Gait Abnormality Detection Using Deep Convolution Network

Introduction

Automatic diagnostic systems for pathological gait based on machine learning (ML) techniques have become a popular approach in rehabilitation centers and clinics (Khokhlova et al., 2019; Rueangsirarak et al., 2018). ML-based automatic gait assessment techniques have surpassed all other classical approaches with its quantitative assessment, preciseness in prediction, and effectiveness to deal with high dimensional data (Figueiredo et al., 2018). Examination of the progress of gait treatment using such automated systems is vital for certain neuromusculoskeletal diseases (Papageorgiou et al., 2019).

Periodic assessment of gait pattern using these systems helps the clinicians to prescribe patient-specific intervention and plan future treatment (Figueiredo et al., 2018). Qualitative analysis of gait pattern is highly laborious and prone to the experience level of the doctors for precise assessment (Rueangsirarak et al., 2018). Quantitative analysis based on statistical methods or mathematical transform often fails to model complex nonlinear relationship of gait data (Figueiredo et al., 2018). Automatic diagnostic systems based on ML algorithms detect gait abnormality by classifying normal and pathological gait using some salient features. In literature, different supervised classification models have been used for gait diagnosis (De Laet et al., 2017; Zhang et al., 2009), out of which support vector machine (SVM) was reported to be the best (Figueiredo et al., 2018). Recently, Convolutional Neural Network (CNN) model has gained popularity due to its effectiveness to deal with high dimensional data (Castro et al., 2017; S. S. Lee et al., 2019). But, the investigation of the usability of this model to diagnose human gait is still in its infant state.

Most of the existing gait diagnosis systems contain highly expensive sensors which make them non-affordable for most of the clinics, especially in the developing countries. An affordable gait diagnosis system is an urgent need for the modern society. Low-cost Microsoft Kinect sensor, demonstrated to be worthy for gait diagnosis (Bei et al., 2018; Khokhlova et al., 2019) due to its portability, affordability, and unobtrusive sensing property, seems to be promising gait diagnosis.

The rest of this chapter is organized as follows. In section 2, relevant state-of-the-art literature is provided. In section 3, data processing, experimental setup and proposed CNN model for the classification of human gait as Normal and Abnormal is provided. In section 4, the results and comparison with other existing models are presented. Finally, this chapter concluded with future research directions in section 5.