Automatic upper-extremity Brunnstrom Classification for Stroke Survivors with a Minimum Number of Tasks.

Abstract

Motor function evaluation plays an important role in post-stroke rehabilitation. However, the traditional evaluation is subjective and laborious, which may bring a heavy burden to both physicians and stroke survivors. Therefore, an automatic and objective rehabilitation evaluation is needed to minimize the burden of physician, so as to achieve a simplified and objective evaluation process. The main purpose of this study is to investigate the minimum number of tasks for upper-extremity actions in objective assessment of stroke survivors with a Brunnstrom stage (BS) based on wearable sensing device, which can achieve a satisfactory result to reduce the burden of stroke survivors. In this study, we employed 20 stroke survivors and 7 healthy participants, performing three types of daily living activities (drinking, teeth brushing, face washing). The acceleration, angular velocity and surface Electromyography signals on five parts of the forearm were simultaneously acquired. Then, we compared the effects of each action combination under multiple classifiers. The results show that the use of a single action can achieve competitive results compared with multiple action combination classifications, and the use of K nearest neighbor (KNN) algorithm for the average recognition accuracy of face washing action shows better performance, with the highest accuracy reaching 85.65±6.21% (mean ± standard error), 23 of the 27 subjects were accurately classified. These findings indicate that the predominant qualitative assessment after stroke can be supplemented by corresponding quantitative solutions, and that stroke rehabilitation can be automated with less professional therapist involvement.