Journal of Biomedical Engineering and Medical Devices
Open Access

Abstract

A Mobile, Smart Gait Assessment System for Asymmetry Detection Using Machine Learning-Based Classification

Sebastian Márquez J, Roozbeh Atri, Masudur R Siddiquee, Connie Leung and Ou Bai

Gait asymmetry is characterized as the dynamic differences between contralateral limbs, it has been shown to be caused by disease, age, clinical interventions and limb dominance. In this study, a mobile gait assessment system was developed for the evaluation of gait asymmetry in persons with simulated leg length discrepancy (sLLD). LLD is a disorder that affects 40-70% of the population requiring clinical intervention when the dissimilarity between limbs exceeds 3.7%. In out of clinic applications, an ambulatory gait symmetry system may be used to monitor postsurgical outcomes based on objective temporal and kinetic features. For this, a wireless gait symmetry system was designed and tested to measure ground reaction forces from insole worn pressure sensors. Thirteen metrics were extracted from a group of 9 subjects and a linear discriminant analysis performed for feature selection. Machine learning classifiers were used to differentiate between normal walking and sLLD. Applying majority voting to an Ensemble AdaBoost Tree classifier resulted in an overall accuracy of 89.9%, a false positive rate of 3.9%, and a sensitivity of 83.6%. Results indicate the wearable sensor is a viable option for out-of-clinic monitoring of asymmetry using machine learning.