Electromyogram Features Research Articles

Simultaneous hand/wrist motion recognition and continuous grasp force estimation based on nonlinear spectral sEMG features for transradial amputees

Simultaneously and accurately estimating both grasp motion classes and force based on surface electromyogram (EMG) of residual arms is essential for precise grasping with a dexterous prosthetic hand. However, accurate grasping-mode prediction and grasping-force estimation often contradict each other. When performing hand grasping motion, different grasping forces will generate different EMG patterns, which would decay the grasp motion prediction accuracy. In this study, we proposed a framework of fusing motion classification and continuous grasp force estimation based on a new set of nonlinear Mel-Frequency Spectrum features of EMG signals. The performance of the proposed method was tested on eight able bodied (AB) subjects and four transradial amputees (TR) who are the end-users of dexterous prostheses. The experimental results showed that the newly proposed features obtained a significant reduction in the average classification error rate compared to other well-known feature sets, achieving improvements of about 5 % to 28 % in the average classification accuracy across all subjects and force levels. Simultaneously, the proposed features obtained the best continuous grasp force estimation accuracies with R2 (coefficient of determination) of 0.96 and 0.83, and root mean square of maximum voluntary contractions of grasp force (std%GFMVC) of 6.2 % and 10.5 % for the AB and TR subjects, respectively, compared to the previous EMG features. This study suggested that the proposed method would be helpful for the stable and fine manipulation of multifunctional prosthetic hand.

Generation of Pulse Sequence Using EMG Signals for Application in Transfemoral Prosthesis

The percentage of people having a lower leg amputation is high, and the incidence of unemployment among these amputees is likewise rising. Hence, it requires the intervention of an innovative solution to serve the function of a lost limb. Electromyogram (EMG) signals is a result of the potential generated by muscles during contraction. In this work, an attempt has been made to extract EMG signals from four set of muscle groups and the acquired signals were pre-processed and transformed to pulses to extract the contraction phase of the signal. Furthermore, the processed signals were subject to feature extraction process where in the Mean Absolute Value (MAV), Integrated EMG Feature (IEMG) and various statistical parameters associated with the signal such as the mean, median, standard deviation, variance, kurtosis, skewness was calculated in order to serve as an input to drive the stepper motor of a transfemoral prosthesis. To promote real time acquisition and control, a transfemoral socket with an ischial containment has been designed.

A resistive force correlated electromyogram feature selection method for muscle strength prediction

The need for standardization of muscle strength measurement techniques has been of interest over many years. The relationship between Electromyography (EMG) and strength of isometric muscle contraction provides an alternative method of a more natural means of analyzing muscle strength. Quadriceps and Erector Spinae muscle groups are considered in this study. Muscle specific isometric protocol is followed where dynamometer resisted force and EMG are simultaneously acquired. A three level feature selection process based on the measured force and EMG is proposed. The first and second level selection is based on visible shape similarity and statistical Spearmans’ correlation respectively between the resisted force and EMG features, followed by feature ranking using Laplacian Score. A very weak but statistically insignificant (p > 0.05) correlation was observed for SSI and Log features for both Quadriceps (‘SSI’ ρ = 0.04, ‘LOG’ ρ = 0.13) and Erector Spinae (‘SSI’ ρ = 0.08, ‘LOG’ ρ = 0.002) muscle groups. The proposed feature set including IEMG, RMS, WL, and FE consistently top ranked in Laplacian score for all the muscles. Fuzzy C-means clustering method is used to classify low, moderate and strong muscle strength levels. In comparison to the popular feature set proposed by Hudgin, the performance of the proposed feature set revealed a better cluster separation (SI score > 0.5, DB Index < 0.5, CH Index > 100). The choice of three of clusters for low, moderate and strong muscle strength using the proposed feature set has also resulted in a low Hartington’s index (<3). The proposed feature set proves to be more robust in muscle strength classification using surface electromyogram signals.

A Semi-Supervised Progressive Learning Algorithm for Brain-Computer Interface.

