High-density Surface Electromyography Signals Research Articles

Extensor digitorum communis compartments with a double compensation synergy strategy for constant force control of the index finger.

Precise sustained force control of the fingers is important for achieving flexible hand movements. However, how neuromuscular compartments within a forearm multi-tendon muscle cooperate to achieve constant finger force remains unclear. This study aimed to investigate the coordination strategies across multiple compartments of the extensor digitorum communis (EDC) during index finger sustained constant extension. Nine subjects performed index finger extensions of 15%, 30%, and 45% maximal voluntary contractions, respectively. High-density surface electromyography signals were recorded from the EDC and then analyzed using non-negative matrix decomposition to extract activation patterns and coefficient curves of EDC compartments. The results showed two activation patterns with stable structures during all tasks: one pattern corresponding to the index finger compartment was named master pattern; whereas the other corresponding to other compartments was named auxiliary pattern. Further, the intensity and stability of their coefficient curves were assessed using the root mean square value (RMS) and coefficient of variation (CV). The RMS and CV values of the master pattern increased and decreased with time, respectively, while the corresponding values of the auxiliary pattern were both negatively correlated with the formers. These findings suggested a special coordination strategy across EDC compartments during index finger constant extension, manifesting as two compensations of the auxiliary pattern for the intensity and stability of the master pattern. The proposed method provides new insight into the synergy strategy across multiple compartments within a forearm multi-tendon during sustained isometric contraction of a single finger and a new approach for constant force control of prosthetic hands.

The effects of α-lactalbumin supplementation and handgrip contraction on soleus motoneuron excitability

We tested two strategies that hypothetically increase serotonin availability (α-lactalbumin consumption and a remote submaximal handgrip contraction) on estimates of persistent inward currents (PICs) amplitude of soleus muscle in healthy participants. With a randomised, double-blind, and cross-over design, 13 healthy participants performed triangular-shaped ramp contractions with their plantar flexors (20% of maximal torque), followed by a 30-s handgrip sustained contraction (40% of maximal force) and consecutive repeated triangular-shaped contractions. This was performed before and after the consumption of either 40g of α-lactalbumin, an isonitrogenous beverage (Zein) or an isocaloric beverage (Corn-starch). Soleus motor units discharge rates were analysed from high-density surface electromyography signals. PICs were estimated by calculating the delta frequency (ΔF) of motor unit train spikes using the paired motor unit technique. ΔF (0.19 pps; p = 0.001; d = 0.30) and peak discharge rate (0.20 pps; p < 0.001; d = 0.37) increased after the handgrip contraction, irrespective of the consumed supplement. No effects of α-lactalbumin were observed. Our results indicate that 40g of α-lactalbumin was unable to modify intrinsic motoneuron excitability. However, performing a submaximal handgrip contraction before the plantar flexion triangular contraction was capable of increasing ΔF and discharge rates on soleus motor units. These findings highlight the diffused effects of serotonergic input, its effects on motoneuron discharge behaviour, and suggest a cross-effector effect within human motoneurons.

Reliability of high-density surface electromyography for assessing characteristics of the thoracic erector spinae during static and dynamic tasks

PurposeTo establish intra- and inter-session reliability of high-density surface electromyography (HDEMG)-derived parameters from the thoracic erector spinae (ES) during static and dynamic goal-directed voluntary movements of the trunk, and during functional reaching tasks. MethodsTwenty participants performed: 1) static trunk extension, 2) dynamic trunk forward and lateral flexion, and 3) multidirectional functional reaching tasks on two occasions separated by 7.5 ± 1.2 days. Muscle activity was recorded bilaterally from the thoracic ES. Root mean square (RMS), coordinates of the barycentre, mean frequency (MNF), and entropy were derived from the HDEMG signals. Reliability was determined with intraclass correlation coefficient (ICC), coefficient of variation, and standard error of measurement. ResultsGood-to-excellent intra-session reliability was found for all parameters and tasks (ICC: 0.79-0.99), whereas inter-session reliability varied across tasks. Static tasks demonstrated higher reliability in most parameters compared to functional and dynamic tasks. Absolute RMS and MNF showed the highest overall reliability across tasks (ICC: 0.66-0.98), while reliability of the barycentre was influenced by the direction of the movements. ConclusionRMS and MNF derived from HDEMG show consistent inter-session reliability in goal-directed voluntary movements of the trunk and reaching tasks, whereas the measures of the barycentre and entropy demonstrate task-dependent reliability.

