Voice analysis as a digital biomarker: A machine learning approach for automated multiple sclerosis classification.

Area of Science:

• Neurology
• Biomedical Engineering
• Data Science

Background:

• Subtle motor impairments in Multiple Sclerosis (MS) can be detected non-invasively using voice analysis.
• Acoustic voice analysis offers a potential method for early detection and classification of MS.

Purpose of the Study:

• To develop and validate a machine learning (ML) framework for automated Multiple Sclerosis (MS) classification using acoustic voice analysis.
• To assess the efficacy of ML models in discriminating between individuals with and without MS based on voice parameters.

Main Methods:

• A cohort of 300 participants (200 MS, 100 controls) provided vocal recordings; 15 acoustic features were extracted.
• An elastic network model identified key voice parameters, which trained five supervised ML classifiers.
• The best model was validated on an independent cohort of 100 individuals.

Main Results:

• The Random Forest model achieved robust performance in independent validation.
• The model demonstrated strong discriminative ability with an ROC AUC of 0.85 and balanced accuracy of 0.80.
• Performance was independent of MS sample prevalence.

Conclusions:

• Acoustic voice analysis combined with ML provides a non-invasive, cost-effective method for MS discrimination.
• This approach has significant potential as a classification support tool for Multiple Sclerosis.
• Further research can explore the integration of voice analysis into routine MS diagnostics.