Brain-computer interface (BCI) usually suffers from the problem of low recognition accuracy and large calibration time, especially when identifying motor imagery tasks for subjects with indistinct features and classifying fine grained motion control tasks by electroencephalogram (EEG)-electromyogram (EMG) fusion analysis. To fill the research gap, this paper presents an end-to-end semi-supervised learning framework for EEG classification and EEG-EMG fusion analysis. Benefiting from the proposed metric learning based label estimation strategy, sampling criterion and progressive learning scheme, the proposed framework efficiently extracts distinctive feature embedding from the unlabeled EEG samples and achieves a 5.40% improvement on BCI Competition IV Dataset IIa with 80% unlabeled samples and an average 3.35% improvement on two public BCI datasets. By employing synchronous EMG features as pseudo labels for the unlabeled EEG samples, the proposed framework further extracts deep level features of the synergistic complementarity between the EEG signals and EMG features based on the deep encoders, which improves the performance of hybrid BCI (with a 5.53% improvement for the Upper Limb Motion Dataset and an average 4.34% improvement on two hybrid datasets). Moreover, the ablation experiments show that the proposed framework can substantially improve the performance of the deep encoders (with an average 5.53% improvement). The proposed framework not only largely improves the performance of deep networks in the BCI system, but also significantly reduces the calibration time for EEG-EMG fusion analysis, which shows great potential for building an efficient and high-performance hybrid BCI for the motor rehabilitation process.

Combined Dynamic Time Warping and Spatiotemporal Attention for Myoelectric Control.

The success of pattern recognition based upper-limb prostheses control is linked to their ability to extract appropriate features from the electromyogram (EMG) signals. Traditional EMG feature extraction (FE) algorithms fail to extract spatial and inter-temporal information from the raw data, as they consider the EMG channels individually across a set of sliding windows with some degree of overlapping. To tackle these limitations, this paper presents a method that considers the spatial information of multi-channel EMG signals by utilising dynamic time warping (DTW). To satisfy temporal considerations, inspired by Long Short-Term Memory (LSTM) neural networks, our algorithm evolves the DTW feature representation across long and short-term components to capture the temporal dynamics of the EMG signal. As such the contribution of this paper is the development of a recursive spatio-temporal FE method, denoted as Recursive Temporal Warping (RTW). To investigate the performance of the proposed method, an offline EMG pattern recognition study with 53 movement classes performed by 10 subjects wearing 8 to 16 EMG channels was considered with the results compared against several conventional as well as deep learning-based models. We show that the use of the RTW can reduce classification errors significantly, paving the way for future real-time implementation.

Surface Electromyogram Feature Set Optimization for Lower Limb Activity Classification

Surface electromyogram (SEMG) shows the vital information of human motor activity. This information can identify the motor activity and controlling assistive and rehabilitative devices. However, its efficient extraction and interpretation plays a very critical role. The present study is focused on lower limb activity classification by selecting the efficient feature vector of the lower limb SEMG signal. Here, the SEMG signal-dependent continuous locomotion mode classification method has been proposed along with a dual-stage Feature Selection Algorithm (FSA). To evaluate the proposed method, SEMG signals of fifteen subjects were recorded from two lower limb muscles (Fibularis longus and Biceps Femoris) for five lower limb activities. The performance of six different classifiers was compared for intuitive feature vectors and FSA. The study analysed the performance of the proposed model for a single muscle approach and a dual muscle approach of activity classification. The time domain features were found more effective over other features, whereas Biceps Femoris muscle came out as a dominant muscle. The FSA enhanced the performance of classification model with fewer features as compared to intuitive feature subsets. Moreover, the dual muscle approach outperformed (p-value<0.05) over the single muscle approach. The best performance, for the dual muscle approach, was achieved 97.73 ± 2.47%, 98.99 ± 1.26% and 96.33 ± 2.70% for Neural Network, Linear Discriminant Analysis and Support Vector Machine classifiers, respectively (p-value > 0.05).

