P =.026); a KPS of less than 80 (P =.025); extent of resection, including subtotal, partial, and biopsy only (P =.028); and enhancing tumor volume of 39.9 cm 2 or more (P =.020) were significant predictors of poor OS. Similarly, an age of 69 years or more (P =.023) was a significant predictor of poor PFS, and KPS of less than 80 (P =.055); extent of resection, including subtotal, partial, and biopsy only (P =.069); and enhancing tumor volume of 30.8 cm 2 or more (P =.070) tended to indicate poor PFS. The presence of eloquent brain involvement (P =.098) and deep white matter invasion (P =.089) showed trends toward poor OS but not PFS. No significant differences in OS and PFS were found according to sex, degree of edema, or the presence of nonenhancing tumor. With the use of multivariate Cox regression, models without ADC histogram parameters for predicting OS Radiology: Volume 281: Number 1 October 2016 n radiology.rsna.org 179

6 Figure 2 Figure 2: Kaplan-Meier curves for OS show, A, ADC mean, B, MGMT status, and, C, both (n = 74). ADC mean and MGMT status were significant prognostic factors. Combining ADC mean and MGMT status further stratified prognosis in patients with glioblastomas. Table 2 Multivariate Prognostic Models with ADC Histogram Parameters for Predicting OS Model 1 Model 2 Model 3 Model 4 Parameter HR 95% CI P Value HR 95% CI P Value HR 95% CI P value HR 95% CI P Value ADC (per mm 2 /sec) Mean , 1.75, th percentile , 1.77, th percentile , 1.67, th percentile , 1.56,.001 MGMT , , 7.51, , 7.29, , Age (per 10 years) , , , , KPS (per 210) , , , , Extent of resection * , , , , Enhancing tumor volume (per 10 cm 3 ) , , , , Note. HR = hazard ratio. * Subtotal, partial resection, and biopsy only versus total resection. and PFS were also built; four models were each built for OS and PFS, with the mean and 25th, 50th, and 75th percentiles of ADC as the respective covariates (Tables 2, 3). All variables in models without ADC parameters were maintained in the models with ADC parameters. All ADC parameters remained independently prognostic, with MGMT promoter methylation status and other clinical prognostic factors in multivariate prognostic models of OS and PFS (P,.001 for all). Incremental Prognostic Value of ADC Parameters The c indices, which indicate prognostic model performances, are presented in Table 4. For OS, all four models with ADC parameters had significantly higher c indices than did the model without ADC parameters (P =.026, P =.022, P =.017, and P =.024 with mean and 25th, 50th, and 75th percentiles of ADC, respectively). For PFS, the models with ADC parameters tended to have higher c indices than did the models without ADC parameters, with borderline significance for all ADC parameters except the 75th percentile of ADC (P =.055, P =.079, P =.082, and P =.100 with mean and the 25th, 50th, and 75th percentiles of ADC, respectively). The diagnostic accuracies for 12- and 16-month OS and 12-month PFS increased when ADC parameters were added (area under the receiver operating characteristic curve [AUC] = for 12-month OS, AUC = for 16-month OS, and AUC = for 12-month PFS), which was significant for all ADC parameters for predicting both OS and PFS (P,.05 for all) (Table E2 [online]). Eight living patients with a follow-up period of less than 16 months were excluded from the training set for the evaluation of 16-month OS. Validation in the Test Set The diagnostic accuracy for 12- and 16-month OS and 12-month PFS for the test set are presented in Table radiology.rsna.org n Radiology: Volume 281: Number 1 October 2016

7 Table 3 Multivariate Prognostic Models with ADC Histogram Parameters for Predicting PFS Model 1 Model 2 Model 3 Model 4 Parameter HR 95% CI P Value HR 95% CI P Value HR 95% CI P Value HR 95% CI P Value ADC (per mm 2 /sec) Mean , 1.57, th percentile , 1.54, th percentile , 1.48, th percentile , 1.44,.001 MGMT , , , , Age (per 10 years) , , , , KPS (per 210) , , , , Extent of resection * , , , , Note. HR = hazard ratio. * Includes subtotal resection, partial resection, and biopsy only versus total resection. Table 4 Performance Comparison of Multivariate Prognostic Models with and without ADC Parameters for Predicting OS and PFS Model Included Parameters C Index 95% CI P Value OS Model 0 * MGMT and other clinical factors , 0.78 Model 1 + mean of ADC , Model th percentile of ADC , Model th percentile of ADC , Model th