Radiogenomics and machine learning predict oncogenic signaling pathways in glioblastoma

Background Glioblastoma (GBM) is a highly aggressive brain tumor with poor prognosis. Despite standard therapies, survival remains low, underscoring the need for novel strategies. MRI is central to GBM assessment. Disruptions in oncogenic pathways (RTK-RAS-ERK, PI3K, TP53, NOTCH) contribute to tumor development with distinct phenotypes. Identifying genetic abnormalities for targeted therapy often requires invasive procedures; radiogenomics offers a non-invasive alternative. This study evaluates radiogenomics and machine learning (ML) to predict oncogenic signaling pathways in GBM. Methods Post-operative MRI scans (T1w, T1c, FLAIR, T2w) from BRATS-19 linked to TCGA/CPTAC were used. Pathway alteration data were manually extracted from cBioPortal. Radiomic features were computed with PyRadiomics, followed by standardization, dimensionality reduction/feature selection, and class balancing with SMOTE. Five ML classifiers were trained with Grid Search hyperparameter tuning and 5-fold cross-validation; performance was evaluated on test data. Results Most signaling pathways showed positive association with MRI-derived radiomic features. Best AUCs achieved were 0.7 (RTK-RAS), 0.8 (PI3K), 0.75 (TP53), and 0.4 (NOTCH), indicating potential for ML models to predict pathway deregulation and inform personalized therapy. Conclusion We present a non-invasive approach to predict oncogenic pathway deregulation in GBM by integrating radiogenomics with ML. This research contributes to advancing precision medicine in GBM management, highlighting the importance of combining radiomics with genomic data to better understand tumor behavior and treatment response.

Abdul Basit Ahanger, Syed Wajid Aalam, Tariq Ahmad Masoodi, Asma Shah, Meraj Alam Khan, Ajaz A. Bhat, Assif Assad, Muzafar Ahmad Macha, Muzafar Rasool Bhat