Kernel Density Estimation of Electromyographic Signals and Ensemble Learning for Highly Accurate Classification of a Large Set of Hand/Wrist Motions.

The performance of myoelectric control highly depends on the features extracted from surface electromyographic (sEMG) signals. We propose three new sEMG features based on the kernel density estimation. The trimmed mean of density (TMD), the entropy of density, and the trimmed mean absolute value of derivative density were computed for each sEMG channel. These features were tested for the classification of single tasks as well as of two tasks concurrently performed. For single tasks, correlation-based feature selection was used, and the features were then classified using linear discriminant analysis (LDA), non-linear support vector machines, and multi-layer perceptron. The eXtreme gradient boosting (XGBoost) classifier was used for the classification of two movements simultaneously performed. The second and third versions of the Ninapro dataset (conventional control) and Ameri's movement dataset (simultaneous control) were used to test the proposed features. For the Ninapro dataset, the overall accuracy of LDA using the TMD feature was 98.99 ± 1.36% and 92.25 ± 9.48% for able-bodied and amputee subjects, respectively. Using ensemble learning of the three classifiers, the average macro and micro-F-score, macro recall, and precision on the validation sets were 98.23 ± 2.02, 98.32 ± 1.93, 98.32 ± 1.93, and 98.88 ± 1.31%, respectively, for the intact subjects. The movement misclassification percentage was 1.75 ± 1.73 and 3.44 ± 2.23 for the intact subjects and amputees. The proposed features were significantly correlated with the movement classes [Generalized Linear Model (GLM); P-value < 0.05]. An accurate online implementation of the proposed algorithm was also presented. For the simultaneous control, the overall accuracy was 99.71 ± 0.08 and 97.85 ± 0.10 for the XGBoost and LDA classifiers, respectively. The proposed features are thus promising for conventional and simultaneous myoelectric control.

High-density surface electromyography signals during isometric contractions of elbow muscles of healthy humans.

This paper presents a dataset of high-density surface EMG signals (HD-sEMG) designed to study patterns of sEMG spatial distribution over upper limb muscles during voluntary isometric contractions. Twelve healthy subjects performed four different isometric tasks at different effort levels associated with movements of the forearm. Three 2-D electrode arrays were used for recording the myoelectric activity from five upper limb muscles: biceps brachii, triceps brachii, anconeus, brachioradialis, and pronator teres. Technical validation comprised a signals quality assessment from outlier detection algorithms based on supervised and non-supervised classification methods. About 6% of the total number of signals were identified as "bad" channels demonstrating the high quality of the recordings. In addition, spatial and intensity features of HD-sEMG maps for identification of effort type and level, have been formulated in the framework of this database, demonstrating better performance than the traditional time-domain features. The presented database can be used for pattern recognition and MUAP identification among other uses.

Analysis of Muscle Load-Sharing in Patients With Lateral Epicondylitis During Endurance Isokinetic Contractions Using Non-linear Prediction.

The aim of this paper is to analyze muscle load-sharing in patients with Lateral Epicondylitis during dynamic endurance contractions by means of non-linear prediction of surface EMG signals. The proposed non-linear cross-prediction scheme was used to predict the envelope of an EMG signal and is based on locally linear models built in a lag-embedded Euclidean space. The results were compared with a co-activation index, a common measure based on the activation of a muscle pair. Non-linear prediction revealed changes in muscle coupling, that is load-sharing, over time both in a control group and Lateral Epicondylitis (p < 0.05), even when subjects did not report pain at the end of the exercise. These changes were more pronounced in patients, especially in the first part of the exercise and up to 50% of the total endurance time (p < 0.05). By contrast, the co-activation index showed no differences between groups. Results reflect the changing nature of muscular activation strategy, presumably because of the mechanisms triggered by fatigue. Strategies differ between controls and patients, pointing to an altered coordination in Lateral Epicondylitis.

A Novel Spatial Feature for the Identification of Motor Tasks Using High-Density Electromyography.

