Artificial intelligence-based monitoring of multiple sclerosis: A comprehensive survey

Multiple Sclerosis (MS) is a chronic, highly heterogeneous neurological disease whose fluctuating course is poorly captured by infrequent clinic visits and subjective scales such as the Expanded Disability Status Scale (EDSS). Advances in artificial intelligence (AI), inertial sensing, smartphones, and computer vision now enable continuous, objective, and remote monitoring of motor, cognitive, and behavioural function. This paper surveys 74 peer-reviewed studies, published up to 2025, that apply AI and machine learning (ML) to MS monitoring, organizing them by data-acquisition modality: smartphone-based assessment, wearable sensors, multimodal platforms, video and computer vision, digital screen interaction, and longitudinal clinical and registry-based modeling. For each modality we summarize the monitored tasks, signals, models, ground-truth measures, and validation strategies, and synthesize the characteristic strengths and limitations. Across the extracted review dataset, Random Forest is the most frequently reported model and EDSS the dominant ground-truth measure, while deep-learning and multimodal approaches are growing rapidly after 2021. A central finding is that external validation is critically underused: among studies with extractable validation-strategy information (n = 74), only 5.41% report external validation against an independent cohort. We distil cross-study trends, research gaps, and concrete future directions toward clinically validated, interpretable, multimodal, and privacy-aware MS monitoring systems.