AI Tool Predicts Glioblastoma Recurrence Site After Surgery

Researchers at the University of California, San Francisco, and the University of Michigan have developed a novel artificial intelligence system capable of predicting the precise locations where glioblastoma is most likely to recur following surgical resection. Published in Science Advances and funded by the National Institutes of Health, the breakthrough introduces a predictive framework designed to enable preemptive therapeutic interventions before recurrent tumors become detectable via conventional magnetic resonance imaging. Glioblastoma remains the most lethal primary brain tumor in adults, carrying a median post-diagnosis survival of approximately seventeen months. Despite maximal safe resection and adjuvant therapies, recurrence is nearly universal and typically emerges at the tumor margin. To address this clinical challenge, the research team engineered FastGlioma, a machine learning model trained on stimulated Raman histology imaging. Unlike traditional pathology methods that require lengthy chemical processing and staining, stimulated Raman histology generates high-resolution microscopic images of fresh, unprocessed tissue in under a minute. FastGlioma analyzes these images to quantify tumor cell infiltration across surgical samples. Initial testing demonstrated that AI-derived infiltration scores matched the predictive accuracy of conventional pathology. However, the model performance significantly improved when infiltration metrics were integrated with clinical, radiological, and molecular data across six distinct machine learning configurations. In five of these configurations, the AI infiltration measurement emerged as the strongest independent predictor of recurrence, surpassing traditional molecular biomarkers. Furthermore, the system demonstrated high spatial precision, accurately forecasting recurrence within five to ten millimeters of sampled tissue boundaries. The clinical implications of this technology center on altering the standard of care for neuro-oncology. By identifying high-risk zones prior to imaging evidence of regrowth, surgical and oncological teams can deploy targeted interventions during the initial operation or immediately thereafter. Potential strategies include extended resection of high-infiltration margins when anatomically feasible, or the delivery of focal radiation therapy and convection-enhanced drug delivery directly into predicted progression zones. The research team emphasized that extending the interval to the first recurrence is a critical therapeutic objective, as subsequent disease progression often complicates treatment efficacy and accelerates clinical decline. Co-first author Sanjeev Herr noted that the system offers neurosurgeons actionable real-time guidance during tumor removal, while senior author Shawn Hervey-Jumper highlighted its potential to inform personalized adjuvant strategies for unresectable disease areas. Co-senior author Todd Hollon reinforced that the primary aim is to prolong the recurrence-free period, ultimately translating into extended overall survival. The model was validated using a training cohort of approximately thirty samples from sixty patients and independently tested on one hundred samples from twenty additional patients, reflecting a median time to first recurrence of five and a half months. As glioblastoma management continues to evolve, this integrated imaging and artificial intelligence platform represents a significant step toward precision neuro-oncology, transforming post-surgical surveillance from reactive monitoring to proactive threat mitigation.