PREDICT-GBM: A multicenter platform advancing personalized glioblastoma radiotherapy planning

Abstract Glioblastoma recurrence is largely driven by diffuse infiltration beyond radiologically visible margins, yet current radiotherapy guidelines rely on uniform margin expansions that ignore patient-specific biology and anatomy. While computational models promise to map this invisible growth and guide personalized planning, their clinical translation is hindered by a lack of standardized benchmarking and reproducible validation. To bridge this gap, we present PREDICT-GBM, an open-source platform integrating a curated, longitudinal, multi-center dataset of 243 patients with a standardized evaluation pipeline. We benchmark a novel U-Net-based recurrence prediction model against state-of-the-art biophysical and data-driven methods. Under iso-volumetric constraints, both biophysical and deep-learning approaches achieved modest but statistically significant gains in geometric coverage of future recurrence over guideline-based plans. On the combined cohort, our U-Net achieved the highest mean coverage of enhancing recurrence (79.37 ± 2.08%), surpassing guideline-based plans (paired Wilcoxon signed-rank test, Benjamini-Hochberg adjusted p = 2.9 × 10 −5 ). The biophysical model GliODIL reached 78.91 ± 2.08% ( p = 1.0 × 10 −3 ), validating the platform’s ability to compare diverse modeling paradigms. By providing a reproducible ecosystem for model training and validation, PREDICT-GBM addresses a major bottleneck toward personalized, computationally guided radiotherapy. The platform, models, and data are openly available at github.com/BrainLesion/PredictGBM .