Recognition of Emotion in Indian Classical Dance Using EMG Signal

Indian classical dance forms like Kathak are an enrichment of Indian culture and tradition. These dance forms glorify its beauty by expressing nine emotions (Navras) such as Adbhut (amazed), Bhayanaka (fearful), Hasya (humorous), Karuna (tragic), Raudra (fierce), Shringar (loving smile), Veer (heroic), Bibhatsa (disgusted), and Shant (peaceful). Identifying correct emotions is an important task. The objective of this research work is to recognize Navras in the Kathak dance. Proposed research work can assist dance teachers in an accurate and unbiased evaluation process of dance examination. This research work analyzed the Electromyogram (EMG) signals acquired from eleven subjects. The EMG signals collected from the various locations on the face and neck represent the emotions and head movement. The EMG signals are processed to extract integrated EMG (IEMG) features. This research introduced a new feature named 'difference in IEMG feature' for improving the accuracy of emotion recognition. For the classification of nine emotions, the Least Square Support Vector Machine (LSSVM), Nonlinear Autoregressive Exogenous Network (NARX), and Long- and Short-Term Memory (LSTM) classifiers were used. The classifiers' performance is judged with head motion and without head motion. The classification accuracies are calculated using a maximum, variance, and mean of the feature. LSSVM, NARX, and LSTM classifiers achieved 60.80%, 81.67%, and 92.28% classification accuracies, respectively, using the IEMG feature and head motion. Using the new feature, LSSVM, NARX, and LSTM classifiers achieved 64.29%, 81.27%, and 93.63% classification accuracies, respectively. The overall classification accuracy improved by 1.46% by using the new feature.

Reduction of Limb Position Invariant of SEMG Signals for Improved Prosthetic Control Using Spectrogram

Prostheses are artificial devices that replace a missing body part, which might be lost through injury, infection, or a condition present at birth. It is proposed to re-establish the normal functions of the missing body part and can be made by hand or with a computer-aided design. As per the World Health Organization, around 160,000 individuals in Malaysia are required to use prostheses. One of the elements that influence the current prosthesis control is that the variety in the limb position and normal use results in different electromyogram (EMG) signals with the same movement at various positions. Consequently, the objective of this study is to ensure that amputees can control their prosthetics in an exact manner regardless of their hand movement and limb position. The raw EMG signals are taken from eight different hand movements’ classes at five different limb positions and each of these hand movements has seven electrodes attach to it. This paper utilizes time-frequency distribution which is spectrogram to extract the EMG feature and six SVM classification learners; linear, quadratic, cubic, fine Gaussian, medium Gaussian, and coarse Gaussian were compared to find the most reasonable one for this application. The analysis performance is then verified based on classification accuracy. From the results, the overall accuracy for the classification is 65% (linear), 87.5% (quadratic) and 97.5% (cubic), 100% (fine Gaussian), 70% (medium Gaussian, and 45% (coarse Gaussian), respectively. It is believed that the study could fill in as knowledge to improve conventional prosthetic control strategies.

A long short-term recurrent spatial-temporal fusion for myoelectric pattern recognition

