Wearable-Enabled Screening of RBD Through HRV-Based Machine Learning

Rapid Eye Movement Sleep Behavior Disorder (RBD) is one of the strongest prodromal markers of neurodegenerative diseases such as Parkinson's disease. However, its current diagnosis relies on overnight video-polysomnography, which is costly, intrusive, and unsuitable for population-level screening. Heart Rate Variability (HRV), which reflects autonomic regulation, offers a promising non-invasive alternative based on wearable ECG sensing, enabling scalable detection in pervasive healthcare environments.This PhD research investigates the feasibility of HRV-based automatic RBD screening through Machine Learning (ML). Four polysomnography datasets (184 subjects) were analyzed to evaluate the robustness of HRV features under realistic conditions. A comprehensive set of 528 time, frequency, and nonlinear-domain attributes was extracted from FullNight and SleepOnly ECG segments, and multiple ML pipelines were trained and optimized for two binary tasks: healthy vs. RBD, and non-RBD (healthy+OSAS) vs. RBD.Results show that HRV biomarkers enable accurate discrimination, achieving F1-scores above 90% in healthy-RBD classification and above 80% even when confounding sleep disorders are included. Moreover, stage-agnostic FullNight extraction improved generalization across heterogeneous datasets, supporting wearable deployment without expert scoring.These findings highlight the potential of autonomic-based pervasive screening for prodromal neurodegeneration and motivate further validation on real-world wearable data toward home-based early diagnosis.