Audio-Based Neural Network to Classify Patients With Early Parkinson’s Disease

Parkinson’s disease is a progressive neurodegenerative disorder, and identifying patients at the premotor stage enables early intervention and improved treatment outcomes. Dysarthria affects over 90% of Parkinson’s patients, making speech a valuable biomarker. In this study, we propose an end-to-end deep learning model to detect early-stage Parkinson’s disease based on speech signals. The model was trained and evaluated using recordings from 131 early-stage PD patients and 42 healthy controls, including sustained vowels (e.g., /A/, /O/) and repetitive syllables (e.g., /pa/, /ta/). Experimental results demonstrate that our model outperforms various deep learning and ensemble learning classifiers in terms of detection accuracy and F1-score, Among indicators, ACC reached 0.78, and F1-score reached 0.831. Furthermore, we explore the temporal dynamics of speech sequences to reveal their correlation with disease progression.