Scientists Identify Multiple Sclerosis Subtypes, Predict Progression and Treatment Response Using Artificial Intelligence

MRI-based subtypes identified by unsupervised machine learning algorithms can be used to predict disease progression and response to treatment in patients with multiple sclerosis, according to findings from scientists at University College London recently published in Nature Communications.

Researchers aimed to define patterns in brain images of patients with MS to further influence diagnosis and treatment choice beyond the four disease courses of MS currently diagnosed.

The algorithm, Subtype and Staging Inference (SuStaIn), can uncover disease subtypes with distinct temporal progression patterns, creating the ability to decipher temporal and phenotypic progression patterns.

“Our primary hypothesis was that a model based on MRI rather than solely on clinical data helps to improve a biological understanding of MS disease progression,” wrote Arman Eshaghi, MD, PhD, Queen Square Multiple Sclerosis Centre, University College London, London, UK, and co-investigators. “There were differences in the risk of disability progression, disease activity and treatment response across subtypes, which suggested that they reflected different pathobiological mechanisms relevant to the manifestations of the disease.”

Researchers used a training dataset from 6322 patients with MS and an independent cohort of 3068 patients with MS. SuStaIn was applied to the set of training scans to define MRI based subtypes and the results were corroborated with the independent cohort.

The three discovered subtypes include cortex-led, normal appearing white matter-led, and lesion-led. A further analysis on patient records was completed to see how patients with the MS subtypes responded to various treatment.

“The patterns of MRI abnormality in these subtypes provide insights into disease mechanisms and, alongside clinical phenotypes, they may aid stratification of patients for future interventional studies,” Dr Eshaghi et al wrote. —Erin McGuinness

Source: Eshaghi A, Young AL, Wijeratne PA, et al. Identifying multiple sclerosis subtypes using unsupervised machine learning and MRI data. Nat Commun. 2021;12(1):2078. Published 2021 Apr 6. doi:10.1038/s41467-021-22265-2