Machine learning finds three MRI-based subtypes of multiple sclerosis

Multiple sclerosis is usually described in four clinical phenotypes — clinically isolated syndrome, relapsing-remitting, secondary progressive and primary progressive. Those labels describe how the disease behaves over time, not what is happening in the brain, and the boundaries between them are blurred enough to make treatment decisions harder than they should be.

In this Nature Communications study, first-authored by Dr Arman Eshaghi, unsupervised machine learning was applied to brain MRI to look for groups defined by pathology rather than by clinical history. The model was trained on 6,322 people with MS and validated in an independent cohort of 3,068. Grouping people by which abnormality appeared earliest produced three data-driven subtypes: cortex-led, normal-appearing white matter-led, and lesion-led.

The subtypes carried real prognostic weight. People with the lesion-led subtype had the highest risk of confirmed disability progression and the highest relapse rate — and, in selected clinical trials, showed a positive treatment response. That combination is the point: an MRI-based grouping that predicts both how the disease will progress and who is likely to benefit from treatment is exactly the kind of tool needed to stratify participants in future interventional trials. This work is a direct forerunner of the MS-PINPOINT programme.

Read the paper: Identifying multiple sclerosis subtypes using unsupervised machine learning and MRI data.