Comparing Online Performance of EMG Pattern Recognition with and Without Joint Movements

Our previous study suggested that the subcutaneous muscle displacement caused by joint movements might alter muscle activation patterns and thus affect the classification performance. To further analyze the effect of joint movements on the online performance of Electromyography (EMG) Pattern Recognition (PR), this study assessed online classification performance with and without joint movements. EMG signals were recorded from the dominant forearm of 10 able-bodied subjects under two motion scenarios: Hand and Wrist Joints Unconstrained (HAWJU) and Constrained (HAWJC). Sixth-order autoregressive coefficients and four time-domain features were extracted from EMG signals. Linear Discriminant Analysis (LDA) models were trained to perform an online performance evaluation of the limb motions. The experimental results showed that the four online performance metrics: Motion Selection Time (MST), Motion Completion Time (MCT), Motion Completion Rate (MCR), and Online Classification Accuracy (ONCA) were 0.35 s, 1.44 s, 97.40%, and 82.61% for HAWJU and 0.37 s, 1.47 s, 89.70%, and 73.57% for HAWJC, respectively. The outcomes of this study indicated that subcutaneous muscle displacement due to joint movements has a positive effect on online classification performance. The absence of joint movements may be a physiological factor contributing to the poor online performance of the EMG-PR of transradial amputees. This study can provide a new perspective for improving the online performance of EMG-PR for transradial amputees.