Current state-of-the-art myoelectric interfaces employ traditional pattern recognition (PR) algorithms to decode the Electromyogram (EMG) signals into hand movements for controlling artificial limbs. Recently, deep learning (DL) models have also been exploited for EMG feature learning/extraction. Models like Convolutional Neural Networks (CNN), which capture the spatial correlations, and Long Short-Term Memory (LSTM), which capture the non-linear temporal dynamics of EMG time-series data, have been shown to outperform traditional EMG PR systems. Nevertheless, the large number of model parameters, long training times, and large amounts of data required to train these DL models remain limiting factors that may hinder their translation into clinically viable prostheses. Consequently, rather than applying DL directly, this paper leverages concepts derived from these models to build upon our proposed concept of a Fusion of Time Domain Descriptors (FTDD). The FTDD are augmented with Range Spatial Filtering (RSF) to capture the spatial correlations and combined into an LSTM-style framework. This process, denoted as Recurrent Spatial-Temporal Fusion (RSTF), can be applied in combination with any traditional feature extraction method to exploit temporal and spatial correlations, with the potential for bi-directional applications. The advantages of the proposed RSTF method include (1) the memory concept, capturing long- and short-term spatial and temporal dependencies of the EMG signals, (2) significantly improved performance outperforming other state-of-the-art models and (3) the simplicity and the fairly low computational costs for feature extraction. Results are bench-marked against several feature extraction methods, proving the power of the RSTF using data from 82 subjects from five EMG databases with varying recording characteristics. The proposed method significantly outperforms all other methods tested for EMG pattern recognition, including a deep LSTM and other CNN methods previously reported in the literature and at a fracture of the computational cost. On the most challenging dataset, improvements of as much as 15% were found.

A Neural Network Based on the Johnson S U Translation System and Related Application to Electromyogram Classification

Electromyogram (EMG) classification is a key technique in EMG-based control systems. Existing EMG classification methods, which do not consider EMG features that have distribution with skewness and kurtosis, have limitations such as the requirement to tune hyperparameters. In this paper, we propose a neural network based on the Johnson SU translation system that is capable of representing distributions with skewness and kurtosis. The Johnson system is a normalizing translation that transforms non-normal distribution data into normal distribution data, thereby enabling the representation of a wide range of distributions. In this study, a discriminative model based on the multivariate Johnson SU translation system is transformed into a linear combination of coefficients and input vectors using log-linearization; then, it is incorporated into a neural network structure. This allows the calculation of the posterior probability of each class given the input vectors and the determination of model parameters as weight coefficients of the network. The uniqueness of convergence of the network learning is theoretically guaranteed. In the experiments, the suitability of the proposed network for distributions including skewness and kurtosis was evaluated using artificially generated data. Its applicability to real biological data was also evaluated via EMG classification experiments. The results showed that the proposed network achieved high classification performance (e.g., 99.973% accuracy using Khushaba’s dataset) without the need for hyperparameter optimization.

A new framework for classification of multi-category hand grasps using EMG signals

Electromyogram (EMG) signals have had a great impact on many applications, including prosthetic or rehabilitation devices, human-machine interactions, clinical and biomedical areas. In recent years, EMG signals have been used as a popular tool to generate device control commands for rehabilitation equipment, such as robotic prostheses. This intention of this study was to design an EMG signal-based expert model for hand-grasp classification that could enhance prosthetic hand movements for people with disabilities. The study, thus, aimed to introduce an innovative framework for recognising hand movements using EMG signals. The proposed framework consists of logarithmic spectrogram-based graph signal (LSGS), AdaBoost k-means (AB-k-means) and an ensemble of feature selection (FS) techniques. First, the LSGS model is applied to analyse and extract the desirable features from EMG signals. Then, to assist in selecting the most influential features, an ensemble FS is added to the design. Finally, in the classification phase, a novel classification model, named AB-k-means, is developed to classify the selected EMG features into different hand grasps. The proposed hybrid model, LSGS-based scheme is evaluated with a publicly available EMG hand movement dataset from the UCI repository. Using the same dataset, the LSGS-AB-k-means design model is also benchmarked with several classifications including the state-of-the-art algorithms. The results demonstrate that the proposed model achieves a high classification rate and demonstrates superior results compared to several previous research works. This study, therefore, establishes that the proposed model can accurately classify EMG hand grasps and can be implemented as a control unit with low cost and a high classification rate.

EMG-based Estimation of Wrist Motion Using Polynomial Models.