percentile of ADC , PFS Model 0 * MGMT and other clinical factors , 0.75 Model 1 + mean of ADC , Model th percentile of ADC , Model th percentile of ADC , Model th percentile of ADC , Note. P values were calculated for comparison with c index of Model 0. MGMT includes MGMT promoter methylation status. * The reference model. Other clinical factors include age, KPS, extent of resection, and enhancing tumor volume. Other clinical factors include age, KPS, and extent of resection. Four living patients with a follow-up period of less than 16 months were excluded from the test set for the evaluation of 16-month OS. The prognostic models with ADC parameters for the test set returned similar or higher AUCs than did those for the training set, with the exception of the mean of ADC for 12-month OS and PFS and the 75th percentile of ADC for 12- and 16-month OS and 16-month PFS. The diagnostic accuracy for 12- and 16-month OS and 12-month PFS was improved when ADC parameters were added. This improvement was significant for the 25th and 50th percentiles of ADC for 16-month OS prediction (P =.040 and P =.047 for the 25th and 50th percentiles, respectively) and the 25th percentile of ADC for 12-month PFS prediction (P =.026) and borderline significant for the 25th percentile of ADC for 12-month OS (P =.058) and the 50th percentile of ADC for 12-month PFS prediction (P =.055). ADC and MGMT Promoter Methylation Status We found no significant differences in ADC histogram parameters according to MGMT status, with P =.694, P =.185, P =.256, and P =.875 for the mean and 25th, 50th, and 75th percentiles of ADC, respectively (Fig 3, Table E3 [online]). Discussion In this study we found that preoperative ADC histogram analysis had incremental prognostic value over MGMT promoter methylation status for predicting OS and PFS in patients with glioblastomas. We found no significant differences in ADC parameters according to MGMT status. Our results imply that preoperative ADC histogram analysis can be useful postoperatively along with MGMT status, which is of clinical relevance given that most patients with glioblastomas undergo surgery in practice. Our results also imply that ADC parameters may not be an appropriate imaging biomarker for predicting MGMT status. The prognostic value of the ADC in patients with glioblastomas has also been previously investigated, with conflicting results (11,13 15,31,32). Romano et al (15) reported that the minimum ADC of the entire enhancing lesion was a prognostic factor, whereas Radiology: Volume 281: Number 1 October 2016 n radiology.rsna.org 181

8 Saksena et al (14) reported that, rather than the minimum ADC, the ADC mean obtained from whole enhancing and nonenhancing lesions had prognostic importance. In contrast, Pope et al (13) analyzed ADC histograms with binomial distribution and found that a Table 5 lower ADC mean was prognostic only in patients with glioblastomas who underwent bevacizumab therapy, not those who underwent temozolomide therapy. Additionally, Wen et al (33) suggested that a lower ADC mean of a binomial distribution within an enhancing lesion Incremental Value of ADC Parameters for Predicting 12- and 16-Month OS and 12-Month PFS in the Test Set Survival and Model Added Parameters * AUC IDI P Value 12-month OS Model (0.588, 0.902) Model 1 + mean (0.645, 0.943) Model th percentile (0.714, 0.973) Model th percentile (0.692, 0.966) Model th percentile (0.600, 0.919) month OS Model (0.693, 0.971) Model 1 + mean (0.806, 1.000) Model th percentile (0.798, 1.000) Model th percentile (0.803, 1.000) 0.090`.047 Model th percentile (0.723, 0.977) month PFS Model (0.637, 0.951) Model 1 + mean (0.680, 0.973) Model th percentile (0.735, 1.000) Model th percentile (0.710, 0.982) Model th percentile (0.655, 0.970) Note. Data in parentheses are 95% CIs. For 12-month OS. n = 38; for 16-monnth OS, n = 34; and for 12-month PFS, n = 38. IDI = integrated discrimination improvement. * Of ADC at initial postoperative MR imaging was prognostic but not significant when adjusting for clinical prognostic factors. Nevertheless, the incremental prognostic value of preoperative ADC histogram analysis over MGMT status was not well established in previous research. Because we attempted to determine the practical prognostic value of ADC histogram analysis, ADC parameters were strictly adjusted for MGMT status and other prognostic factors, including