Multi-Modality Modeling of Glioblastoma Progression: Integrating DTI and Prognostic Biomarkers for Personalized Radiation Therapy Targeting

About This Grant

Project Abstract Glioblastoma outcomes have not improved substantially over the past decades, with median survival remaining at 12-15 months despite aggressive therapy. A major limitation in the current radiotherapy (RT) planning is it defines clinical target volumes (CTVs) using uniform geometric expansions around MRI-visible tumor boundaries. This conventional approach fails to capture GBM’s diffuse infiltrative spread and ignores patient-specific tumor biology. As a result, microscopic disease often extends beyond the treated field while normal brain is unnecessarily irradiated, leading to universal recurrences and treatment-induced toxicity. Although diffusion tensor imaging (DTI) can visualize white matter tracts, current tractography does not distinguish between tract pathways that facilitate tumor cell migration and those that are anatomically present but rarely involved in tumor spread. Additionally, molecular biomarkers such as MGMT methylation and TERT promoter mutations reflect distinct tumor progression patterns, yet these factors are not incorporated into RT target delineation. This project investigates DTI-based infiltrative risk mapping integrated with molecular biomarkers to improve glioblastoma progression prediction and RT CTV definition using a dataset of over 500 patients with pre-operative DTI, anatomical MRI, and molecular biomarker data. We will develop White Matter Infiltrative Risk maps by identifying population-level infiltration patterns across major white matter tracts and combining these with patient-specific fiber density maps. The infiltrative risk maps will be integrated with anatomical MRI, MGMT methylation and TERT promoter mutation status through a transformer-based deep learning framework with cross-attention mechanisms. Validation will be conducted via spatial accuracy assessment against ground truth progression, comparison with standard RT targets, and histopathological correlation using tissue samples with matched imaging coordinates from 298 patients. This fusion of advanced DTI mapping and genomics with state-of-the-art Artificial Intelligence modeling will produce voxel-level risk maps that reveal otherwise occult tumor infiltration pathways and can be directly incorporated into RT planning. This work will provide proof-of-concept for integrating infiltration pathways and biological factors into RT target definition and establish the foundation for future clinical trials testing personalized radiation therapy strategies. The goal is to transition from geometric margins to biology-guided targeting that improves GBM control while preserving healthy brain tissue.