Therapeutic Hybrid Intelligence with Neural and Knowledge-based Expert Reasoning for SRS (THINKERS): an AI Model for Functional Outcome Prediction After Gamma Knife Thalamotomy for Essential Tremor.

Outcome prediction after Gamma Knife thalamotomy for essential tremor remains largely guided by baseline tremor severity, target anatomy, radiosurgical dose parameters, and institutional experience rather than individualized estimates of functional improvement, latency to benefit, and adverse radiation effect risk. We developed THINKERS-ET, a mixture-of-experts (MoE) artificial intelligence framework for personalized functional outcome prediction and dose-policy evaluation after Gamma Knife thalamotomy. We performed a retrospective single-center treatment-level study of patients with medically refractory essential tremor treated with Gamma Knife thalamotomy. Variables available before or at treatment were used to train a MoE neural network. Maximum dose was incorporated as a queryable input to enable repeated candidate dose evaluation. Internal validation used holdout testing and 5-fold cross-validation. Performance was assessed using AUC and Brier score for binary outcomes, and MAE, RMSE, and R² for continuous outcomes. The cohort included 163 Gamma Knife thalamotomy procedure. Clinical benefit was observed in 152 procedures (93.3%), complete tremor arrest in 39 (23.9%), recurrence in 16 (9.8%), and adverse radiation effects in 6 (3.7%). THINKERS-ET predicted total FTM improvement with a MAE of 1.78 points, percent FTM improvement with a MAE of 2.21%, and time to benefit with a MAE of 1.17 months. Brier scores were 0.0025 for clinical benefit and 0.0057 for adverse radiation effects. Adverse radiation effect prediction achieved an AUC of 0.857 ± 0.033. THINKERS-ET provides an internally validated framework for personalized functional outcome prediction and dose-policy evaluation after Gamma Knife thalamotomy for essential tremor.