conventional MR imaging findings, to simulate clinical settings in survival analyses. Given that ADC histogram analysis showed almost perfect interobserver agreement, and no operator-dependent input was required during DT image processing, we suggest that standardization and optimization of ADC histogram analysis can be achieved and may improve preoperative prognostication of glioblastomas, which also have incremental value postoperatively. MGMT is a key enzyme in DNA repair that removes alkyl groups from the O6 position of guanine, a site of alkylation by temozolomide (34). Epigenetic silencing by methylation of the MGMT promoter reduces MGMT protein activity (35,36). The MGMT promoter is methylated in 30% 60% of glioblastomas, a characteristic that is associated with better response to temozolomide therapy and better Figure 3 Figure 3: A, Box plot of ADC mean and, B, relative frequency histogram of ADC value according to MGMT status (n = 112). No differences were found according to MGMT status, and relative frequency histograms of ADC values according to MGMT status showed overlapping and similar distributions. 182 radiology.rsna.org n Radiology: Volume 281: Number 1 October 2016

9 prognosis (6,7,36). Previous studies reported a relationship between ADC values and MGMT status, although with controversial results. Several studies reported that ADC ratios and ADC minimum values were higher in tumors with methylated MGMT promoters than with unmethylated promoters and that mean ADC had a positive relationship with the MGMT promoter methylation ratio (16,17). In contrast, one study reported that a lower ADC mean value from binomial distributions was higher in tumors with an unmethylated MGMT promoter than in tumors with methylated MGMT promoters (13), and another study found no significant correlation between ADC values and MGMT status, a finding consistent with our results (18). These conflicting results may be partially attributed to different methodologies, including ROI selection, statistical analysis, and different cohorts (high-grade glioma versus glioblastomas only). We found no significant correlation between ADC parameters and MGMT status. Assuming that both ADC and the Ki-67 index are indicative of tumor cellularity and that they correlate with one another, our results agree with those from previous reports that found no association between MGMT status and Ki- 67 (37,38). We propose that MGMT status and ADC values may contribute to prognosis by two different mechanisms in terms of chemosensitivity and tumor cellularity, through which these two factors add prognostic value to one another. Thus, in our opinion, ADC values may be more useful as an adjunct prognostic factor in combination with MGMT status than as an imaging biomarker that predicts MGMT status. In addition to the retrospective study design, several limitations should be mentioned. First, b = 600 sec/mm 2 was used instead of a b = 1000 sec/ mm 2 or higher to perform DT imaging. This low b value may have resulted in overestimation of ADC values because of the effects of perfusion. Second, various treatments for tumor progression after standard chemoradiation therapy were not considered for OS analysis. However, we believe that this pitfall may have been mitigated by performing PFS analysis. Third, no locus-specific pathologic-radiologic correlations were made, and MGMT can be heterogeneous in a single glioblastoma (39). However, histogram analysis of the entire enhancing tumor appeared to partially reduce this limitation. Fourth, conventional MR imaging findings were not extensively analyzed according to the Visually Accessible Repository for Molecular Brain Neoplasia Data Images (VASARI) lexicon. A recent study (27) found that, in VASARI, only eloquent brain involvement remained independently prognostic for worse OS after adjusting for age and KPS, along with quantitative measurement of enhancing tumor volume instead of qualitative measurement of tumor size from VA- SARI. Thus, we attempted to mitigate this limitation by including such independently prognostic findings. In our study, survival models included more variable factors (eg, extent of tumor resection and MGMT status) than the previous study (27), and only enhancing tumor volume remained, whereas the other conventional MR imaging findings were excluded when selecting variables for multivariate prognostic