Parkinson's Disease Gait Evaluation Leveraging Wearable Insoles and Deep Learning Approach

Abstract

Gait evaluation is important for apprehension and management of different neurocognitive disorders (NCD). The gait events are changing with the age factor and this variability is being incorrectly linked with people with NCD. So, there is a high need to analyze gait events correctly. The gait analysis is mostly performed on temporal and spectral feature extraction in which there is a high rate of missing important features. Apart from this, monitoring and quantification of Parkinson's disease patients raise many therapeutic challenges in terms of severity analysis of motor symptoms i.e. freezing of gait (FOG), bradykinesia and continuous remote monitoring of patients. The objective of this study is to use a smart insole dataset for the assessment of computational techniques focusing on gait evaluation. The objective of this research study is to use continuous wavelet transform to convert time series signals into an images instead of using more traditional techniques for dealing with time series based on e.g. recurrent architectures. The results evidence that the proposed system works efficiently with the accuracy of 96.5% in gait variability analyzing three cohorts i.e. adults, elderly, and patients with Parkinson's disease (PwPD) and 91% for analyzing the gait symptoms in different severity stages of PD patients.