Neural decoding from surface high-density EMG signals: influence of anatomy and synchronization on the number of identified motor units

Objective. High-density surface electromyography (HD-sEMG) allows the reliable identification of individual motor unit (MU) action potentials. Despite the accuracy in decomposition, there is a large variability in the number of identified MUs across individuals and exerted forces. Here we present a systematic investigation of the anatomical and neural factors that determine this variability. Approach. We investigated factors of influence on HD-sEMG decomposition, such as synchronization of MU discharges, distribution of MU territories, muscle-electrode distance (MED—subcutaneous adipose tissue thickness), maximum anatomical cross-sectional area (ACSAmax), and fiber cross-sectional area. For this purpose, we recorded HD-sEMG signals, ultrasound and magnetic resonance images, and took a muscle biopsy from the biceps brachii muscle from 30 male participants drawn from two groups to ensure variability within the factors—untrained-controls (UT = 14) and strength-trained individuals (ST = 16). Participants performed isometric ramp contractions with elbow flexors (at 15%, 35%, 50% and 70% maximum voluntary torque—MVT). We assessed the correlation between the number of accurately detected MUs by HD-sEMG decomposition and each measured parameter, for each target force level. Multiple regression analysis was then applied. Main results. ST subjects showed lower MED (UT = 5.1 ± 1.4 mm; ST = 3.8 ± 0.8 mm) and a greater number of identified MUs (UT: 21.3 ± 10.2 vs ST: 29.2 ± 11.8 MUs/subject across all force levels). The entire cohort showed a negative correlation between MED and the number of identified MUs at low forces (r = −0.6, p = 0.002 at 15% MVT). Moreover, the number of identified MUs was positively correlated to the distribution of MU territories (r = 0.56, p = 0.01) and ACSAmax (r = 0.48, p = 0.03) at 15% MVT. By accounting for all anatomical parameters, we were able to partly predict the number of decomposed MUs at low but not at high forces. Significance. Our results confirmed the influence of subcutaneous tissue on the quality of HD-sEMG signals and demonstrated that MU spatial distribution and ACSAmax are also relevant parameters of influence for current decomposition algorithms.

High-density Surface Electromyography as Biomarker of Muscle Aging.

Sarcopenia is a muscle disease with adverse changes that increase throughout the lifetime but with different chronological scales between individuals. Addressing "early muscle aging" is becoming a critical issue for prevention. Through the CHRONOS study, we demonstrated the ability of the high-density surface electromyography (HD-sEMG), a noninvasive, wireless, portable technology, to detect both healthy muscle aging and accelerated muscle aging related to a sedentary lifestyle, one of the risk factors of sarcopenia. The HD-sEMG signals were analyzed in 91 healthy young, middle-aged, and old subjects (25-75 years) distributed according to their physical activity status (82 active and 9 sedentary; International Physical Activity Questionnaire) and compared with current methods for muscle evaluation, including muscle mass (dual-energy X-ray absorptiometry [DXA], ultrasonography), handgrip strength, and physical performance. The HD-sEMG signals were recorded from the rectus femoris during sit-to-stand trials, and 2 indexes were analyzed: muscular contraction intensity and muscle contraction dynamics. The clinical parameters did not differ significantly across the aging and physical activity levels. Inversely, the HD-sEMG indexes were correlated to age and were different significantly through the age categories of the 82 active subjects. They were significantly different between sedentary subjects aged 45-54 years and active ones at the same age. The HD-sEMG indexes of sedentary subjects were not significantly different from those of older active subjects (≥55 years). The muscle thicknesses evaluated using ultrasonography were significantly different between the 5 age decades but did not show a significant difference with physical activity. The HD-sEMG technique can assess muscle aging and physical inactivity-related "early aging," outperforming clinical and DXA parameters.

