Machine learning and statistical approaches to support gait analysis

Abstract

Walking performance deterioration can be caused by neurodegenerative diseases and health disorders such as low back pain (LBP). LBP is a common and costly condition that affects 8 out of 10 people at some point in their lives. The costs associated with LBP lie within the management of the condition and the social impact that comes with it. Existing technologies can be expensive or unsuitable to be used at home remotely. This research aims to investigate machine learning and statistical approaches to support gait analysis using alternative low-cost smart mobile technologies.

This thesis documents the research, development and implementation of a novel gait tele-monitoring and tele-assessment system, iterGait, to identify different patterns of gait using machine learning and statistical approaches. A mobile application has been developed during this project to investigate the feasibility of using smart mobile phones for gait analysis. The mobile application has been tested using a data collection protocol designed to capture the acceleration data of patients with LBP along with matched healthy control subjects. An overall set of 37 features of gait can be extracted from a smart mobile phone accelerometer. A new feature selection method based on the minimum redundancy maximum relevance (mRMR) method has been proposed and implemented. The mRMR feature selection method has been employed along with the classification model, KStar, to successfully differentiate between patients with LBP and the healthy control group with classification accuracy of 92.5%. The context-awareness of gait analysis is investigated to successfully detect contexts such as age, gender and activities using the framework of the system proposed.

The proposed iterGait system utilises cloud technology to provide storage and processing of data captured by smart mobile phones. This low-cost system has the potential to provide a bridge between end-users to self-manage health conditions and also remotely receive intervention from care-givers and doctors.

Date of Award	Apr 2014
Original language	English
Supervisor	Huiru (Jane) Zheng (Supervisor) & Haiying Wang (Supervisor)