Identifying multiple sclerosis subtypes using unsupervised machine learning and MRI data

Abstract

Multiple sclerosis (MS) can be divided into four phenotypes based on clinical evolution. The pathophysiological boundaries of these phenotypes are unclear, limiting treatment stratification. Machine learning can identify groups with similar features using multidimensional data. Here, to classify MS subtypes based on pathological features, we apply unsupervised machine learning to brain MRI scans acquired in previously published studies. We use a training dataset from 6322 MS patients to define MRI-based subtypes and an independent cohort of 3068 patients for validation. Based on the earliest abnormalities, we define MS subtypes as cortex-led, normal-appearing white matter-led, and lesion-led. People with the lesion-led subtype have the highest risk of confirmed disability progression (CDP) and the highest relapse rate. People with the lesion-led MS subtype show positive treatment response in selected clinical trials. Our findings suggest that MRI-based subtypes predict MS disability progression and response to treatment and may be used to define groups of patients in interventional trials.

Phenotypic and genetic associations of quantitative magnetic susceptibility in UK Biobank brain imaging

“…Thus, the dissemination of our pipeline is important to the translational impact of the UK Biobank QSM resource. For example, in order to relate a new subject to a classifier trained using UK Biobank data 81 or to a nomogram derived from UK Biobank 82 , it will be important to use the same QSM processing pipeline. To date, two ongoing COVID-19 brain imaging studies have already adopted our QSM pipeline (C-MORE/PHOSP and COVID-CNS) 83 to process their brain swMRI data.…”

“…Dissemination of our pipeline will thus be crucial for harmonisation of our IDPs with data acquired in novel settings, such as clinical scanners. This will, for example, enable stratification of patients using classifiers 79 or nomograms 80 derived from UK Biobank data. To date, two ongoing COVID-19 brain imaging studies have already adopted our QSM pipeline (C-MORE/PHOSP and COVID-CNS) 81 to process their brain swMRI data.…”

pySuStaIn: a Python implementation of the Subtype and Stage Inference algorithm

“…This is valuable because many progressive diseases are heterogeneous in nature and can naturally be described by a set of distinct subtypes [1], [2],[3], [4]. SuStaIn has been applied to a number of neurodegenerative diseases, including Alzheimer’s disease [5][6], frontotemporal dementia [5] and multiple sclerosis (MS) [7]. It has also been being applied to progressive lung disease [8].…”

“…Importantly, the study also showed that only one of the identified subtypes showed a significant treatment response in randomised controlled trials. This study used the z-score likelihood [7].…”

Ordinal SuStaIn: Subtype and Stage Inference for Clinical Scores, Visual Ratings, and Other Ordinal Data

“…SuStaIn simultaneously clusters individuals into subgroups and characterises the trajectory that best defines each subgroup, thus capturing heterogeneity in both disease subtype and disease stage. The SuStaIn algorithm has been applied in a range of conditions including Alzheimer's disease (Young et al, 2018;Aksman et al, 2020;Garcia et al, 2020;Vogel et al, 2021), frontotemporal dementia (Young et al, 2018;Young et al, 2020a), Multiple Sclerosis (Eshaghi et al, 2020) and Chronic Obstructive Pulmonary disease (Young et al, 2020b). From a mathematical perspective any disease progression model can be used in combination with SuStaIn, but in practice some disease progression models may be unfeasibly computationally intensive.…”