Detecting motor unit abnormalities in amyotrophic lateral sclerosis using high-density surface EMG

ObjectiveThe purpose of this study was to detect specific motor unit (MU) abnormalities in people with amyotrophic lateral sclerosis (ALS) compared to controls using high-density surface electromyography (HD-SEMG). MethodsSixteen people with ALS and 16 control subjects. The participants performed ramp up and sustained contractions at 30% of their maximal voluntary contraction. HD-SEMG signals were recorded in the vastus lateralis muscle and decomposed into individual MU firing behavior using a convolution blind source separation method. ResultsIn total, 339 MUs were detected (people with ALS; n = 93, control subjects; n = 246). People with ALS showed significantly higher mean firing rate, recruitment threshold, coefficient of variation of the MU firing rate, MU firing rate at recruitment, and motoneurons excitability than those of control subjects (p < 0.001). The number of MU, MU firing rate, recruitment threshold, and MU firing rate at recruitment were significantly correlated with disease severity (p < 0.001). Multivariable analysis revealed that an increased MU firing rate at recruitment was independently associated with ALS. ConclusionsThese results suggest increased excitability at recruitment, which is consistent with neurodegeneration results in a compensatory increase in MU activity. SignificanceAbnormal MU firing behavior provides an important physiological index for understanding the pathophysiology of ALS.

ViT-HGR: Vision Transformer-based Hand Gesture Recognition from High Density Surface EMG Signals.

Recently, there has been a surge of significant interest on application of Deep Learning (DL) models to autonomously perform hand gesture recognition using surface Electromyogram (sEMG) signals. Many of the existing DL models are, however, designed to be applied on sparse sEMG signals. Furthermore, due to the complex structure of these models, typically, we are faced with memory constraint issues, require large training times and a large number of training samples, and; there is the need to resort to data augmentation and/or transfer learning. In this paper, for the first time (to the best of our knowledge), we investigate and design a Vision Transformer (ViT) based architecture to perform hand gesture recognition from High Density (HD-sEMG) signals. Intuitively speaking, we capitalize on the recent breakthrough role of the transformer architecture in tackling different com-plex problems together with its potential for employing more input parallelization via its attention mechanism. The proposed Vision Transformer-based Hand Gesture Recognition (ViT-HGR) framework can overcome the aforementioned training time problems and can accurately classify a large number of hand gestures from scratch without any need for data augmentation and/or transfer learning. The efficiency of the proposed ViT-HGR framework is evaluated using a recently-released HD-sEMG dataset consisting of 65 isometric hand gestures. Our experiments with 64-sample (31.25 ms) window size yield average test accuracy of 84.62 ± 3.07%, where only 78,210 learnable parameters are utilized in the model. The compact structure of the proposed ViT-based ViT-HGR framework (i.e., having significantly reduced number of trainable parameters) shows great potentials for its practical application for prosthetic control.

Cancelable HD-SEMG Biometric Identification via Deep Feature Learning.

Conventional biometric modalities, such as the face, fingerprint, and iris, are vulnerable against imitation and circumvention. Accordingly, secure biometric modalities with cancelable properties are needed for personal identification, especially in smart healthcare applications. Here we developed a person identification model using high-density surface electromyography (HD-sEMG) as biometric traits. In this model, the HD-sEMG biometric templates are cancelable and could be customized by the users through finger isometric contractions. A deep feature learning approach, implemented by convolutional neural networks (CNNs) is used to capture user-specific patterns from HD-sEMG signals and make identification decisions. This model has been validated on twenty-two subjects, with training and testing data acquired from two different days. The rank-1 identification accuracy and equal error rate for 44 identities (22 subjects × 2 accounts) can reach 87.23% and 4.66%, respectively. The cross-day identification accuracy of the proposed model is higher than the results of previous methods reported in the literature. The usability and efficiency of the proposed model are also investigated, indicating its potentials for practical applications.

