A convolutional neural network to identify motor units from high-density surface electromyography signals in real time

Abstract

Objectives. This paper aims to investigate the feasibility and the validity of applying deep convolutional neural networks (CNN) to identify motor unit (MU) spike trains and estimate the neural drive to muscles from high-density electromyography (HD-EMG) signals in real time. Two distinct deep CNNs are compared with the convolution kernel compensation (CKC) algorithm using simulated and experimentally recorded signals. The effects of window size and step size of the input HD-EMG signals are also investigated. Approach. The MU spike trains were first identified with the CKC algorithm. The HD-EMG signals and spike trains were used to train the deep CNN. Then, the deep CNN decomposed the HD-EMG signals into MU discharge times in real time. Two CNN approaches are compared with the CKC: (a) multiple single-output deep CNN (SO-DCNN) with one MU decomposed per network, and (b) one multiple-output deep CNN (MO-DCNN) to decompose all MUs (up to 23) with one network. Main results. The MO-DCNN outperformed the SO-DCNN in terms of training time (3.2-21.4 s epoch −1 vs 6.5-47.8 s epoch −1 , respectively) and prediction time (0.04 vs 0.27 s sample −1 , respectively). The optimal window size and step size for MO-DCNN were 120 and 20 data points, respectively. It results in sensitivity of 98% and 85% with simulated and experimentally recorded HD-EMG signals, respectively.There is a high cross-correlation coefficient between the neural drive estimated with CKC and that estimated with MO-DCNN (range of r-value across conditions: 0.88-0.95). Significance. We demonstrate the feasibility and the validity of using deep CNN to accurately identify MU activity from HD-EMG with a latency lower than 80 ms, which falls within the lower bound of the human electromechanical delay. This method opens many opportunities for using the neural drive to interface humans with assistive devices.

A deep CNN framework for neural drive estimation from HD-EMG across contraction intensities and joint angles

“…We have previously proposed a deep CNN approach to identify individual MU spike trains from HD-EMG signals [26]. In this previous study, the structure of the deep CNN was customized for each contraction intensity based on the number of MUs identified during preliminary training.…”

“…This threshold ensured a sensitivity higher than 90% and a FP rate lower than 2% [30]. In addition, MU spike trains with less than 150 spikes were not considered for the analysis [26].…”

“…With the BSS method and manual editing, we identified, across all contraction tasks and subjects in each scenario, 20±6 MUs (range 8-30) with 12269±3715 spikes in scenario 1, 29±5 MUs (range [23][24][25][26][27][28][29][30][31][32][33][34][35] with 16616±5462 in scenario 2, and 19±7 MUs (range 7-29) with 31754±9261 spikes in scenario 3. Table . I shows the detailed numbers of MUs and spikes per participant for each contraction task.…”

“…[13]. Deep recurrent neural networks (RNN) [25] and deep CNN [26] can also accurately identify the MU spike trains from HD-EMG. Specifically, these neural networks extract spatio-temporal features from the HD-EMG signal and then identify the spike train for individual MUs after a training based on EMG decomposition algorithms [25], [26].…”

“…Deep recurrent neural networks (RNN) [25] and deep CNN [26] can also accurately identify the MU spike trains from HD-EMG. Specifically, these neural networks extract spatio-temporal features from the HD-EMG signal and then identify the spike train for individual MUs after a training based on EMG decomposition algorithms [25], [26]. However, these methods rely heavily on the features of the MUAP shapes, which are different for different MUs and could be altered by the muscle length even for the same MU [27].…”