Machine-learning method accurately classifies patients with MS

A combination of different machine-learning principles applied to functional MRI (fMRI) scans accurately classified patients with MS by clinical phenotype, and distinguished them from healthy controls. Distinct sub-network abnormalities contributed to accurate phenotype classification.

Graph theoretical analysis helps to gain insight into functional reorganisation in MS. Italian researchers developed advanced machine-learning methods to analyse data on resting-state functional connectivity and classify MS patients according to disease phenotype [1]. They obtained fMRI scans from 46 healthy controls and 113 MS patients (62 with relapsing-remitting MS and 51 with progressive MS). By way of dominant set clustering, functional connectivity matrices were grouped into patients with similar network configurations. Disease phenotypes were classified using linear support vector machines.

This approach helped to distinguish relapsing-remitting MS patients from healthy controls with an accuracy rate of 72.5%. A sensitivity analysis revealed the following key features that differentiated relapsing-remitting MS as well as progressive MS patients from healthy controls: increased connectivity within the basal ganglia sub-network and decreased functional connectivity within the temporal sub-network. Decreased functional connectivity within the occipital and parietal sub-networks contributed to differentiate progressive MS patients from healthy controls. Altered thalamic and frontal resting-state functional connectivity occurred in all phenotypes and may be a hallmark of MS. The involvement of occipitotemporal subnetworks in relapsing-remitting MS patients may be secondary to damage of associative sensory regions. The involvement of the parietal regions in progressive MS suggests a spreading of damage to high-order, associative regions, leading to impaired network integration.

In another very recent study, machine learning applied to brain MRI scans from 6,322 MS patients resulted in the definition of 3 MS subtypes: cortex-led, normal-appearing white matter-led, and lesion-led [2]. The lesion-led subtype had the highest risk of confirmed disability progression and the highest relapse rate, but also predicted positive treatment response in clinical trials.

1. Rocca MA, et al. Classifying and characterizing multiple sclerosis disease phenotypes with functional connectivity and machine learning. OPR-112, EAN 2021 Virtual Congress, 19–22 June.
2. Eshaghi A, et al. Nat Commun. 2021;12(1):2078.

 

Copyright ©2021 Medicom Medical Publishers



Posted on 18 August 202118 August 2021