Artificial Intelligence in Central Nervous System Autoimmune Diseases Under Investigation

A new review has been published examining the role of artificial intelligence technologies in the diagnosis, monitoring, and prognosis assessment of central nervous system autoimmune diseases. The study, featured in the journal EMBO Molecular Medicine, highlights advancements in multiple sclerosis research while drawing attention to translational gaps in clinical applications.

The research covered conditions such as multiple sclerosis, neuromyelitis optica spectrum disorder, autoimmune encephalitis, and myelin oligodendrocyte glycoprotein antibody-associated disease. It was noted that these disorders create diagnostic and therapeutic challenges due to overlapping clinical features, flawed biomarkers, and the requirement for long-term follow-up. It was reported that artificial intelligence is being evaluated in areas such as automated neuroimaging analysis, differential diagnosis, disease classification, relapse prediction, prognostic stratification, and risk-informed care planning. Furthermore, it was stated that it is also being applied to laboratory data, electronic health records, molecular datasets, and multimodal patient information sources.

While it was determined that evidence is unevenly distributed across diseases, it was noted that the largest volume of publications is in the field of multiple sclerosis. Conversely, it was reported that studies on neuromyelitis optica spectrum disorder, autoimmune encephalitis, and myelin oligodendrocyte glycoprotein antibody-associated diseases are smaller and more heterogeneous. It was emphasized that although many studies demonstrate promising diagnostic and prognostic performance, they lack external validation and are based on retrospective datasets. The study stated that diagnostically ambiguous, seronegative, and overlapping populations are limitedly represented in current research.

It was noted that there is growing interest in advanced applications such as federated learning, digital biomarkers, multi-omics integration, foundation models, and large language models, but these approaches offer hypothesis-generating or exploratory evidence rather than proven clinical benefit. It was emphasized that no prospectively validated artificial intelligence model has been identified that would allow for comparisons between the diseases studied or guide individual treatment selection. It was stated that most current models focus on diagnosis, monitoring, and risk prediction rather than guiding therapeutic decisions.