A Novel Approach to Distinguish Parkinson's Disease Patients From Healthy Control Subjects Using Speech-Based Task Analysis.

Patients with Parkinson's disease (PD) often exhibit speech and voice impairments early in the disease course, making these characteristics potential biomarkers for diagnosis. Additionally, PD speech analysis offers a promising avenue for monitoring disease progression in response to therapeutic interventions. In this study, we propose a novel method for distinguishing PD patients from healthy controls (HCs) through the analysis of speech task recordings. Our method integrates recurrence plots (RPs) and their corresponding quantitative descriptors with established speech features, namely Mel-spectrograms and Mel-frequency Cepstral Coefficients (MFCCs). To enrich the representation of speech signals, RPs and Mel-spectrograms are further processed to extract features using a Convolutional Neural Network (CNN). The resulting feature sets are then classified with a Support Vector Machine (SVM). Experimental evaluations on the PC-GITA speech database, as well as an analogous set of PD tasks in Greek, demonstrate the effectiveness of the proposed approach, achieving classification accuracies above 90% on the examined tasks.