AI-Driven Classification of Medulloblastoma Tumor Subtypes Using Novel Features Derived From Apparent Diffusion Coefficient MRI

Motivation: Identifying medulloblastoma subtypes is critical for effective treatment, as different subtypes exhibit unique prognostic profiles. Goal(s): This study aims to develop a novel set of features derived from ADC MRI images that can be used by a machine learning model to classify medulloblastoma subtypes. Approach: Using retrospective MRI data from medulloblastoma patients, we extracted ADC-based features and trained a classifier to distinguish among medulloblastoma subtypes. Results: Our developed features found significant relationships between the different subtypes and were able to independently predict tumor subtype with 68% accuracy. Impact: This AI-based approach could enable early, non-invasive classification of medulloblastoma subtypes, aiding personalized treatment planning. It provides novel features that have interoperable meaning. This work adds another tool to aid in medulloblastoma tumor classification.