Evoked resonant neural activity outperforms spectral markers in decoding sleep from the subthalamic nucleus

Abstract

Background: Deep brain stimulation is a treatment for advanced Parkinson's disease and currently tuned to target motor symptoms during daytime. Parkinson's disease is associated with multiple nocturnal symptoms such as akinesia, insomnia and sleep fragmentation which may require adjustments of stimulation during sleep for best treatment outcome. Objectives: There is a need for a robust biomarker to guide stimulation titration across sleep stages. This study aimed to investigate whether evoked resonant neural activity (ERNA) is modulated during sleep. Methods: We recorded local field potentials from the subthalamic nucleus of four Parkinson's patients with externalised electrodes while applying single stimulation pulses to investigate the effect of sleep on ERNA. Results: We found that ERNA features change with wakefulness and sleep stages, and are correlated with canonical frequency bands and heart rate. We further evaluated the performance of machine learning models in classifying non-REM sleep versus wakefulness and found that ERNA amplitude outperforms all spectral markers. Conclusions: Given the heterogeneity of spectral features during sleep, their susceptibility to movement artefacts and superior classification accuracy of models using ERNA features, this study paves the way for ERNA as a marker for automatic stimulation titration during sleep and improved patient care.