Laterality of the activation of the vastus lateralis muscle in females with Parkinson's disease during the medication state compared with healthy controls.

The aim of this study was to quantify the laterality of motor unit (MU) activation properties in people with Parkinson's disease (PD) during force production (low- to high-intensity contraction) using high-density surface electromyography (HD-SEMG). Sixteen females with PD (age = 69.9 ± 7.6 years, disease duration = 4.9 ± 5.1 years) and 14 healthy female subjects (age = 68.6 ± 3.6 years) were enrolled in the study and performed submaximal ramp-up contractions during isometric knee extension. HD-SEMG signals were recorded from both vastus lateralis muscles. We calculated the level of heterogeneity in the spatial distribution patterns of the HD-SEMG signals and determined the modified entropy, coefficient of variation of the root mean square (RMS), and correlation coefficient to evaluate MU activation properties. Pearson's correlation coefficients were calculated to examine the relationships between disease severity and the RMS and EMG variables. The RMS value and heterogeneity were significantly higher and lower on the more-affected side in people with PD than on the other side in people with PD or either side in control subjects (p < 0.05). People with PD exhibited the temporal changes of spatial MUs activation properties showed significant laterality when compared to healthy control subjects not only in the low-intensity contractions but also in high-intensity contraction. Moderate-to-strong correlations were observed between disease severity and RMS and EMG variables in people with PD (r > 0.6, p < 0.001). We compared the laterality of MU activation properties between the people with PD and the control subjects. These findings suggest that people with PD have asymmetrical MU activation properties, regardless of the magnitude of force production.

Alterations of tibialis anterior muscle activation pattern in subjects with type 2 diabetes and diabetic peripheral neuropathy

Diabetic peripheral neuropathy (DPN) is associated with loss of motor units (MUs), which can cause changes in the activation pattern of muscle fibres. This study investigated the pattern of muscle activation using high-density surface electromyography (HD-sEMG) signals from subjects with type 2 diabetes mellitus (T2DM) and DPN. Thirty-five adults participated in the study: 12 healthy subjects (HV), 12 patients with T2DM without DPN (No-DPN) and 11 patients with T2DM with DPN (DPN). HD-sEMG signals were recorded in the tibialis anterior muscle during an isometric contraction of ankle dorsiflexion at 50% of the maximum voluntary isometric contraction (MVIC) during 30-s. The calculated HD-sEMG signals parameters were the normalised root mean square (RMS), normalised median frequency (MDF), coefficient of variation (CoV) and modified entropy (ME). The RMS increased significantly (p = 0.001) with time only for the DPN group, while the MDF decreased significantly (p < 0.01) with time for the three groups. Moreover, the ME was significantly lower (p = 0.005), and CoV was significantly higher (p = 0.003) for the DPN group than the HV group. Using HD-sEMG, we have demonstrated a reduction in the number of MU recruited by individuals with DPN. This study provides proof of concept for the clinical utility of this technique for identifying neuromuscular impairment caused by DPN.

Design and Validation of a Wireless Body Sensor Network for Integrated EEG and HD-sEMG Acquisitions.

Sensorimotor integration is the process through which the human brain plans the motor program execution according to external sources. Within this context, corticomuscular and corticokinematic coherence analyses are common methods to investigate the mechanism underlying the central control of muscle activation. This requires the synchronous acquisition of several physiological signals, including EEG and sEMG. Nevertheless, physical constraints of the current, mostly wired, technologies limit their application in dynamic and naturalistic contexts. In fact, although many efforts were made in the development of biomedical instrumentation for EEG and High Density-surface EMG (HD-sEMG) signal acquisition, the need for an integrated wireless system is emerging. We hereby describe the design and validation of a new fully wireless body sensor network for the integrated acquisition of EEG and HD-sEMG signals. This Body Sensor Network is composed of wireless bio-signal acquisition modules, named sensor units, and a set of synchronization modules used as a general-purpose system for time-locked recordings. The system was characterized in terms of accuracy of the synchronization and quality of the collected signals. An in-depth characterization of the entire system and an head-to-head comparison of the wireless EEG sensor unit with a wired benchmark EEG device were performed. The proposed device represents an advancement of the State-of-the-Art technology allowing the integrated acquisition of EEG and HD-sEMG signals for the study of sensorimotor integration.

