• The ICI score derived from gene expression profile of immune cells infiltrating GBM correlates with overall survival and is an effective prognostic biomarker.• In this study, the authors developed a radiomics-based machine learning model able to identify gene expression profiles of GBM intratumoral stromal and immune cells and predict the ICI score on the preoperative MRI scans with high accuracy.• Radiogenomics could potentially be applied in primary brain tumors to noninvasively assess the specific tumor immune characteristics, predict patients' prognosis and identify those patients with higher probability to respond to immunotherapy.