Supervoxel-based multimodal MRI biomarkers reveal tumor heterogeneity in high-grade glioma for prognostic stratification and therapy response prediction

High-grade glioma (HGG) exhibits substantial biological heterogeneity, which complicates prognosis and treatment, and clarifying the interplay between proliferating tumor cells and tumor-associated microvasculature may improve patient outcomes. We prospectively enrolled 221 patients with HGG from four institutions and integrated diffusion-weighted and dynamic contrast-enhanced MRI within a supervoxel-based framework to delineate tumor subregions termed density-enhancement compounded voxels (DECV). Four DECV subregions (DECV1-4) with distinct imaging characteristics were identified. DECV4, characterized by low apparent diffusion coefficient and gradual enhancement, was strongly associated with tumor aggressiveness and treatment resistance. Clustering analysis further revealed two DECV phenotypes that differed significantly in progression-free and overall survival; phenotype II showed a higher DECV4 proportion and a poorer prognosis. DECV phenotypes outperformed conventional imaging markers as independent predictors of survival and were validated in an independent cohort. These DECV-based imaging phenotypes provide a robust, non-invasive biomarker for characterizing HGG heterogeneity and show potential to enhance prognostic stratification and guide personalized therapy.