Optimizing Input for Gesture Recognition using Convolutional Networks on HD-sEMG Instantaneous Images.

Hand gesture recognition using high-density surface electromyography (HD-sEMG) has gained increasing attention recently due its advantages of high spatio-temporal resolution. Convolutional neural networks (CNN) have also recently been implemented to learn the spatio-temporal features from the instantaneous samples of HD-sEMG signals. While the CNN itself learns the features from the input signal it has not been considered whether certain pre-processing techniques can further improve the classification accuracies established by previous studies. Therefore, common pre-processing techniques were applied to a benchmark HD-sEMG dataset (CapgMyo DB-a) and their validation accuracies were compared. Monopolar, bipolar, rectified, common-average referenced, and Laplacian spatial filtered configurations of the HD-sEMG signals were evaluated. Results showed that the baseline monopolar HD-sEMG signals maintained higher prediction accuracies versus the other signal configurations. The results of this study discourage the use of extra pre-processing steps when using convolutional networks to classify the instantaneous samples of HD-sEMG for gesture recognition.

The force-generation capacity of the tibialis anterior muscle at different muscle\u2013tendon lengths depends on its motor unit contractile properties

PurposeMuscle–tendon length can influence central and peripheral motor unit (MU) characteristics, but their interplay is unknown. This study aims to explain the effect of muscle length on MU firing and contractile properties by applying deconvolution of high-density surface EMG (HDEMG), and torque signals on the same MUs followed at different lengths during voluntary contractions.MethodsFourteen participants performed isometric ankle dorsiflexion at 10% and 20% of the maximal voluntary torque (MVC) at short, optimal, and long muscle lengths (90°, 110°, and 130° ankle angles, respectively). HDEMG signals were recorded from the tibialis anterior, and MUs were tracked by cross-correlation of MU action potentials across ankle angles and torques. Torque twitch profiles were estimated using model-based deconvolution of the torque signal based on composite MU spike trains.ResultsMean discharge rate of matched motor units was similar across all muscle lengths (P = 0.975). Interestingly, the increase in mean discharge rate of MUs matched from 10 to 20% MVC force levels at the same ankle angle was smaller at 110° compared with the other two ankle positions (P = 0.003), and the phenomenon was explained by a greater increase in twitch torque at 110° compared to the shortened and lengthened positions (P = 0.002). This result was confirmed by the deconvolution of electrically evoked contractions at different stimulation frequencies and muscle–tendon lengths.ConclusionHigher variations in MU twitch torque at optimal muscle lengths likely explain the greater force-generation capacity of muscles in this position.

Deficit in knee extension strength following anterior cruciate ligament reconstruction is explained by a reduced neural drive to the vasti muscles.

