Deep Learning as a Tool in Current MRI-Based Approaches for Predicting Clinical Disability Progression in Multiple Sclerosis

Multiple sclerosis (MS) is a chronic neurological disorder and one of the leading causes of long-term disability in young adults. Despite therapeutic advances, predicting disability progression remains challenging. Magnetic resonance imaging (MRI) plays a central role in disease monitoring; however, conventional prognostic models show limited predictive accuracy. Deep learning (DL) has emerged as a promising approach for extracting complex imaging patterns and improving outcome prediction. This systematic review, conducted in accordance with PRISMA guidelines, evaluated the potential of MRI-based deep learning models for predicting clinical disability progression in MS. A comprehensive literature search was performed in PubMed, Google Scholar, Cochrane Library, and IEEE Xplore using PICO criteria and relevant MeSH terms. Studies applying deep learning techniques to MRI data for disability progression prediction were included, and methodological quality was assessed using the Prediction Model Risk of Bias Assessment Tool (PROBAST). Of 226 identified records, seven studies met the inclusion criteria, with most demonstrating low risk of bias. Predictive performance ranged from moderate to strong, with area under the curve (AUC) values between 0.69 and 0.84. Regression models achieved root mean square error (RMSE) values as low as 1.33, while survival-based approaches reported concordance indices up to 0.72. Longitudinal deep learning models generally outperformed single timepoint approaches. Overall, MRI-based deep learning models show strong potential for predicting disability progression in MS and may support personalized disease monitoring and clinical decision-making.