Gamma knife radiosurgery for cerebellar brain metastases: clinical outcomes and artificial intelligence-based predictive modeling.

Cerebellar brain metastases pose unique management challenges due to the risk of rapid neurological deterioration. Resection is often considered for large posterior fossa tumors. Gamma Knife radiosurgery (GKRS) is an established treatment for intracranial metastases, yet data dedicated to posterior fossa tumors remain limited. We retrospectively analyzed 490 patients harboring 1296 cerebellar metastases treated with GKRS between 2014 and 2024. Demographic, tumor, and dosimetry variables were collected. Overall survival (OS), local control (LC), and treatment-related toxicity were evaluated. Subgroup analyses examined tumors ≥ 10 cc. In parallel, a feedforward neural network (FNN) was developed to predict the appropriate prescription dose expected to result in the best OS, and LC for a specific patient. Across the cohort, LC was 82.5%, median OS was 10.2 months. Immunotherapy significantly improved OS (13.5 vs. 8.1 months, p < 0.001) and LC (89.4% vs. 78.1%, p = 0.012). Tumors ≥ 10 cc (n = 72) achieved outcomes comparable to smaller tumors, with OS of 11.1 months, LC of 80.6%, and minimal toxicity. Immunotherapy further improved survival and LC in this subgroup. Tumor progression was managed with repeat SRS for 10%, resection for 5%, and WBRT for 2% tumors. The FNN achieved strong predictive performance (R2 = 0.81 for dose, R2 = 0.77 for OS, AUC = 0.83 for LC), demonstrating feasibility of artificial intelligence for radiosurgical planning. GKRS provides safe and effective treatment for cerebellar metastases, including large tumors. This is the first study to integrate an FNN for outcome prediction in Gamma Knife radiosurgery, establishing a foundation for personalized, data-driven neurosurgery.