The persistence of quadriceps weakness represents a major concern following anterior cruciate ligament reconstruction (ACLR). The underlying adaptations occurring in the activity of spinal motoneurons are still unexplored. This study examined the discharge patterns of large populations of motor units (MUs) in the vastus lateralis (VL) and vastus medialis muscles following ACLR. Nine ACLR individuals and 10 controls performed unilateral trapezoidal contractions of the knee extensor muscles at 35%, 50% and 70% of the maximal voluntary isometric force (MVIF). High-density surface electromyography (HDsEMG) was used to record the myoelectrical activity of the vasti muscles in both limbs. HDsEMG signals were decomposed with a convolutive blind source separation method and MU properties were extracted and compared between sides and groups. The ACLR group showed a lower MVIF on the reconstructed side compared to the contralateral side (28.1%; P<0.001). This force deficit was accompanied by reduced MU discharge rates (∼21%; P<0.05), lower absolute MU recruitment and derecruitment thresholds (∼22% and ∼22.5%, respectively; P<0.05) and lower input-output gain of motoneurons (27.3%; P=0.009). Deficits in MU discharge rates of the VL and in absolute recruitment and derecruitment thresholds of both vasti MUs were associated with deficits in MVIF (P<0.05). A strong between-side correlation was found for MU discharge rates of the VL of ACLR individuals (P<0.01). There were no significant between-group differences (P>0.05). These results indicate that mid- to long-term strength deficits following ACLR may be attributable to a reduced neural drive to vasti muscles, with potential changes in excitatory and inhibitory synaptic inputs. KEY POINTS: Impaired expression and control of knee extension forces is common after anterior cruciate ligament reconstruction and is related to high risk of a second injury. To provide novel insights into the neural basis of this impairment, the discharge patterns of motor units in the vastus lateralis and vastus medialis were investigated during voluntary force contractions. There was lower knee extensor strength on the reconstructed side with respect to the contralateral side, which was explained by deficits in motor unit discharge rate and an altered motoneuronal input-output gain. Insufficient excitatory inputs to motoneurons and increased inhibitory afferent signals potentially contributed to these alterations. These results further our understanding of the neural underpinnings of quadriceps weakness following anterior cruciate ligament reconstruction and can help to develop effective rehabilitation protocols to regain muscle strength and reduce the risk of a second injury.

Automatic decomposition of pediatric high density surface EMG: A pilot study

This pilot study presents a novel application of high density surface electromyography (EMG) decomposition in pediatric patients, with a view toward promoting its potential clinical application for examination of pediatric neuromuscular diseases. Automatic progressive FastICA peel-off (APFP) framework was used to decompose high density surface EMG signals recorded from the first dorsal interosseous and abductor pollicis brevis muscles of children with spinal muscular atrophy. The performance of the decomposition with reduced EMG recording channels was also investigated. In total 131 motor units were extracted from 15 trials of 64-channel (8 ​× ​8) surface EMG. The decomposition yield reduced to 102 motor units when 36-channel (6 ​× ​6) was used, among which 84 motor units were common ones with an average matching rate of (95.68 ​± ​5.26) %. The decomposition yield further reduced to 65 motor units when only 16-channel (4 ​× ​4) was used, among which 57 were common ones with an average matching rate of (95.99 ​± ​4.56) %. These results indicate that using the APFP framework single motor unit activity can be reliably and automatically extracted from pediatric surface EMG signals recorded by an electrode array, which can facilitate potential clinical application of surface EMG as an alternative or supplement to invasive needle EMG for examination of neuromuscular diseases in children.

Elicitation hybrid spatial features from HD-sEMG signals for robust classification of gestures in real-time

ABSTRACT Recently, using high-density surface electromyography (HD-sEMG) electrodes in prosthetics have outdone the challenges of electrodes sites on the muscle, while the high accuracy of HD-sEMG signals classification will improve prosthetics performance. A new concept has emerged that the robust features extraction methods increase the efficiency of the classification regardless of the classifier. In addition, there are many factors affecting the quality of the signal, and thus the quality of classification such as stress, fatigue, disease, muscular dystrophy … etc. In this paper, these challenges will be reduced by the proposed approach for extraction hybrid features from the HD-EMG signal based on the histogram-oriented gradient (HOG) algorithm and signal intensity features, where the support vector machine (SVM) classifier is used for the classification process. The results showed high accuracy of the classification and successful in real-time tests. Also, the classification results of these experiments have overcome the challenge of long term classification.

