Truace
Subscribe

AI model predicts tremor relief and side effect risk for patients after Gamma Knife treatment

Study title: 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

Researchers built THINKERS-ET, a mixture-of-experts neural network trained on variables known before or at treatment for patients with medically refractory essential tremor who underwent Gamma Knife thalamotomy. It predicts functional improvement, time to benefit, and adverse radiation effects, and lets clinicians query different maximum doses.

Read the study abstract

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 perfo…

By the Truvace desk1 min read1 sourcePeer-reviewedTRV-2026-2236· cite & verify
Peer-reviewed study: The Cerebellum

In brief

Current outcome prediction relies on baseline severity, anatomy, dose parameters, and institutional experience, so individualized estimates could support dose planning and patient counseling. The evidence so far comes from one center and internal validation only, so performance in other settings and in clinical decisions remains untested.

Main points

  1. The cohort covered 163 Gamma Knife thalamotomy procedures for medically refractory essential tremor at a single center.
  2. Clinical benefit occurred in 152 procedures (93.3%), complete tremor arrest in 39 (23.9%), recurrence in 16 (9.8%).
  3. Adverse radiation effects occurred in 6 procedures (3.7%), and the model reached an AUC of 0.857 for predicting them.
  4. Time to benefit was predicted with a mean absolute error of 1.17 months, and percent FTM improvement with an error of 2.21%.

The gain

An internally validated AI model estimated individual tremor improvement, time to benefit, and adverse radiation effect risk, and could test candidate doses, in 163 thalamotomy procedures.

The rundown

The study was retrospective and treatment-level. Inputs available before or at treatment were used to train the model, with maximum dose as a queryable input so that candidate doses could be evaluated repeatedly.

Validation used holdout testing and 5-fold cross-validation. Binary outcomes were assessed with AUC and Brier score, continuous ones with MAE, RMSE, and R². Brier scores were 0.0025 for clinical benefit and 0.0057 for adverse radiation effects.

Outcomes in the cohort were mostly favorable, with clinical benefit in 93.3% of procedures. Adverse radiation effects were rare at 6 events, which is a small number for a prediction target.

Sources

  1. Peer-reviewedThe Cerebellum2026-10-23

The debate