Tutorial on MUedit: An open-source software for identifying and analysing the discharge timing of motor units from electromyographic signals

We introduce the open-source software MUedit and we describe its use for identifying the discharge timing of motor units from all types of electromyographic (EMG) signals recorded with multi-channel systems. MUedit performs EMG decomposition using a blind-source separation approach. Following this, users can display the estimated motor unit pulse trains and inspect the accuracy of the automatic detection of discharge times. When necessary, users can correct the automatic detection of discharge times and recalculate the motor unit pulse train with an updated separation vector. Here, we provide an open-source software and a tutorial that guides the user through (i) the parameters and steps of the decomposition algorithm, and (ii) the manual editing of motor unit pulse trains. Further, we provide simulated and experimental EMG signals recorded with grids of surface electrodes and intramuscular electrode arrays to benchmark the performance of MUedit. Finally, we discuss advantages and limitations of the blind-source separation approach for the study of motor unit behaviour during tonic muscle contractions.

Larger and Denser: An Optimal Design for Surface Grids of EMG Electrodes to Identify Greater and More Representative Samples of Motor Units

The spinal motor neurons are the only neural cells whose individual activity can be noninvasively identified. This is usually done using grids of surface electromyographic (EMG) electrodes and source separation algorithms; an approach called EMG decomposition. In this study, we combined computational and experimental analyses to assess how the design parameters of grids of electrodes influence the number and the properties of the identified motor units. We first computed the percentage of motor units that could be theoretically discriminated within a pool of 200 simulated motor units when decomposing EMG signals recorded with grids of various sizes and interelectrode distances (IEDs). Increasing the density, the number of electrodes, and the size of the grids, increased the number of motor units that our decomposition algorithm could theoretically discriminate, i.e., up to 83.5% of the simulated pool (range across conditions: 30.5-83.5%). We then identified motor units from experimental EMG signals recorded in six participants with grids of various sizes (range: 2-36 cm2) and IED (range: 4-16 mm). The configuration with the largest number of electrodes and the shortest IED maximized the number of identified motor units (56 ± 14; range: 39-79) and the percentage of early recruited motor units within these samples (29 ± 14%). Finally, the number of identified motor units further increased with a prototyped grid of 256 electrodes and an IED of 2 mm. Taken together, our results showed that larger and denser surface grids of electrodes allow to identify a more representative pool of motor units than currently reported in experimental studies.

Evaluation method of motor unit number index based on optimal muscle strength combination

Repeatability is an important attribute of motor unit number index (MUNIX) technology. This paper proposes an optimal contraction force combination for MUNIX calculation in an effort to improve the repeatability of this technology. In this study, the surface electromyography (EMG) signals of the biceps brachii muscle of eight healthy subjects were initially recorded with high-density surface electrodes, and the contraction strength was the maximum voluntary contraction force of nine progressive levels. Then, by traversing and comparing the repeatability of MUNIX under various combinations of contraction force, the optimal combination of muscle strength is determined. Finally, calculate MUNIX using the high-density optimal muscle strength weighted average method. The correlation coefficient and the coefficient of variation are utilized to assess repeatability. The results show that when the muscle strength combination is 10, 20, 50 and 70% of the maximum voluntary contraction force, the repeatability of MUNIX is greatest, and the correlation between MUNIX calculated using this combination of muscle strength and conventional methods is high (PCC > 0.99), the repeatability of the MUNIX method improved by 11.5-23.8%. The results indicate that the repeatability of MUNIX differs for various combinations of muscle strength and that MUNIX, which is measured with a smaller number and lower-level contractility, has greater repeatability.

Analytical Modelling of Surface EMG Signals Generated by Curvilinear Fibers with Approximate Conductivity Tensor

OBJECTIVE

Mathematical modelling of surface electromyographic (EMG) signals has been proven a valuable tool to interpret experimental data and to validate signal processing techniques. Most analytical EMG models only consider muscle fibers with specific and fixed arrangements. However, the fiber orientation and curvature may change along the fiber paths and may differ from fiber to fiber. Here we propose a subject-specific EMG model that simulates the fiber trajectories in muscles of the upper arm and analytically derives the action potentials assuming an approximate conductivity tensor.

METHODS

Magnetic Resonance (MR) images were acquired to identify and generate muscle fiber paths and to determine the muscle locations in a cylindrical volume conductor. While the propagation of the action potentials followed the identified curvilinear fiber paths, the conductivity tensor was not adapted to the fiber direction but approximated along the longitudinal axis of the cylindrical volume conductor. Single fiber action potentials (SFAPs) were computed by simulating the generation, propagation, and extinction of membrane current sources. To validate the assumption of the approximate conductivity tensor, two numerical models were implemented for comparison with the analytical solution. The first numerical model reproduced the analytical model and therefore included an approximation for the conductivity tensor. The second numerical model included the exact conductivity tensor derived from the fiber curvatures.

RESULTS

The motor unit action potentials generated by the proposed analytical model and the two numerical models were highly similar (cross-correlation >0.98, normalized root mean square error, nRMSE 0.04, relative error in the median frequency of the simulated waveforms of approximately 3%). The proposed analytical model was also evaluated by comparing simulated and experimentally recorded compound muscle action potentials (CMAPs). The CMAPs simulated with the proposed model better matched the experimental data (cross-correlation >0.90 and nRMSE <0.25 for the majority of the channels) than a model with straight fibers. Finally, the proposed model was representatively used to test the accuracy of an EMG decomposition algorithm, providing a realistic benchmark.

CONCLUSIONS AND SIGNIFICANCE

The proposed analytical model generates action potentials that reflect the spatial distributions of muscle fibers with curvilinear paths. The simulated signals are more realistic than signals generated by analytical models with straight fibers and can therefore be applied for testing EMG processing algorithms with a trade-off between simulation accuracy and computational speed.

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

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: 1) multiple single-output deep CNN (SO-DCNN) with one MU decomposed per network, and 2) 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 to 21.4 s/epoch vs. 6.5 to 47.8 s/epoch, respectively) and prediction time (0.04 vs. 0.27 s/sample, 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.