Multimodal AI model for Detecting Parkinson's Disease based on Nocturnal Multichannel Physiological Signals

In this study, we propose a multimodal hybrid AI model for detecting Parkinson's disease (PD) based on nocturnal multichannel physiological signals, specifically electrocardiography and electromyography, to address both motor and non-motor symptoms. This study focused on three biomarkers: sleep disturbances including circadian dysfunction, muscle abnormalities, and reduced heart rate variability. Data were extracted from 39 participants in a sleep health study, including 13 patients with PD and 26 healthy controls. Physiological signals were segmented into 10-second windows and divided in a ratio of 8:1:1 for training, validation, and test sets with 5-fold cross-validation. The proposed model achieved an average F1-score of 99.40%, demonstrating a superior performance in the automatic detection of PD. In conclusion, we applied a multimodal AI model to detect PD and showed its potential for diagnosing PD that can be extended to polysomnography (PSG) study. In future studies, we plan to expand the predictive model to PSG and develop a multimodal AI model that replaces Long Short-Term Memory with attention.Clinical RelevanceThis study shows possibility of automatic detecting Parkinson's disease though a multimodal AI model utilizing nocturnal multichannel physiological signals and extends polysomnography research in neurological studies.