Abstract
Although high-frequency deep brain stimulation (DBS) is effective at relieving many motor symptoms of Parkinson's disease (PD), its effects on gait can be variable and unpredictable. This is due to 1) a lack of standardized and robust metrics for gait assessment in PD patients, 2) the challenges of performing a thorough evaluation of all the stimulation parameters space that can alter gait, and 3) a lack of understanding for impacts of stimulation on the neurophysiological signatures of walking. In this study, our goal was to develop a data-driven approach to identify optimal, personalized DBS stimulation parameters to improve gait in PD patients and identify the neurophysiological signature of improved gait. Local field potentials from the globus pallidus and electrocorticography from the motor cortex of three PD patients were recorded using an implanted bidirectional neural stimulator during overground walking. A walking performance index (WPI) was developed to assess gait metrics with high reliability. DBS frequency, amplitude, and pulse width on the "clinically-optimized" stimulation contact were then systemically changed to study their impacts on gait metrics and underlying neural dynamics. We developed a Gaussian Process Regressor (GPR) model to map the relationship between DBS settings and the WPI. Using this model, we identified and validated personalized DBS settings that significantly improved gait metrics. Linear mixed models were employed to identify neural spectral features associated with enhanced walking performance. We demonstrated that improved walking performance was linked to the modulation of neural activity in specific frequency bands, with reduced beta band power in the pallidum and increased alpha band pallidal-motor cortex coherence synchronization during key moments of the gait cycle. Integrating WPI and GPR to optimize DBS parameters underscores the importance of developing and understanding personalized, data-driven interventions for gait improvement in PD.