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

Area of Science:

• Neurology
• Artificial Intelligence
• Biomedical Engineering

Background:

• Parkinson's disease (PD) diagnosis relies on clinical symptoms, often appearing after significant neurodegeneration.
• Current diagnostic methods lack objective biomarkers for early detection of both motor and non-motor symptoms.
• Nocturnal physiological signals offer a rich, underutilized data source for PD assessment.

Purpose of the Study:

• To develop and validate a multimodal hybrid AI model for early Parkinson's disease detection.
• To integrate electrocardiography and electromyography signals for comprehensive PD symptom analysis.
• To investigate the utility of sleep disturbances, muscle abnormalities, and heart rate variability as PD biomarkers.

Main Methods:

• Utilized nocturnal multichannel physiological signals (electrocardiography, electromyography) from 39 participants (13 PD patients, 26 controls).
• Segmented signals into 10-second windows, with data split for training, validation, and testing using 5-fold cross-validation.
• Developed a multimodal hybrid AI model focusing on sleep disturbances, muscle abnormalities, and heart rate variability.

Main Results:

• The AI model achieved a high average F1-score of 99.40% in detecting Parkinson's disease.
• Demonstrated superior performance in the automatic detection of PD using physiological signals.
• Highlighted the potential of the model for early and accurate PD diagnosis.

Conclusions:

• A multimodal AI model effectively detects Parkinson's disease using nocturnal physiological signals.
• The model shows promise for extending diagnostic capabilities to polysomnography (PSG) studies.
• Future work includes expanding the model to PSG data and incorporating attention mechanisms.