Glioblastoma (GBM) is an aggressive brain tumor with significant survival variability, necessitating advanced prognostic tools for personalized treatment. Radiomic features extracted from medical imaging offer a non-invasive approach to characterizing tumor heterogeneity and predicting clinical outcomes. This study aimed to use radiomic analysis to assess survival variability in GBM and develop a predictive tool to support clinical decision-making.

We analyzed a cohort of GBM patients from the TCGA-GBM dataset, extracting radiomic features from contrast-enhanced T1-weighted MRI (T1ce) scans. A total of 1,213 radiomic features were extracted, and feature selection was performed in two stages, first with univariate analysis (P < 0.05) followed by LASSO regression to refine the set. The selected features were weighted using LASSO coefficients to develop the “Radiomics Survival Predictor (RaSPr)” score, stratifying patients into low- and high-risk categories. Kaplan-Meier analysis was performed to evaluate the survival prediction efficiency. Independent datasets, UCSF-GBM and UPENN-GBM, were used for validation. Differential gene expression (DEG) analysis provided biological insights into tumor characteristics, and the RaSPr score was integrated into our self-developed "RadGLO" web platform for personalized survival predictions.

From the initial 1,213 radiomic features, 8 key features were identified to construct the Radiomics Survival Predictor (RaSPr) score, stratifying patients into low- and high-risk categories. Kaplan-Meier analysis showed that high RaSPr scores were associated with worse survival (P < 0.05), demonstrating its predictive power. We validated RaSPr using the UCSF-GBM and UPENN-GBM datasets, confirming its robustness and generalizability across cohorts. Differential gene expression analysis revealed significant upregulation of MDM2 in high RaSPr patients. MDM2, a negative regulator of the tumor suppressor protein p53 that promotes tumorigenesis, links radiomic risk with molecular markers. To enhance clinical adoption, RaSPr is integrated into "RadGLO" (https://project.iith.ac.in/cgntlab/radglo/), allowing clinicians and researchers to submit MRI images or radiomic features for personalized survival risk predictions.

These findings underscore the potential of RaSPr score in personalizing GBM prognosis and guiding risk-based treatment strategies. By offering a non-invasive, efficient method for survival risk assessment, RaSPr can significantly enhance clinical decision-making and facilitate tailored treatment plans to improve patient outcomes. Integrating RaSPr into the 'RadGLO' web platform further enhances accessibility, providing clinicians and researchers with an easy-to-use tool for submitting MRI images or radiomic features and receiving personalized survival predictions.