Machine Learning-Based Detection of Parkinson's Disease From Arabic Speech: A Cross-Linguistic Validation Study

Area of Science:

• Neurology
• Computational Linguistics
• Machine Learning

Background:

• Early detection of Parkinson's disease (PD) is crucial for effective management.
• A significant gap exists in validated speech analysis tools for Arabic-speaking populations.
• Previous machine learning (ML) approaches for PD detection used limited features, potentially introducing bias.

Purpose of the Study:

• To develop and validate an optimal Parkinson's disease prediction pipeline using speech analysis for Arabic speakers.
• To create the first Arabic PD speech dataset and test the generalizability of the methodology on a Spanish cohort.
• To identify vocal biomarkers for PD through advanced feature extraction and explainability techniques.

Main Methods:

• Developed a novel PD prediction pipeline by evaluating multiple ML classifiers and feature extraction methods.
• Created the first Arabic PD speech dataset (40 subjects) and validated on a Spanish cohort (100 subjects).
• Utilized traditional, audio-to-text (Whisper model), and deep voice features, followed by feature selection, dimensionality reduction, and twelve classifiers with cross-validation. SHAP analysis identified feature importance.

Main Results:

• Linear Discriminant Analysis achieved 90% accuracy, precision, recall, and F1-score on the Arabic dataset using leave-one-out cross-validation.
• Linear Support Vector Classification demonstrated strong performance with 87.7% precision and 87.5% recall.
• The methodology achieved 83% accuracy on the independent Spanish dataset, confirming cross-linguistic generalizability. Audio-to-text features provided contextual insights, while traditional features captured acoustic variations.

Conclusions:

• Established the first validated Arabic speech classification system for Parkinson's disease detection.
• Demonstrated the universal applicability and cross-linguistic generalizability of the developed speech analysis methodology.
• Laid the groundwork for global, accessible, and early screening of Parkinson's disease through speech analysis.