Gait Clustering Analysis in Patients after Stroke using Gait Kinematics Data

Abstract

In rehabilitation of the patients after stroke, gait types are important to know the characteristics of the patient. To know gait types, a systematic methodology for direct measurement and interpretation of gait motion are required. In this study, the patient's kinetic data were collected eight times over six months after onset using motion capture equipment. Features for gait type classification were extracted from time series gait cycle data and used for machine learning analysis. We utilized the simultaneous clustering and classification method to determine gait types that ensure classification performance. The optimal number of gait groups was four, which shows 0.1504 and 0.9142 in silhouette score and F1 score. We present a novel work to find the gait groups of patients after stroke, and showed the potential for use in the rehabilitation field.