Artificial Intelligence Advances CNS Autoimmune Disease Care

Artificial intelligence is increasingly being studied as a way to improve the diagnosis, monitoring, and prediction of future health outcomes in autoimmune diseases of the central nervous system (CNS), according to a new review of current research.

The review looked at how artificial intelligence is being used in multiple sclerosis, neuromyelitis optica spectrum disorder, autoimmune encephalitis, and myelin oligodendrocyte glycoprotein antibody-associated disease.

These conditions can be difficult to diagnose and treat because they can cause similar symptoms, available biological markers are not always reliable, and patients need long-term follow-up.

Researchers reported that artificial intelligence has been studied for several clinical tasks. These include automatically analysing brain scans, helping doctors tell diseases apart, classifying diseases, predicting relapses, estimating likely health outcomes, and planning care based on a patient's risks.

The technology is also being used to study laboratory test results, electronic health records, molecular data, and different types of patient information combined together.

Artificial Intelligence Research Is Strongest in Multiple Sclerosis

The amount of evidence varied across the diseases. Multiple sclerosis had the largest body of published research, especially on brain imaging. These applications include identifying and measuring lesions, tracking disease activity, measuring brain volume, and predicting disability.

Some imaging tools have advanced to testing across multiple medical centres and in studies that assess their use in clinical practice over time.

In comparison, studies of neuromyelitis optica spectrum disorder, autoimmune encephalitis, and myelin oligodendrocyte glycoprotein antibody-associated disease were generally smaller and more varied.

Although several studies showed promising results for diagnosis and predicting health outcomes, many had not been tested on independent groups of patients and relied on data collected in the past.

The review highlighted that current research includes too few patients whose diagnoses are uncertain, who test negative for the relevant antibodies, or who have features that overlap between different diseases.

Researchers also noted growing interest in more advanced methods, including federated learning, digital biomarkers, combining different types of molecular data, foundation models, and large language models.

However, these methods currently offer early findings that help researchers develop and explore ideas, rather than proof that they improve clinical care.

Key Challenges in Bringing Artificial Intelligence into Clinical Care

Despite progress, important gaps remain between research and everyday medical practice. The review found no artificial intelligence model that had been tested in advance in real clinical settings and shown to reliably compare treatment effects or select the best treatment for individual patients across the diseases studied.

Most existing models focus on diagnosis, monitoring, and predicting risks rather than helping doctors decide which treatment to use.

The review stressed the need for larger datasets from multiple medical centres, stronger testing on independent patient groups, studies conducted in real clinical settings, and assessments of how these tools affect medical decisions and patient outcomes.

According to the researchers, near-term progress is most likely to come from combining different types of patient data, making brain imaging analysis more consistent, and improving the assessment of patient risks. Current evidence does not yet support artificial intelligence making diagnoses or choosing treatments on its own.

Reference

Cao Z et al. Artificial intelligence in central nervous system autoimmune diseases: from clinical decision support to mechanistic discovery and translation. EMBO Mol Med. 2026; DOI:10.1038/s44321-026-00527-0.