COMPUTATIONAL MODELLING OF NEUROINFLAMMATION IN MULTIPLE SCLEROSIS: MACHINE LEARNING APPROACHES TO DISEASE ACTIVITY PREDICTION AND TREATMENT RESPONSE

Abstract

Multiple sclerosis (MS) is the most common non-traumatic neurological disability cause in young adults, affecting approximately 2.8 million individuals globally. The relapsing-remitting course characteristic of most patients creates significant prognostic uncertainty and challenges for treatment selection — between 15 currently approved disease-modifying therapies with substantially different efficacy and safety profiles. Machine learning applied to clinical, MRI, and biomarker data has demonstrated significant potential for predicting disease activity, treatment response, and long-term disability accumulation, potentially enabling precision MS medicine. This review analyses 33 studies of machine learning in MS disease activity prediction and treatment response modelling published between 2017 and 2024. Predictive models for 2-year relapse risk achieve AUC of 0.72–0.88, while models predicting response to specific disease-modifying therapies achieve AUC of 0.76–0.91. MRI-based deep learning models for lesion segmentation demonstrate Dice similarity coefficients of 0.82– 0.94, approaching human expert performance. Implications for MS management in Uzbekistan, where diagnosis is frequently delayed and disease-modifying therapy access is limited, are discussed.