Prediction of molecular markers based on glioma subregions through radiomics and machine learning

Motivation: Intratumoral heterogeneity is a crucial feature of gliomas, and its potential to aid in molecular subtyping and disease prediction warrants further investigation. Goal(s): To develop a radiomics model that predicts glioma molecular biomarkers by studying subregions, thereby improving understanding and treatment. Approach: A retrospective study employing machine learning on MRI data from 316 glioma patients to predict WHO classification, IDH mutation status, and 1p/19q co-deletion status through radiomics feature extraction. Results: Support Vector Machine (SVM) predicted 1p/19q status in the edema region, while Random Forest (RF) effectively predict IDH status in both the edema and enhancing region. Impact: Our study predicts biomarkers using glioma subregion radiomics and machine learning , enabling physicians to personalize treatment non-invasively. This approach will inspire other researchers to validate larger datasets and incorporate clinical genetic data.