Non-Invasive Molecular Classification of Gliomas Using Explainable ai and Radiomic Features from Multi-Sequence MRI

Accurate molecular classification of brain gliomas is crucial for effective diagnosis and treatment, but existing methods mainly rely on invasive biopsies. To address this issue, we developed an explainable AI-based diagnostic system that combines radiomic features from multi-sequence MRI and clinical data to predict key molecular subtypes based on WHO 2021 criteria. Analyzing MRI and molecular data from 117 glioma patients, we extracted intensity, shape, and texture features. Our model was trained to classify various molecular subtypes separately, achieving high classification accuracy with AUC scores exceeding 0.8 for IDH mutations, CDKN2A/B deletions, and chromosomal alterations, as determined by 5-fold cross-validation. Features from T1 precontrast and FLAIR imaging of enhancing tumors and edematous regions were particularly predictive. This noninvasive method shows promise for glioma molecular classification, enhancing personalized treatment and clinical decision-making.