NIMG-83. ELUCIDATING INTRA-TUMORAL HETEROGENEITY AND DISTINCT RADIOMIC FEATURES IN GLIOBLASTOMA METHYLATION SUBCLASSES

Abstract

Abstract Glioblastoma, IDH-wildtype (GBM), is characterized by the inter- and intra-tumoral molecular heterogeneity. Genome-wide DNA methylation profiling has recently emerged as a promising tool offering complementary value to the classification of central nervous system tumors. In adult GBM patients, the most common methylation subclasses comprise RTK I, RTK II, and MES. We hypothesize that quantification of intricate and spatially complex radiomic features extracted from multi-parametric MRI (mpMRI) is informative to non-invasively determining characteristics of GBM methylation subclasses and their heterogeneity. In this study, mpMRI scans (T1, T1-Gd, T2, T2-FLAIR, DSC, DTI) of newly diagnosed 23 GBM patients were retrospectively collected. We derived 6339 radiomic features, including histograms, morphologic and textural descriptors. For each case, 3-4 multiple samplings from different parts of the tumor were conducted. DNA was extracted from FFPE for each of sample and analyzed for genome-wide DNA methylation patterns using the Illumina EPIC array. Classification of methylation subclasses was performed using the MNP brain tumor classifier v12b6 of the DKFZ. Analysis of variance (ANOVA) with Bonferroni correction was employed to extract subclass-specific radiomic features. Of the 23 patients, 7 indicated intra-tumoral heterogeneity in methylation subclasses. We identified 68 radiomic features (p< 0.05) and 12 radiomic features (p< 0.01) that showed statistically significant differences among the methylation subclasses as determined by ANOVA with Bonferroni correction. A heatmap illustrating the relationships between subclasses and extracted radiomic features displayed distinct trends among subclasses and between heterogeneous and homogeneous cases within the same subclass. Our findings suggest the presence of heterogeneity within methylation subclasses and distinctive imaging patterns for each subclass. These findings potentially pave the way for a more personalized approach to glioblastoma management, highlighting the importance of further validation through the analysis of additional cases.