A data-driven scientific approach to explore the relationship between MGMT methylation and imaging phenotype in glioblastoma.

Glioblastoma is a highly malignant brain tumor, and patient survival remains short despite multimodal treatment. One important prognostic factor is the methylation status of the O⁶-methylguanine DNA methyltransferase (MGMT) gene, as MGMT promoter methylation is associated with a better response to temozolomide chemotherapy. The aim of this study was to evaluate the association between noninvasively obtained imaging features and MGMT methylation status. Magnetic resonance (MR) images and MGMT methylation data for 55 glioblastoma cases (33 methylated, 22 unmethylated) were obtained from the publicly available TCGA-GBM database. After brain morphological standardization, tumor centroid coordinates were calculated and used as inputs for a quadratic discriminant classifier to assess the separability of MGMT methylation status based on anatomical location. In addition, 279 radiomic features, including histogram- and texture-based metrics, were extracted from tumor regions. Using Lasso regression, nine features were selected and subsequently analyzed using linear discriminant analysis. The areas under the receiver operating characteristic curve (AUCs) were 0.71 for anatomical location-based classification and 0.85 for radiomic feature-based classification. Because the correlation between anatomical location and radiomic features was low, integrating both feature sets improved discrimination performance to an AUC of 0.90. These results suggest that imaging examinations capture complementary information regarding tumor morphology and anatomical origin. The anatomical location of glioblastoma may therefore provide useful clues for classifying MGMT gene methylation status.