Electrophysiological signatures predict the therapeutic window of deep brain stimulation electrode contacts

Deep brain stimulation (DBS) is an effective treatment for Parkinson's disease. Identifying the optimal parameters is a complex task. Here, we investigated whether electrophysiology, combined with machine learning, can support contact selection. We applied tree learning to resting-state magnetoencephalographic and local field potential recordings from the subthalamic nucleus (STN). STN power and STN-cortex coherence in various frequency bands served to predict the therapeutic window. The model successfully predicted therapeutic windows in the original (r = 0.45, p < 0.001, N = 45) and in an independent cohort (r = 0.30, p < 0.001, N = 8). It relied mostly on fast (>35 Hz) subthalamic activity and on STN-cortex coherence in several bands. Furthermore, it was able to order contacts such that the optimal contact can be found faster. Our study demonstrates the feasibility of predicting therapeutic windows from electrophysiological features and could contribute to automated contact selection in the future.