Myoelectric control is a method of decoding the motor intent from the electromyogram (EMG) data and using the estimated intent to control prostheses and robots. This work investigates estimation of the wrist kinematics from EMG signals using polynomial models. Due to their low complexity, polynomial models are potentially the perfect choice for EMG-kinematics modeling. Ten ablebodied individuals participated in this study, where the EMG signals from the forearm and the wrist kinematics from the contralateral wrist were measured during mirrored contractions. Two sets of EMG features were employed including the time domain (TD) set, and TD features along with autoregressive coefficients (TDAR). Polynomial models of order 1 to 4 were applied to map the EMG signals to the wrist motions. The performance was directly compared to that of a multilayer perceptron (MLP) neural network. The estimation accuracy of the wrist kinematics improved with increasing the order of the model, but saturated at the 4th order. When using the TD set, the MLP significantly outperformed all polynomial models. However, when using the TDAR set, the polynomial models' performance improved so that the 4th order model performance was not significantly different than that of the MLP in two DoFs, although it was lower than MLP in one DoF. These results indicate that polynomial models are not as effective as more complex models such as neural networks, in learning the highly nonlinear mapping between the EMG data and motion intent. However, using a sufficiently high number of various EMG features, would reduce the mapping nonlinearities, and thereby may increase the polynomial models' performance to levels similar to those of complex black box models.

Dimensionality Reduction in EMG-Based Estimation of Wrist Kinematics.

Pattern recognition has shown remarkable success in decoding motor information from electromyogram (EMG) signals. To decrease the computational complexity in EMG pattern recognition, it may be useful to reduce the dimensionality of the model input. This paper investigates the effect of reducing the dimensionality of EMG features in a regression-based motion intent estimation model. Ten able-bodied subjects participated in this analytic study. EMG signals from the right forearm and angle of the left wrist in three degrees of freedom (DoF) were measured, concurrently. The TD features were extracted from eight EMG channels, resulting in a total of 32 features. Three dimensionality reduction methods including principal component analysis (PCA), non-negative matrix factorization (NNMF), and canonical correlation analysis (CCA) were applied to the EMG features. Reducing the dimension of the EMG features below a certain threshold degraded the performance of the EMG pattern recognition model. Otherwise, dimensionality reduction did not change the performance. These thresholds for the PCA, NNMF, and CCA methods were 25, 26, and 13, respectively. Based on the results, CCA substantially outperformed PCA and NNMF, as it allowed a significant reduction of the EMG features size, from 32 to 13, with no adverse impact on the performance.

Wearable EMG-Based Gesture Recognition Systems During Activities of Daily Living: An Exploratory Study.

Recent advancements in wearable technologies have increased the potential for practical gesture recognition systems using electromyogram (EMG) signals. However, despite the high classification accuracies reported in many studies (> 90%), there is a gap between academic results and industrial success. This is in part because state-of-the-art EMG-based gesture recognition systems are commonly evaluated in highly-controlled laboratory environments, where users are assumed to be resting and performing one of a closed set of target gestures. In real world conditions, however, a variety of non-target gestures are performed during activities of daily living (ADLs), resulting in many false positive activations. In this study, the effect of ADLs on the performance of EMG-based gesture recognition using a wearable EMG device was investigated. EMG data for 14 hand and finger gestures, as well as continuous activity during uncontrolled ADLs (>10 hours in total) were collected and analyzed. Results showed that (1) the cluster separability of 14 different gestures during ADLs was 171 times worse than during rest; (2) the probability distributions of EMG features extracted from different ADLs were significantly different (p <; 0.05). (3) of the 14 target gestures, a right angle gesture (extension of the thumb and index finger) was least often inadvertently activated during ADLs. These results suggest that ADLs and other non-trained gestures must be taken into consideration when designing EMG-based gesture recognition systems.

Adaptive robot mediated upper limb training using electromyogram-based muscle fatigue indicators.