models. Fifth, contrast-enhanced T1- weighed images and ADC maps were acquired with different section thicknesses, and coregistration of these two images may have been inaccurate. However, we confirmed fair coregistration by visual assessment and excluded patients with poor coregistration. We also confined ADC values of to mm 2 /sec for histogram analysis to exclude cystic components or artifacts in ROIs. In conclusion, our study demonstrated the incremental prognostic value of preoperative ADC histogram analysis as an adjunct to MGMT promoter methylation status in patients with glioblastomas. We found no association between ADC values and MGMT status. We believe that preoperative ADC histogram analysis is a useful imaging biomarker that holds incremental prognostic value over MGMT status after surgery. Disclosures of Conflicts of Interest: Y.S.C. disclosed no relevant relationships. S.S.A. disclosed no relevant relationships. D.W.K. disclosed no relevant relationships. J.H.C. disclosed no relevant relationships. S.-G.K. disclosed no relevant relationships. E.H.K. disclosed no relevant relationships. S.H.K. disclosed no relevant relationships. T.H.R. disclosed no relevant relationships. S.-K.L. disclosed no relevant relationships. References 1. Mohin G, Madajewicz S, Manzione J, Franceschi D. Glioblastoma multiforme: advances in postsurgical management. Community Oncol 2006;3(10): Oh J, Henry RG, Pirzkall A, et al. Survival analysis in patients with glioblastoma multiforme: predictive value of choline-to-n-acetylaspartate index, apparent diffusion coefficient, and relative cerebral blood volume. J Magn Reson Imaging 2004;19(5): Marijnen CA, van den Berg SM, van Duinen SG, Voormolen JH, Noordijk EM. Radiotherapy is effective in patients with glioblastoma multiforme with a limited prognosis and in patients above 70 years of age: a retrospective single institution analysis. Radiother Oncol 2005;75(2): Zikou AK, Alexiou GA, Kosta P, et al. Diffusion tensor and dynamic susceptibility contrast MRI in glioblastoma. Clin Neurol Neurosurg 2012;114(6): Liu Y, Shete S, Etzel CJ, et al. Polymorphisms of LIG4, BTBD2, HMGA2, and RTEL1 genes involved in the double-strand break repair pathway predict glioblastoma survival. J Clin Oncol 2010;28(14): Hegi ME, Diserens AC, Godard S, et al. Clinical trial substantiates the predictive value of O-6-methylguanine-DNA methyltransferase promoter methylation in glioblastoma patients treated with temozolomide. Clin Cancer Res 2004;10(6): Hegi ME, Diserens AC, Gorlia T, et al. MGMT gene silencing and benefit from temozolomide in glioblastoma. N Engl J Med 2005;352(10): Gupta RK, Cloughesy TF, Sinha U, et al. Relationships between choline magnetic resonance spectroscopy, apparent diffusion coefficient and quantitative histopathology in human glioma. J Neurooncol 2000;50(3): Gupta RK, Sinha U, Cloughesy TF, Alger JR. Inverse correlation between choline magnetic resonance spectroscopy signal intensity and the apparent diffusion coefficient in human glioma. Magn Reson Med 1999;41(1): Gauvain KM, McKinstry RC, Mukherjee P, et al. Evaluating pediatric brain tumor cel- Radiology: Volume 281: Number 1 October 2016 n radiology.rsna.org 183

10 lularity with diffusion-tensor imaging. AJR Am J Roentgenol 2001;177(2): 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): Sugahara T, Korogi Y, Kochi M, et al. Usefulness of diffusion-weighted MRI with echo-planar technique in the evaluation of cellularity in gliomas. J Magn Reson Imaging 1999;9(1): Pope WB, Lai A, Mehta R, et al. Apparent diffusion coefficient histogram analysis stratifies progression-free survival in newly diagnosed bevacizumab-treated glioblastoma. AJNR Am J Neuroradiol 2011;32(5): Saksena S, Jain R, Narang J, et al. Predicting survival in glioblastomas using diffusion tensor imaging metrics. J Magn Reson Imaging 2010;32(4): Romano A, Calabria LF, Tavanti F, et al. Apparent diffusion coefficient obtained by magnetic resonance imaging as a prognostic marker in glioblastomas: correlation with MGMT promoter methylation status. Eur Radiol 2013;23(2): Moon WJ, Choi JW, Roh HG, Lim SD, Koh YC. Imaging parameters of high grade gliomas in relation to the MGMT promoter methylation status: the CT, diffusion tensor imaging, and perfusion MR imaging. Neuroradiology 2012;54(6): Sunwoo L, Choi SH, Park CK, et al. Correlation of apparent diffusion coefficient values measured by diffusion MRI