Ji Eun Park1, Ho Sung Kim, Myeong Ju Goh, Sang Joon Kim, Jeong Hoon Kim. 1. From the Department of Radiology and Research Institute of Radiology (J.E.P., H.S.K., M.J.G., S.J.K.) and Department of Neurosurgery (J.H.K.), University of Ulsan College of Medicine, Asan Medical Center, 86 Asanbyeongwon-Gil, Songpa-Gu, Seoul 138-736, Korea.
Abstract
PURPOSE: To validate a volume-weighted voxel-based multiparametric clustering (VVMC) method for magnetic resonance imaging data that is designed to differentiate between pseudoprogression and early tumor progression (ETP) in patients with glioblastoma in an independent test set. MATERIALS AND METHODS: This retrospective study was approved by the local institutional review board, with waiver of the need to obtain informed consent. The study patients were grouped chronologically into a training set (108 patients) and a test set (54 patients). The reference standard was pathologic findings or subsequent clinical-radiologic study results. By using the optimal cutoff determined in the training set, the diagnostic performance of VVMC was subsequently tested in the test set and was compared with that of single-parameter measurements (apparent diffusion coefficient [ADC], normalized cerebral blood volume [nCBV], and initial area under the time-signal intensity curve). RESULTS: Interreader agreement was highest for VVMC (intraclass correlation coefficient, 0.87-0.89). Receiver operating characteristic curve analysis revealed that VVMC performed the best as a classifier, although statistical significance was not demonstrated with respect to the nCBV in the training set. In the test set, the diagnostic accuracy of VVMC was higher than that of any single-parameter measurements, but this trend reached significance only for the ADC. When the entire population was considered, VVMC had significantly better diagnostic accuracy than did any single parameter (P = .003-.046 for reader 1; P = .002-.016 for reader 2). Results of fivefold cross validation confirmed the trends in both the training set and the test set. CONCLUSION: VVMC is a superior and more reproducible imaging biomarker than single-parameter measurements for differentiating between pseudoprogression and ETP in patients with glioblastoma. Online supplemental material is available for this article. RSNA, 2015
PURPOSE: To validate a volume-weighted voxel-based multiparametric clustering (VVMC) method for magnetic resonance imaging data that is designed to differentiate between pseudoprogression and early tumor progression (ETP) in patients with glioblastoma in an independent test set. MATERIALS AND METHODS: This retrospective study was approved by the local institutional review board, with waiver of the need to obtain informed consent. The study patients were grouped chronologically into a training set (108 patients) and a test set (54 patients). The reference standard was pathologic findings or subsequent clinical-radiologic study results. By using the optimal cutoff determined in the training set, the diagnostic performance of VVMC was subsequently tested in the test set and was compared with that of single-parameter measurements (apparent diffusion coefficient [ADC], normalized cerebral blood volume [nCBV], and initial area under the time-signal intensity curve). RESULTS: Interreader agreement was highest for VVMC (intraclass correlation coefficient, 0.87-0.89). Receiver operating characteristic curve analysis revealed that VVMC performed the best as a classifier, although statistical significance was not demonstrated with respect to the nCBV in the training set. In the test set, the diagnostic accuracy of VVMC was higher than that of any single-parameter measurements, but this trend reached significance only for the ADC. When the entire population was considered, VVMC had significantly better diagnostic accuracy than did any single parameter (P = .003-.046 for reader 1; P = .002-.016 for reader 2). Results of fivefold cross validation confirmed the trends in both the training set and the test set. CONCLUSION: VVMC is a superior and more reproducible imaging biomarker than single-parameter measurements for differentiating between pseudoprogression and ETP in patients with glioblastoma. Online supplemental material is available for this article. RSNA, 2015