Studies on improving the adaptability of upper limb rehabilitation training do not often consider the implications of muscle fatigue sufficiently. In this study, electromyogram features were used as fatigue indicators in the context of human-robot interaction. They were utilised for auto-adaptation of the task difficulty, which resulted in a prolonged training interaction. The electromyogram data was collected from three gross-muscles of the upper limb in 30 healthy participants. The experiment followed a protocol for increasing the muscle strength by progressive strength training, that was an implementation of a known method in sports science for muscle training, in a new domain of robotic adaptation in muscle training. The study also compared how the participants in three experimental conditions perceived the change in task difficulty levels. One task benefitted from robotic adaptation (Intervention group) where the robot adjusted the task difficulty. The other two tasks were control groups 1 and 2. There was no difficulty adjustment at all in Control 1 group and the difficulty was adjusted manually in Control 2 group. The results indicated that the participants could perform a prolonged progressive strength training exercise with more repetitions with the help of a fatigue-based robotic adaptation, compared to the training interactions, which were based on manual/no adaptation. This study showed that it is possible to alter the level of the challenge using fatigue indicators, and thus, increase the interaction time. The results of the study are expected to be extended to stroke patients in the future by utilising the potential for adapting the training difficulty according to the patient’s muscular state, and also to have a large number repetitions in a robot-assisted training environment.

Identification of long thoracic nerve on high-resolution 3T MRI

ObjectiveLong thoracic neuropathy results in serratus anterior muscle denervation and presents with scapular winging. Previously published studies have been unable to identify the long thoracic nerve on MRI; instead, secondary imaging features of serratus anterior muscle denervation are used to infer nerve injury. Our study's purpose was to evaluate the ability of high-resolution MRI to depict the long thoracic nerve. Materials and methodsIn this HIPAA-compliant, IRB-approved retrospective study, two musculoskeletal radiologists reviewed the brachial plexus MRI exans of 24 subjects performed for clinical suspicion of long thoracic neuropathy. The radiologists evaluated whether the long thoracic nerve could be identified and for the presence of serratus anterior denervation; when the nerve was seen, assessment for nerve enlargement, signal hyperintensity, and morphologic change was performed. Inter-observer reliability was estimated with Cohen's kappa (κ). Clinical presentation and electromyogram (EMG) were then reviewed. ResultsThe long thoracic nerve was identified in 18 cases (75%), with high inter-observer reliability for nerve visualization. Kappa values of 1.0, 0.9, 1.0, and 0.9 were obtained for identification of the LTN on coronal sequences of the brachial plexus, identification of the LTN on proximal and mid segments of the nerve on oblique sagittal sequences, and identification of the distal segment of the LTN on axial sequences through the chest, respectively. The nerve was identified in 91% of patients with positive EMG findings for a long thoracic neuropathy. In patients with EMG features of long thoracic neuropathy, 70% had corresponding abnormal MRI features. When denervation edema was present, the nerve was identified 86% of the time. ConclusionHigh-resolution MRI can be used to visualize segments of the long thoracic nerve.

Insights into paradoxical (REM) sleep homeostatic regulation in mice using an innovative automated sleep deprivation method.

Identifying the precise neuronal networks activated during paradoxical sleep (PS, also called REM sleep) has been a challenge since its discovery. Similarly, our understanding of the homeostatic mechanisms regulating PS, whether through external modulation by circadian and ultradian drives or via intrinsic homeostatic regulation, is still limited, largely due to interfering factors rendering the investigation difficult. Indeed, none of the studies published so far were able to manipulate PS without significantly altering slow-wave sleep and/or stress level, thus introducing a potential bias in the analyses. With the aim of achieving a better understanding of PS homeostasis, we developed a new method based on automated scoring of vigilance states-using electroencephalogram and electromyogram features-and which involves closed-loop PS deprivation through the induction of cage floor movements when PS is detected. Vigilance states were analyzed during 6 and 48 h of PS deprivation as well as their following recovery periods. Using this new automated methodology, we were able to deprive mice of PS with high efficiency and specificity, for short or longer periods of time, observing no sign of stress (as evaluated by plasma corticosterone level and sleep latency) and requiring no human intervention or environmental changes. We show here that PS can be homeostatically modulated and regulated while no significant changes are induced on slow-wave sleep and wakefulness, with a PS rebound duration depending on the amount of prior PS deficit. We also show that PS interval duration is not correlated with prior PS episode duration in the context of recovery from PS deprivation.

