Cluster-Based classification of Multiple Sclérosis Patients: "Identifiying Gait Biomarkers"

Abstract

Gait dysfunction is a common and disabling symptom of multiple sclerosis (MS). Classic clinical evaluations often lack sensitivity to detect early or subtle impairments in MS and this study is a method to complement for this by combining high-resolution spatiotemporal gait data from GAITRite® electronic walkway with unsupervised machine learning (ML) algorithms for the detection of PwMS gait phenotypes. Method Secondary structure prediction Secondary structure prediction was conducted using the Principal Component Analysis (PCA) for dimensionality reduction and for the orthogonal feature projection. Following the dimensionality reduction, the KMeans clustering technique was used to detect gait clusters. 4 clusters on the basis of differences in walking speed and lateral stability were identified. Such variations reflect underlying biomechanical and neural differences. including spasticity, ataxia, and nearly-normal act trends. This information provides clinical meaningful functional information about gait impairment not otherwise represented through common metrics such as the EDSS or Timed Walk tests. The results encourage the use of unsupervised clustering as a technique for MS motor-impairment classification, helping to guide tailored rehabilitation therapy.