Diffuse gliomas cause over 1,500 deaths annually in Australia and remain among the most aggressive and complex brain cancers. Molecular markers such as IDH mutation, 1p/19q co-deletion, and EGFR amplification are now central to diagnosis, prognosis, and treatment planning. This talk will present two AI-based approaches for identifying molecular biomarkers from digital whole-slide images (WSIs) and magnetic resonance imaging (MRI), respectively. We will discuss the advantages and limitations of current deep learning frameworks in WSI analysis and introduce our prototype learning approach, which leverages morphological features of WSIs to improve EGFR biomarker prediction with greater generalizability, faster inference, and clinically aligned interpretability. We will further present a non-invasive MRI-based approach: MTS-UNET, a SWIN-UNETR–based framework that integrates tumor-aware feature encoding with cross-modality differential cues to segment gliomas and predict IDH mutation status, 1p/19q co-deletion, and tumor grade directly from routine MRI scans.