Ageing Effect Evaluation on HD-sEMG Signals Using CCA Approach

ObjectivesThe objective of the proposed study is to exploit the technology of high-density surface electromyography (HD-sEMG), in order to evaluate the muscle activation in young and elderly subjects during a daily life gesture, namely, Sit To Stand (STS), using wireless connected ambulatory equipment (TMSi©) and Blind Source Separation (BSS) approach with Canonical Correlation Analysis (CCA). Materials and methodsSixteen subjects participated (50% females) divided into two categories (‘H1’: young (30.62 yrs ±5.92, 23.95 kg/m2 ±3.08), versus ‘H2’: old (61.87 yrs ±7.98, 23.4 kg/m2 ±3.38)), in the recording of HD-sEMG signals, using 32-electrodes square grids (4×8), during Sit To Stand (STS) motion, three times at spontaneous speed. The studied muscle is the Rectus Femoris (RF) muscle. The recorded HD-sEMG signals were analyzed with CCA approach to extract correlation coefficient sets according to two age categories (young versus old), in order to evaluate its discriminating power with ageing. Statistical tests (t-test) were used to evaluate the discrimination for these two categories. ResultsThe calculation of CCA correlation coefficients showed a significant difference between young and old category concerning the mean CCA correlation coefficient (P<0.001***) and also the standard deviation of the CCA correlation coefficients (P<0.0001****). ConclusionThe obtained results are promising and indicate a clear difference between the obtained source variability using CCA method between the young and the old tested subjects during daily life motion. Furthermore, these estimated sources seem to be impacted by both anatomical and functional modifications with ageing.

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: (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.

Simultaneous and proportional control of wrist and hand movements by decoding motor unit discharges in real time

Objective. Surface electromyography (EMG) decomposition techniques can be used to establish human-machine interfacing (HMI), but most investigations are implemented offline due to the computational load of the approach. Here, we generalize the offline decomposition algorithm to identify the motor unit (MU) activities in real time, and we propose a MU-based approach for online simultaneous and proportional control (SPC) of multiple motor tasks. Approach. High-density surface EMG signals recorded from forearm muscles were decomposed into motor unit spike trains (MUSTs) with the proposed decomposition method. The MUSTs were first pooled into clusters in the calibration phase and the cumulative discharges of active MUs in each group were extracted as the control signal for each motor task. Then the subjects were instructed to control a virtual cursor with multiple motor tasks involving grasp and wrist movements. Fifteen able-bodied subjects and two patients with limb deficiency participated in the experiments to validate the proposed control scheme. Main results. On average, over 20 MUSTs were identified in real time with an estimated decomposition accuracy 85%. The cumulative discharge in each pool was highly correlated with the activation of the specific motion (R = 0.93 ± 0.05). Moreover, the proposed MU-based method had superior performance in online tests than conventional myo-control methods based on global EMG features. Significance. These results indicate the feasibility of real-time neural decoding in a non-invasive way. Moreover, the superior performance in online tests proves the potential of the MU-based approach for the SPC, promoting the application of EMG decomposition for HMI systems.

Deep Learning for Robust Decomposition of High-Density Surface EMG Signals.

Blind source separation (BSS) algorithms, such as gradient convolution kernel compensation (gCKC), can efficiently and accurately decompose high-density surface electromyography (HD-sEMG) signals into constituent motor unit (MU) action potential trains. Once the separation matrix is blindly estimated on a signal interval, it is also possible to apply the same matrix to subsequent signal segments. Nonetheless, the trained separation matrices are sub-optimal in noisy conditions and require that incoming data undergo computationally expensive whitening. One unexplored alternative is to instead use the paired HD-sEMG signal and BSS output to train a model to predict MU activations within a supervised learning framework. A gated recurrent unit (GRU) network was trained to decompose both simulated and experimental unwhitened HD-sEMG signal using the output of the gCKC algorithm. The results on the experimental data were validated by comparison with the decomposition of concurrently recorded intramuscular EMG signals. The GRU network outperformed gCKC at low signal-to-noise ratios, proving superior performance in generalising to new data. Using 12 seconds of experimental data per recording, the GRU performed similarly to gCKC, at rates of agreement of 92.5% (84.5%-97.5%) and 94.9% (88.8%-100.0%) respectively for GRU and gCKC against matched intramuscular sources.