MRI-based Radiogenomic Classification of Medulloblastoma using Deep Learning Models

Medulloblastoma (MB) is a heterogeneous pediatric brain tumor comprising four molecular subgroups (WNT, SHH, Group 3, Group 4), each with distinct genetic profiles and prognostic outcomes. Current diagnostic techniques rely heavily on invasive biopsy and lack the precision needed for subgroup classification and personalized treatment planning. This study presents a radiogenomic deep learning framework that integrates statistical and texture-based MRI features with Convolutional Neural Networks (CNN) and ResNet-50 for noninvasive MB classification and survival prediction. The proposed models were trained on multimodal MRI datasets and achieved high classification accuracy—99.31% with CNN and 98.75% with ResNet-50—demonstrating strong performance in distinguishing MB subtypes and normal brain scans. The inclusion of Gray- Level Co-occurrence Matrix (GLCM) features and statistical descriptors enhanced interpretability and classification reliability. This research validates the potential of AI-driven radiogenomic models in improving diagnostic precision and enabling risk-adapted, personalized therapy for pediatric medulloblastoma.