Multimodal Physiological Signal for the Diagnosis of Parkinson's Disease

Parkinson's disease (PD) is the second most prevalent neurodegenerative disorder, characterized by a complex etiology and diverse clinical presentations. The intelligent diagnosis of this condition poses a significant challenge. Despite abnormal electrocardiogram (ECG), respiratory (RSP), and pulse signals in PD, which are closely linked to clinical symptoms, there remains a dearth of intelligent diagnostic models centered on multimodal physiological signals. In this study, we gathered multimodal physiological data from 55 PD patients and 30 healthy individuals, comprising ECG, photoplethysmogram (PPG), and RSP. Subsequently, we proposed a random forest classification model employing averaging probabilities, achieving an 89.41% accuracy in distinguishing between PD patients and healthy controls. Moreover, a feature ranking analysis was conducted on the time-domain features across the three modalities. This analysis unveiled the significance of time-domain features from diverse physiological signals in identifying PD, leading to optimization of the classification model and ultimately attaining a classification accuracy of 90.59%. In addition, the pulse and breathing characteristics in the open eyes state contribute more to disease identification, while the closed eyes state is mainly based on ECG characteristics. This research underscores the potential of machine learning techniques in amalgamating multimodal physiological data to enhance the diagnostic precision of PD, offering an objective and efficient clinical diagnostic approach.