Multimodal Electrophysiological Signals for Machine Learning-Aided Parkinson's Disease Diagnosis

Area of Science:

• Biomedical Engineering
• Neuroscience
• Medical Diagnostics

Background:

• Parkinson's disease (PD) is a neurodegenerative disorder impacting motor and autonomic functions.
• Accurate and non-invasive diagnostic tools for PD are crucial for timely intervention.

Purpose of the Study:

• To evaluate the diagnostic performance of unimodal and multimodal biosignal analysis for Parkinson's disease.
• To identify optimal combinations of biosignals for efficient and accurate PD detection.

Main Methods:

• Recorded six synchronized biosignals: electroencephalography (EEG), electrocardiography (ECG), electromyography (EMG), respiration (Resp), photoplethysmography (PPG), and gait (Gait).
• Utilized a Random Forest classifier for both unimodal and multimodal signal classification in 25 PD patients and 25 controls.
• Performed incremental analysis to assess the contribution of each modality.

Main Results:

• Unimodal ECG achieved the highest accuracy (84%).
• Multimodal combinations showed improved performance, with three or more signals significantly enhancing classification.
• The full six-modality model achieved 95.00% accuracy, 97.14% recall, and 0.98 AUC.
• Selecting key complementary modalities maintained high performance while reducing complexity.

Conclusions:

• Multimodal biosignal analysis offers a highly accurate and efficient approach for non-invasive Parkinson's disease diagnosis.
• Optimizing signal combinations can lead to simpler, more comfortable diagnostic procedures.
• This approach provides a foundation for developing practical PD diagnostic tools.