Performance Evaluation of EEG/EMG Fusion Methods for Motion Classification.

Abstract

Wearable robotic systems have shown potential to improve the lives of musculoskeletal disorder patients; however, to be used practically, they require a reliable method of control. The user needs to be able to indicate that they wish to move in a way that feels intuitive and comfortable. One proposed method for detecting motion intention is through the combined use of muscle activity, known as electromyography (EMG), and brain activity, known as electroencephalography (EEG). Other groups have developed various methods of fusing EEG/EMG signals for classification of motion intention, but a comprehensive evaluation of their performance has yet to be completed. This work evaluates EEG/EMG fusion methods during elbow flexion-extension motion while varying parameters, such as speed of motion, weight held, and muscle fatigue. Overall, the use of EEG/EMG fusion was found to not be more accurate than using just EMG alone $(86.81 \pm 3.98$%), with some fusion methods demonstrating equivalent performance to EMG $(p=1.000)$. EEG/EMG fusion was, however, demonstrated to be less sensitive to changes in motion parameters, allowing it to perform more consistently across different speed/weight combinations. The results of this work provide further justification for the use of EEG/EMG fusion for control of a wearable robotic device.