New AI model beats benchmarks for deadly brain cancer prognosis by 9%

Researchers developed TopoGBM, a novel AI framework using topological neural networks to analyze glioblastoma (GBM) brain tumors from 3D MRI scans. It addresses critical failures of standard deep learning by capturing extreme tumor heterogeneity and maintaining accuracy across different hospital scanners. The model achieved a C-index of 0.67 in testing, outperforming baselines, and localized 50% of its prognostic signal to the tumor microenvironment, demonstrating strong clinical reliability in unsupervised learning.