Artificial intelligence is opening a new chapter in the understanding of multiple sclerosis by revealing that the disease consists of at least two biologically distinct subtypes. In research led by UCL and Queen Square Analytics, a UCL spinout company, AI was able for the first time to identify these subtypes using a combination of a simple blood test and standard MRI brain imaging.
The study, published in Brain, focused on serum neurofilament light chain, or sNfL, a blood biomarker that reflects nerve cell damage and provides an indication of disease activity. Researchers combined sNfL measurements with MRI data showing the extent and location of brain damage. These data were analysed using SuStaIn, a machine learning model developed at UCL that infers disease subtypes and stages from complex biological data.
Two types of MS revealed
Analysis of data from 634 participants across two clinical trial cohorts revealed two distinct patterns of multiple sclerosis. One group showed high sNfL levels early in the disease, alongside early damage to the corpus callosum and rapid development of brain lesions, suggesting a more aggressive and active disease course. The second group showed brain atrophy in regions such as the limbic cortex and deep grey matter before sNfL levels increased, indicating a slower disease trajectory in which overt damage emerges later.
According to lead author Dr. Arman Eshaghi, “MS is not one disease and current subtypes fail to describe the underlying tissue changes, which we need to know to treat it.” He adds, “By using an AI model combined with a highly available blood marker with MRI, we have been able to show two clear biological patterns of MS for the first time. This will help clinicians understand where a person sits on the disease pathway and who may need closer monitoring or earlier, targeted treatment.”
Earlier intervention, better treatment
The findings address a long-standing challenge in MS care, where clinical classifications are based on symptoms and disease course rather than underlying biology. This mismatch can lead to treatments that fail to target the mechanisms driving disease progression. By identifying measurable biological subtypes, the approach offers a pathway to earlier intervention, more precise monitoring and better alignment between patients and therapies.
With more than 2.8 million people worldwide living with MS, often developing disability at a young age, the ability to predict disease progression before clinical deterioration becomes visible could have a significant impact. As large datasets and AI methods become more widely available, data driven subtyping based on blood and imaging biomarkers may redefine how MS and other neurological diseases are diagnosed and treated.
Medication effectiveness
Last year, Brazilian and French researchers developed a machine learning based tool that predicts how patients with multiple sclerosis will respond to natalizumab, a widely used but costly therapy with variable effectiveness. By combining high content cell imaging with advanced analytics, the approach supports more precise treatment selection in neurology.
The team analysed immune cell behaviour in blood samples, focusing on changes in CD8+ T cells after treatment. While responders showed expected actin remodelling and rounded cell shapes, non responders retained elongated, highly mobile cells. By extracting more than 400 morphological features and analysing over one million feature combinations, the model achieved high predictive accuracy, reaching 92 percent in the discovery cohort and 88 percent in validation.
According to the researchers, the key innovation was converting complex cell images into quantifiable data suitable for machine learning. Their findings highlight the potential to improve outcomes by avoiding ineffective treatments, reducing side effects and lowering healthcare costs. The team plans further validation in larger populations and sees broader applications for personalized medicine beyond multiple sclerosis.
