A machine learning-based model to evaluate multiple sclerosis predictor factors with emphasis on neurophysiological indices of physical activity

Abstract

AimsThe aim of this study was to establish a model for prediction and early diagnosis of multiple sclerosis (MS) based on motion-dependent neurophysiological variables. Main methodsThe statistical population included 110 volunteers with and without MS in Mazandaran province, Iran. Based on the information provided by the subjects, they were assigned into the following groups; MS and control groups, and based on disease model they were further divided into relapsing-remitting (RR), progressive-relapsing (PR) and control groups, and according to the activity levels they were assigned into active MS, sedentary MS, active control and sedentary control groups. The Support Vector Machine (SVM) method was used to ensure separation and prediction accuracy. All calculations were performed using MATLAB software (version 2016). Key findings99.1% separation accuracy and 90% prediction accuracy were observed in non-kinematic data, while in kinematic and electromyography (EMG) data, this was 66% for separation accuracy and 65% regarding prediction accuracy. Among the measured variables, static balance and strength had the greatest effect on prediction results. SignificanceUsing SVM technique and incorporating early symptoms of MS, we were able to achieve a high precision in predicting MS among the participants. Based on SVM, we achieved a considerably higher prediction accuracy extrapolated from non-kinematic dataset compared to kinematic and EMG datasets. Therefore, this study has opened up a great avenue towards predicting MS based on clinical parameters which could provide the clinicians with information regarding progression of the disease well in advance helping in opting for the best treatment strategies.