Gait based Parkinson’s disease diagnosis and severity rating using multi-class support vector machine

Abstract

To diagnose Parkinson’s disease (PD) in a clinical setting, generally clinicians utilize several clinical manifestations such as motor impairments and non-motor symptoms and predict the severity level based on unified Parkinson’s disease rating scale (UPDRS). Such a clinical evaluation highly depends on the expertise and experience of the clinicians and results in variation in assessment among clinicians. Hence, to assist the clinicians to diagnose the PD and rate the severity level, we present a gait classification based decision support system using multi-class support vector machine (MCSVM). As the gait alterations are the initial manifestations of PD, we utilize the publicly available vertical ground reaction force (VGRF) dataset and perform the kinematic analysis to extract the spatiotemporal features. To identify the prominent gait biomarkers, this work utilizes a correlation based feature selection approach and employs multi-regression approach to normalize the gait time series data. Moreover, transforming the multi-class classification problem into multiple binary classification problem using one-versus-one (OVO) strategy, the proposed PD severity rating framework tests the performance of four SVM kernel functions for three different walking tests. Experimental results highlight that the quadratic SVM classifier offers an average accuracy of 98.65% and outperforms several other state-of-the-art methods that utilized gait dataset for PD diagnosis.