Investigating the Effects of Long-Term Contractions on Myoelectric Recognition of Wrist Movements from Stroke Patients

Abstract

In robotic rehabilitation, the classification of motion intents and detection of fatigue from surface electromyography (sEMG) are important to guarantee safety during the rehabilitation process. The time-varying characteristics of sEMG can induce errors in related applications such as force/torque estimation, detection of muscle fatigue, and pattern recognition. We investigated the effects of long-term wrist contractions on the classification accuracy of stroke patients in fatigue. Seven stroke patients participated to repeatedly perform sessions of four isometric wrist movements, namely, flexion, extension, radial deviation, and ulnar deviation in different sessions until exhaustion over 4 days. Each movement was successively performed by 60 s with 30 s of rest. To avoid excessive muscle fatigue, subjects were asked to perform each movement at 20% of the maximum voluntary contraction. We classified the four types of wrist movements using an artificial neural network and investigated variations of sEMG features in fatigue. The results showed that not only the classification accuracy but also the manifestation of muscle fatigue from sEMG remained consistent during long-term contractions in fatigue. The average classification accuracy for all patients was 0.91 ± 0.07 without significant difference between sessions.