Online prediction of optimal deep brain stimulation contacts from local field potentials in Parkinson's disease

Selecting optimal contacts for chronic deep-brain stimulation (DBS) requires a monopolar review, involving time-consuming manual testing by trained personnel, often causing patient discomfort. Neural biomarkers, such as local field potentials (LFP), could streamline this process. This study aimed to validate LFP recordings from chronically implanted neurostimulators for guiding clinical contact-level selection. We retrospectively analysed bipolar LFP recordings from Parkinson's disease patients across three centres (Netherlands: 68, Switzerland: 21, Germany: 32). Using beta-band power measures (13-35 Hz), we ranked channels based on clinical contact-level choices and developed two prediction algorithms: (i) a "decision tree" method for in-clinic use and (ii) a "pattern based" method for offline validation. The "decision tree" method achieved accuracies of 86.5% (NL), 86.7% (CH), and 75.0% (DE) for predicting the top two contact-levels. Both methods outperformed an existing algorithm. These findings suggest LFP-based approaches can enhance DBS programming efficiency, potentially reducing patient burden and clinical workload.