Objectives/Goals: This project aims to improve gait dysfunction treatment in Parkinson’s disease (PD) by developing adaptive deep brain stimulation (aDBS) that automatically modulates stimulation in response to specific movement states. We use chronic at-home cortical-pallidal recordings to classify walking vs. non-walking and turning vs. straight walking. Methods/Study Population: Local field potentials from globus pallidus (GP) and electrocorticography from premotor (PM) and primary motor (M1) cortices were recorded in four people with PD (2M/2F) implanted unilaterally (n=2) or bilaterally (n=2) with bidirectional neurostimulators (Summit RC+S, Medtronic, Inc). Concurrent at-home movement was captured with wearable ankle sensors (Rover, Sensoplex Inc). Neural-kinematic signals were segmented into 10-second walking/non-walking and 1-second turning/straight walking epochs. Logistic regression and linear discriminant analysis models classified movement states using power within various frequency bands. Random forests identified walking biomarkers compatible with the Summit RC+S device; real-time decoding performance was evaluated using in silico simulations. Results/Anticipated Results: Over 80 hours of at-home neural-kinematic data were analyzed across 6 hemispheres. M1 alpha (8-13Hz) and beta (13-30Hz) power was lower during walking compared to non-walking in all hemispheres. M1 and PM theta (4-8Hz), alpha, and beta power were lower during turning compared to straight walking in all but one hemisphere. Features from GP were most important for walking/non-walking classification, while cortical features predominated for turning/straight walking. Despite these shared features, the most important frequency ranges for classification varied widely across individuals. Within-subject walking/non-walking classifiers achieved strong performance (AUC:0.77-0.96), with simulations of real-time on-board decoding also demonstrating above-chance decoding in all cases (AUC:0.63-0.85). Discussion/Significance of Impact: These results support the hypothesis that cortical-basal ganglia oscillations are modulated by specific movement states. Furthermore, this work demonstrates one pipeline that can identify patient-specific movement biomarkers from long-term naturalistic neural-kinematic recordings, advancing the viability of aDBS for gait dysfunction in PD.
Ramesh et al. (Wed,) studied this question.