Analyzing multiple-sclerosis progression: stage-specific biomarker insights via explainable machine learning.

Abstract

Background: Multiple Sclerosis (MS) is a chronic autoimmune disease where early diagnosis from Clinically Isolated Syndrome (CIS) remains challenging.

Methods: This study investigates stage-specific biomarkers for CIS-to-MS conversion using explainable machine learning on a 10-year prospective dataset of 273 CIS patients, stratified by EDSS scores (1, 2, 3).

Results: Following data preprocessing and 10-fold cross-validation, Shapley analysis identified clinical, MRI, demographic, and environmental variables. Models achieved high accuracy (EDSS = 1: 89.5% via KNN; EDSS = 2/3: 100% via SVM/Ensemble). Periventricular MRI lesions and oligoclonal bands were primary predictors across all stages. Spinal cord lesions became decisive at EDSS = 3, while motor symptoms were critical for early diagnosis. Lower education and lack of breastfeeding increased MS risk; varicella history showed positive correlation.

Conclusion: These AI models effectively identify stage-specific biomarkers, revealing the dynamic importance of MRI findings. The influence of psychosocial and environmental factors underscores a multidisciplinary approach for MS management and early diagnosis.