and MGMT promoter methylation semiquantitatively analyzed with MS-MLPA in patients with glioblastoma multiforme. J Magn Reson Imaging 2013;37(2): Gupta A, Prager A, Young RJ, Shi W, Omuro AM, Graber JJ. Diffusion-weighted MR imaging and MGMT methylation status in glioblastoma: a reappraisal of the role of preoperative quantitative ADC measurements. AJNR Am J Neuroradiol 2013;34(1): E10 E Kim YS, Kim SH, Cho J, et al. MGMT gene promoter methylation as a potent prognostic factor in glioblastoma treated with temozolomide-based chemoradiotherapy: a single-institution study. Int J Radiat Oncol Biol Phys 2012;84(3): Newson RB. Comparing the predictive powers of survival models using Harrell s C or Somers D. Stata J 2010;10(3): Ko SB, Choi HA, Carpenter AM, et al. Quantitative analysis of hemorrhage volu-me for predicting delayed cerebral ischemia after subarachnoid hemorrhage. Stroke 2011; 42(3): Cha J, Kim ST, Kim HJ, et al. Differentiation of tumor progression from pseudoprogression in patients with posttreatment glioblastoma using multiparametric histogram analysis. AJNR Am J Neuroradiol 2014;35(7): Maes F, Collignon A, Vandermeulen D, Marchal G, Suetens P. Multimodality image registration by maximization of mutual information. IEEE Trans Med Imaging 1997;16(2): Pope WB, Sayre J, Perlina A, Villablanca JP, Mischel PS, Cloughesy TF. MR imaging correlates of survival in patients with highgrade gliomas. AJNR Am J Neuroradiol 2005;26(10): Carrillo JA, Lai A, Nghiemphu PL, et al. Relationship between tumor enhancement, edema, IDH1 mutational status, MGMT promoter methylation, and survival in glioblastoma. AJNR Am J Neuroradiol 2012;33(7): Vasari MRI Visual Feature Guide. The Cancer Imaging Archive (TCIA) Web site. display/public/vasari+research+project. Published April 7, Accessed July 15, Wangaryattawanich P, Hatami M, Wang J, et al. Multicenter imaging outcomes study of The Cancer Genome Atlas glioblastoma patient cohort: imaging predictors of overall and progression-free survival. Neuro Oncol 2015;17(11): Landis JR, Koch GG. The measurement of observer agreement for categorical data. Biometrics 1977;33(1): Contal C, O Quigley J. An application of changepoint methods in studying the effect of age on survival in breast cancer. Comput Stat Data Anal 1999;30(3): Pencina MJ, D Agostino RB Sr, D Agostino RB Jr, Vasan RS. Evaluating the added predictive ability of a new marker: from area under the ROC curve to reclassification and beyond. Stat Med 2008;27(2): ; discussion Murakami R, Sugahara T, Nakamura H, et al. Malignant supratentorial astrocytoma treated with postoperative radiation therapy: prognostic value of pretreatment quantitative diffusion-weighted MR imaging. Radiology 2007;243(2): Zulfiqar M, Yousem DM, Lai H. ADC values and prognosis of malignant astrocytomas: does lower ADC predict a worse prognosis independent of grade of tumor? a metaanalysis. AJR Am J Roentgenol 2013;200(3): Wen Q, Jalilian L, Lupo JM, et al. Comparison of ADC metrics and their association with outcome for patients with newly diagnosed glioblastoma being treated with radiation therapy, temozolomide, erlotinib and bevacizumab. J Neurooncol 2015; 121(2): Esteller M, Herman JG. Generating mutations but providing chemosensitivity: the role of O6-methylguanine DNA methyltransferase in human cancer. Oncogene 2004; 23(1): Riemenschneider MJ, Hegi ME, Reifenberger G. MGMT promoter methylation in malignant gliomas. Target Oncol 2010;5(3): Weller M, Stupp R, Reifenberger G, et al. MGMT promoter methylation in malignant gliomas: ready for personalized medicine? Nat Rev Neurol 2010;6(1): Lee SH, Nam SW, Hong YG, Kang CS, Lee SL. O6-methylguanine DNA methyltransferase gene promoter methylation status in glioblastoma and its correlation with other prognostic markers. Mol Cell Toxicol 2012; 7(4): Huang L, Jiang T, Yuan F, et al. Correlation of chromosomes 1p and 19q status and expressions of O6-methylguanine DNA methyltransferase (MGMT), p53 and Ki-67 in diffuse gliomas of World Health Organization (WHO) grades II and III: a clinicopathological study. Neuropathol Appl Neurobiol 2009;35(4): Della Puppa A, Persano L, Masi G, et al. MGMT expression and promoter methylation status may depend on the site of surgical sample collection within glioblastoma: a possible pitfall in stratification of patients? J Neurooncol 2012;106(1): radiology.rsna.org n Radiology: Volume 281: Number 1 October 2016