Estimation of neuromuscular intention using electromyography (EMG) and pattern recognition is still an open problem. One of the reasons is that the pattern-recognition approach is greatly influenced by temporal changes in electromyograms caused by the variations in the conductivity of the skin and/or electrodes, or physiological changes such as muscle fatigue. This paper proposes novel features for task identification extracted from the high-density electromyographic signal (HD-EMG) by applying the mean shift channel selection algorithm evaluated using a simple and fast classifier-linear discriminant analysis. HD-EMG was recorded from eight subjects during four upper-limb isometric motor tasks (flexion/extension, supination/pronation of the forearm) at three different levels of effort. Task and effort level identification showed very high classification rates in all cases. This new feature performed remarkably well particularly in the identification at very low effort levels. This could be a step towards the natural control in everyday applications where a subject could use low levels of effort to achieve motor tasks. Furthermore, it ensures reliable identification even in the presence of myoelectric fatigue and showed robustness to temporal changes in EMG, which could make it suitable in long-term applications.

Prediction of isometric motor tasks and effort levels based on high-density EMG in patients with incomplete spinal cord injury.

OBJECTIVE: The development of modern assistive and rehabilitation devices requires reliable and easy-to-use methods to extract neural information for control of devices. Group-specific pattern recognition identifiers are influenced by inter-subject variability. Based on high-density EMG (HD-EMG) maps, our research group has already shown that inter-subject muscle activation patterns exist in a population of healthy subjects. The aim of this paper is to analyze muscle activation patterns associated with four tasks (flexion/extension of the elbow, and supination/pronation of the forearm) at three different effort levels in a group of patients with incomplete Spinal Cord Injury (iSCI). APPROACH: Muscle activation patterns were evaluated by the automatic identification of these four isometric tasks along with the identification of levels of voluntary contractions. Two types of classifiers were considered in the identification: linear discriminant analysis and support vector machine. MAIN RESULTS: Results show that performance of classification increases when combining features extracted from intensity and spatial information of HD-EMG maps (accuracy = 97.5%). Moreover, when compared to a population with injuries at different levels, a lower variability between activation maps was obtained within a group of patients with similar injury suggesting stronger task-specific and effort-level-specific co-activation patterns, which enable better prediction results. SIGNIFICANCE: Despite the challenge of identifying both the four tasks and the three effort levels in patients with iSCI, promising results were obtained which support the use of HD-EMG features for providing useful information regarding motion and force intention.

Spatial distribution of HD-EMG improves identification of task and force in patients with incomplete spinal cord injury.

BACKGROUND: Recent studies show that spatial distribution of High Density surface EMG maps (HD-EMG) improves the identification of tasks and their corresponding contraction levels. However, in patients with incomplete spinal cord injury (iSCI), some nerves that control muscles are damaged, leaving some muscle parts without an innervation. Therefore, HD-EMG maps in patients with iSCI are affected by the injury and they can be different for every patient. The objective of this study is to investigate the spatial distribution of intensity in HD-EMG recordings to distinguish co-activation patterns for different tasks and effort levels in patients with iSCI. These patterns are evaluated to be used for extraction of motion intention. METHOD: HD-EMG was recorded in patients during four isometric tasks of the forearm at three different effort levels. A linear discriminant classifier based on intensity and spatial features of HD-EMG maps of five upper-limb muscles was used to identify the attempted tasks. Task and force identification were evaluated for each patient individually, and the reliability of the identification was tested with respect to muscle fatigue and time interval between training and identification. RESULTS: Three feature sets were analyzed in the identification: 1) intensity of the HD-EMG map, 2) intensity and center of gravity of HD-EMG maps and 3) intensity of a single differential EMG channel (gold standard). Results show that the combination of intensity and spatial features in classification identifies tasks and effort levels properly (Acc = 98.8 %; S = 92.5 %; P = 93.2 %; SP = 99.4 %) and outperforms significantly the other two feature sets (p < 0.05). CONCLUSION: In spite of the limited motor functionality, a specific co-activation pattern for each patient exists for both intensity, and spatial distribution of myoelectric activity. The spatial distribution is less sensitive than intensity to myoelectric changes that occur due to fatigue, and other time-dependent influences.