Brain MRI-Based Subtypes of MS Predict Disability Progression, Treatment Response

Researchers aimed to define new subtypes of MS based on objective assessments of the pathological changes seen on MRI scans, rather than on the observed changes in the clinical symptoms.

Subtypes of multiple sclerosis (MS) defined by brain magnetic resonance imaging (MRI) scans can predict disability progression and treatment response, according to study findings published in Nature Communications.

Study researchers processed brain scans from 19 datasets (randomized controlled trials: n=16) and used them to train and cross-validate a machine learning model. The scans underwent unsupervised clustering to stratify MS-associated brain patterns. External validation using 5 datasets was used to predict disability progression and treatment response.

The training and validation cohorts comprised 6322 and 3068 patients, respectively. The control and training datasets differed significantly in 13 of the 18 MRI features.

The clustering model identified 3 distinct patterns of evolution that were successfully validated in the independent cohort. The clusters were deemed cortex-led, normal-appearing white matter (NAWM)-led, and lesion-led.

The most frequently encountered subtypes were cortex-led, followed by NAWM-led. Stratified by these subtypes, patients with a lesion-led subtype had smallest baseline grey matter volume, highest baseline lesion load, expanded disability status scale (EDSS), lesion accrual, and longest disease duration (all P <.001). Study researchers observed no age or gender differences between subtypes.

Patients with a lesion-led subtype had a 30% (95% CI, 5% to 62%; P =.01) higher risk for 24-week disability progression than the cortex-led subtype among the training dataset and a 32% (95% CI, 9% to 59%; P =.004) higher risk in the validation cohort. The average relapse rate was highest among the lesion-led subtype for both the training (mean, 0.56; standard error [SE], 0.07) and validation (mean, 0.41; SE, 0.03) cohorts.

These subtypes were associated with 24-week disability progression (b, 0.04; SE, 0.01; P =.02) and stages (b, -0.06; SE, 0.02; P <.001).

The subtypes could be used to predict disease progression with a concordance index of 0.55 (standard deviation [SD], ±0.01), and this index was improved by the inclusion of clinical characteristics to 0.63 (SD, ±0.01; P <.01).

Using data from clinical trials, patients with a lesion-led subtype on active treatment exhibited slower worsening of EDSS compared with placebo among patients with secondary progressive MS (average difference, -66%; SE, ±25.6%; P =.009) or primary progressive MS (average difference, -89%; SE, ±44%; P =.04).

Although this study combined data from multiple independent cohorts, its patient stratification should be replicated with a prospective study.

These findings suggested there were 3 subtypes of MS phenotypes that associate with progression of disability and response to active treatment.

Disclosure: Some study authors declared affiliations with the biotech, pharmaceutical, and/or device companies. Please see the original article for a full list of authors’ disclosures.


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. doi:10.1038/s41467-021-22265-2