Explainable artificial intelligence to diagnose early Parkinson's disease via voice analysis

Area of Science:

• Neuroscience
• Biomedical Engineering
• Artificial Intelligence

Background:

• Parkinson's disease (PD) is a neurodegenerative disorder impacting motor control, necessitating early diagnosis for effective management.
• Current diagnostic methods for PD can be time-consuming and costly, highlighting the need for accessible alternatives.

Purpose of the Study:

• To develop and validate an AI-driven approach for the early detection of Parkinson's disease using voice analysis.
• To assess the performance of a hybrid AI model in identifying PD from vocal biomarkers.

Main Methods:

• A hybrid machine learning model integrating Convolutional Neural Networks (CNN), Recurrent Neural Networks (RNN), Multiple Kernel Learning (MKL), and Multilayer Perceptron (MLP) was employed.
• Acoustic voice features, including Mel-Frequency Cepstral Coefficients (MFCCs), jitter, and shimmer, were extracted and analyzed from 81 voice recordings.
• SHapley Additive exPlanations (SHAP) were utilized for model interpretability, identifying key vocal features indicative of PD.

Main Results:

• The hybrid AI model demonstrated high diagnostic performance, achieving 91.11% accuracy, 92.50% recall, 89.84% precision, and a 91.13% F1 score.
• The Area Under the Curve (AUC) reached 0.9125, indicating strong discriminative power.
• SHAP analysis identified crucial acoustic features contributing to PD detection, enhancing model transparency.

Conclusions:

• AI-powered voice analysis presents a promising, non-invasive, and cost-effective method for early Parkinson's disease detection.
• The developed AI model and probability scoring system can aid clinicians and patients in monitoring disease progression and facilitating personalized treatment strategies.
• Vocal biomarkers, analyzed through advanced AI techniques, offer a valuable tool for improving PD diagnosis and management.