Skip to main content
2025 Neural Interfaces Meeting Main banner
Unparalleled proactive care. Optimizing the SCS experience. Website: neuro.biotronik.com

(16) Automated deep brain stimulation programming via meta-active learning of evoked potentials

Thursday, June 12, 2025
5:50 PM - 6:30 PM ET
Location: Sky View
Introduction: Deep brain stimulation (DBS) parameters must be selected manually for each patient in a months-long trial-and-error process, delaying optimal clinical outcomes. We have developed a fully automated approach to optimize STN-DBS parameters for Parkinson's disease intraoperatively.

Methods: Our approach uses meta-active learning (meta-AL), an AI method that learns patterns of similarity across a prior patient database to interactively apply stimulation parameters, measure the effect on neural biomarkers in real-time, and iterate to optimize parameters most quickly in new patients. Its optimization goal is to maximize DBS local evoked potentials (DLEP) while minimizing EMG-measured motor evoked potentials (mEP)–electrophysiological biomarkers of pathway activation for symptom relief and side-effects, respectively.

Results: Using annotated EMG recordings from 9 patients, we first develop and validate an automated mEP detection algorithm with 94% accuracy in identifying which DBS settings induced mEP. We then apply meta-learning to predict DLEP and mEP values as a function of stimulation parameters in 59 intraoperative patients (61 STN, 10 GPi hemispheres; mean 24 settings/lead), showing that a meta-trained neural network can predict biomarker values for unseen parameter combinations with high cross-validation accuracy (DLEP: 0.879 correlation, mEP: 0.767). We then use this modeled data to provide the basis for a simulation environment where meta-AL can practice DBS programming, showing that meta-AL can identify optimal parameters of simulated patients in silico within 50 tested stimulation settings and more efficiently than various control methods. To deploy meta-active learning intraoperatively, we describe an open-source system for real-time biomarker recording and parameter selection in Python. Last, we demonstrate preliminary results of deploying meta-AL intraoperatively in an initial patient cohort, showing that meta-AL biomarker predictions generalize to newly collected held-out stimulation data and can predict therapeutically effective stimulation settings.

Conclusion: Long-term, our approach could reduce DBS programming time by providing expected best parameters before clinical testing and could enable effective programming for other DBS applications (epilepsy, neuropsychiatric diseases, etc.) where behavioral feedback is limited.