Finger Movement Discrimination of EMG Signals Towards Improved Prosthetic Control using TFD

Prosthetic is an artificially made as a substitute or replacement for missing part of a body. The function of the missing body part can be replaced by using the prosthesis and it can help disabled people do their activities easily. A myoelectric control system is a fundamental part of modern prostheses. The electromyogram (EMG) signals are used in this system to control the prosthesis movements by taking it from a person’s muscle. The problem for the myoelectric control system is when it did not receive the same attention to control fingers due to more dexterous of individual and combined finger control in a signal. Thus, a method to solve the problem of the myoelectric control system by using time-frequency distribution (TFD) is proposed in this paper. The EMG features of the individual and combine finger movements for ten subjects and ten different movements is extracted using TFD, ie. spectrogram. Three machine learning algorithms which are Support Vector Machine (SVM), k-Nearest Neighbor (KNN) and Ensemble Classifier are then used to classify the individuals and combine finger movement based on the extracted EMG feature from the spectrogram. The performance of the proposed method is then verified using classification accuracy. Based on the results, the overall accuracy for the classification is 90% (SVM), 100% (KNN) and 100% (Ensemble Classifier), respectively. The finding of the study could serve as an insight to improve the conventional prosthetic control strategies.

Influence of muscle fatigue on electromyogram-kinematic correlation during robot-assisted upper limb training.

IntroductionStudies on adaptive robot-assisted upper limb training interactions do not often consider the implications of muscle fatigue sufficiently.MethodsTo explore this, we initially assessed muscle fatigue in 10 healthy subjects using two electromyogram features, namely average power and median power frequency, during an assist-as-needed interaction with HapticMaster robot. Since robotic assistance resulted in a variable fatigue profile across participants, a completely tiring experiment, without a robot in the loop, was also designed to confirm the results.ResultsA significant increase in average power and a decrease in median frequency were observed in the most active muscles. Average power in the frequency band of 0.8–2.5 Hz and median frequency in the band of 20–450 Hz are potential fatigue indicators. Also, comparing the Spearman’s correlation coefficients (between the electromyogram average power and the kinematic force) across trials indicated that correlation was reduced as individual muscles were fatigued.ConclusionsConfirming fatigue indicators, this study concludes that robotic assistance based on user’s performance resulted in lesser muscle fatigue, which caused an increase in electromyogram–force correlation. We now intend to utilise the electromyogram and kinematic features for auto-adaptation of therapeutic human–robot interactions.

Investigation of Fatigue Using Different EMG Features.

Rehabilitative exercise for people suffering from upper limb impairments has the potential to improve their neuro-plasticity due to repetitive training. Our study investigates the usefulness of Electroencephalogram and Electromyogram (EMG) signals for incorporation in humanrobot interaction loop. Twenty healthy participants recruited who performed a series of physical and cognitive tasks, with an inherent fatiguing component in those tasks. Here we report observed effects on EMG signals. Participants performed a Biceps curl repetitions using a suitable dumbbell in three phases. In phase 1, the initial weight was set to achieve maximum voluntary contraction (MVC). Phase 2 followed with 80 % MVC and phase 3 had 60% MVC. After each phase, they had a break around 3 minutes. EMG data were acquired from Biceps, Triceps, and Brachioradialis muscles. Different EMG features were explored to inform on muscle fatigue during this interaction. Comparing EMG during the first and last dumbbell of each phase demonstrated that the muscle fatigue had caused an increase in the average power (94% of cases) and amplitude (91%) and a decrease in the mean (80%) and the median frequency (57%) of EMG, which was more noticeable in Biceps. The results from integrated EMG showed a continuous rise in all three muscles which was more pronounced in Biceps muscle. Given these results, we identify EMG average power as the most reliable feature for informing on muscle fatigue.