Predicting multiple sclerosis prognosis using AI and machine learning: integrating clinical, immunological, and...

Area of Science:

• Neuroimmunology
• Computational Neuroscience
• Biostatistics

Background:

• Accurate prediction of multiple sclerosis (MS) progression is crucial for effective clinical management.
• Relapsing-remitting MS (RRMS) requires reliable methods to monitor disease advancement and disability.
• Existing methods for predicting MS progression have limitations in accuracy and timeliness.

Purpose of the Study:

• To investigate the efficacy of supervised machine learning (ML) models in predicting clinical disability (Expanded Disability Status Scale - EDSS) and radiological activity (MRI lesion changes) in RRMS patients.
• To evaluate the predictive performance of various ML classifiers using peripheral cytokine profiles and patient metadata.
• To determine if ML models can offer clinically meaningful predictions for functional and radiological progression in MS.

Main Methods:

• Peripheral cytokine profiles (IL-12, TNF-α, IFN-γ, IL-4, IL-10) and patient metadata were utilized.
• 43 machine learning classifiers were trained and evaluated.
• Models were assessed for their ability to discriminate between mild and moderate disability (EDSS thresholds) and predict new MRI lesions in 15 RRMS patients.

Main Results:

• Ensemble ML models demonstrated superior performance compared to simpler algorithms.
• For EDSS prediction, Random Forest achieved 90.1% sensitivity and 89.7% specificity; Simple Logistic Regression reached 92.6% with patient ID.
• Random Subspace classifiers excelled in predicting new MRI lesions, achieving 82.4% sensitivity and specificity.

Conclusions:

• Combining cytokine profiles with ML strategies provides accurate predictions of functional and radiological progression in RRMS.
• These predictive tools can enhance patient monitoring, therapeutic decision-making, and risk stratification.
• Further validation in prospective cohorts is necessary for clinical implementation of these ML-based predictive models.