LAUSR

Abstract The development of a reliable but cost-effective Parkinson’s disease (PD) detection and severity stage classification system is one of the important aspects of medical application. Hence, this paper has provided a new cost-effective approach based on time-varying singular value decomposition method for the analysis of gait signals. The main idea is to separate different components of the signal, select the most relevant components which are used to quantify inter-limb deviation in singular value space. For this purpose, gait data are rearranged in two trajectory matrixes and a new symmetric feature revealing transient characteristic has been extracted. The proposed method has been evaluated and compared across three gait datasets recorded during different walking tasks. The experimental results have demonstrated that the use of new features outperforms most of the previous methods. Average accuracy rates of 97.22% and 95.59% have been obtained across all datasets for PD detection and stage classification, respectively.