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

BACKGROUND/OBJECTIVES: Voice analysis is a non-invasive tool that can capture subtle motor impairments in Multiple Sclerosis (MS). The objective of this study was to develop and validate a machine learning (ML) framework for the automated classification of MS through acoustic voice analysis. METHODS: A cohort of 300 gender-balanced participants (200 with MS and 100 healthy controls) provided sustained vocal recordings. Fifteen acoustic features were extracted. An elastic network model first identified the most relevant parameters from the development cohort (n = 200), which were then used to train five supervised ML classifiers. During the variable selection and ML model training phase, the sample was divided into an 80/20 split and cross-validation was used to minimize overfitting. The best-performing model was subsequently validated in an independent, unknown clinical cohort (n = 100). RESULTS: The Random Forest model demonstrated robust performance, which was confirmed in the independent validation (ROC AUC = 0.85 [95% CI=0.76-0.93] and balanced accuracy = 0.80), showing strong discriminative ability independent of sample prevalence. CONCLUSION: Voice analysis combined with ML presents a non-invasive, low-cost, and effective method for MS discrimination, offering significant potential as a classification support tool.