MRI-based radiomic clustering identifies a glioblastoma subtype enriched for neural stemness and proliferative programs

Introduction Glioblastoma (GBM) is a highly aggressive brain tumor with a median survival of only 15 months. A major challenge in GBM management is the pronounced inter- and intratumoral heterogeneity, which complicates prognosis and therapy. Radiomics, the quantitative extraction of features from medical images, can capture this heterogeneity across the entire tumor volume, but the biological basis of radiographic phenotypes remains poorly understood. Methods We integrated preoperative MRI-based radiomic stratification with multi-platform transcriptomics (bulk RNA-seq, single-cell RNA-seq, and spatial transcriptomics) in IDH-wildtype GBM patients. Unsupervised clustering of radiomic features identified four imaging subtypes. Results Group 4 emerged as a high-risk subtype associated with significantly worse survival and a distinctive MRI pattern of peripheral contrast enhancement. Transcriptomic analyses revealed that Group 4 tumors were enriched in cell-cycle and proliferation markers and exhibited neural stem cell–like gene expression signatures. Single-cell profiling confirmed an elevated proportion of stem-like malignant cells in this subtype. Spatial transcriptomics further demonstrated that these proliferative, stem-like programs were localized predominantly to the tumor periphery, corresponding to the rim-enhancing regions on MRI. Finally, we identified the developmental transcription factor VAX2 as a candidate driver of the Group 4 gene network; functional assays showed that VAX2 promotes GBM cell proliferation in vitro. Discussion Our findings link a radiomics-defined MRI phenotype to specific molecular programs and cell populations in GBM, suggesting that radiomic subtypes can serve as noninvasive biomarkers of tumor biology and highlighting potential therapeutic targets in aggressive, stem-like tumor cell populations.