Abstract
Hemiplegia is a symptom that is caused by reduced sensory and motor ability on one side of the body due to stroke-related neural defects. Muscular weakness and abnormal sensation that is induced by hemiplegia usually lead to motor impairments, such as difficulty in controlling the trunk, unstable balance, and poor walking ability. Therefore, most hemiplegia patients show defective and asymmetric gait pattern. The purpose of this study is to distinguish hemiplegic gait by extracting simple characteristics of acceleration signals that are caused by asymmetry during walking using a wearable system. The devised wearable system was equipped with a three-axis accelerometer and a three-axis gyroscope. We selected 165 candidate features without step detection. A random forest algorithm was used for the classification, and the forward search algorithm was also used for optimal feature selection. The developed system and algorithms were verified clinically in 15 normal subjects and 20 hemiplegia patients that were undergoing stroke treatment, and 26 subject's data was used for training, including validation, and nine subject's data used for test. As a result of test set, the accuracy, sensitivity, specificity and positive predictive value were 100.0%, with the two classification attributes of standard deviation of points perpendicular to the axis of line of identity of Poincaré plot of angular velocity around vertical axis and kurtosis of frequency of angular velocity around longitudinal axis.