Machine Learning for sEMG Facial Feature Characterization

Electromyogram (EMG) signals are biopotential signals, which occur as the result of electro-chemical circumstances dependent on contraction of muscles. In this study, a new method is presented to determine the motor unit action potentials of EMG signals recorded with surface electrodes. Compared to needle EMG signals, surface EMG signals have more active noise component. Results, which are gained by the process train including wavelet de-noising method for noise anlysis at first step and then ICA algorithms to provide independent components, form four different datas to determine the motor unit action potentials. Performed study, which is aimed to make ineffective the problems that make the analysis of EMG signals difficult and then to reach the desired results, provides simultaneous comparison possibility submitting four results. After the determination of the results, with the feedback between input and output, examination of accuracy ratio of results are possible and this examination eliminates the questions about the accuracy of the study.

In recent years, there has been extensive interest in the revelation of different technologies for artificial neural network (ANN) application in various classification of biomedical signals. Several publications have focused on diverse applications of ANN techniques for biomedical signals classification, detection, and processing. This review work gives an overview on the multilayer perceptron (MLP) technique of ANN and how this technology is useful for the classification of EMG signal. This paper will be interesting for those researchers who are studying the classification EMG signal for tremor detection in Parkinson’s disease. Many researchers have worked and also working on Parkinson’s disease tremor by using EMG signal and classifying its feature using ANN. Among those various techniques, MLP technique is highlighted in this work for the classification of EMG signal. The core center of this paper is to review the evolution and research works related to the topic mentioned above.

This paper describes the estimation and analysis of variance distribution of surface electromyogram (EMG) signals based on a stochastic EMG model. With the assumption that EMG signals at a certain time follow Gaussian distribution, their variance is handled as a random variable that follows inverse gamma distribution, and noise superimposed onto this variance can be expressed accordingly. The paper proposes variance distribution estimation based on marginal likelihood maximization of EMG signals. A simulation experiment using artificially generated signals to verify its accuracy indicated that the method can estimate variance distribution with high accuracy for a wide range of variance distribution shaping. Analysis of variance distribution using measured EMG signals revealed the relationship between muscle force and variance distribution involving signal-dependent noise.

Electromyogram (EMG) signals have been increasingly used for hand and finger gesture recognition. However, most studies have focused on the wrist and whole-hand gestures and not on individual finger (IF) gestures, which are considered more challenging. In this study, we develop EMG-based hand/finger gesture classifiers based on fixed electrode placement using machine learning methods. Ten healthy subjects performed ten hand/finger gestures, including seven IF gestures. EMG signals were measured from three channels, and six time-domain (TD) features were extracted from each channel. A total of 18 features was used to build personalized classifiers for ten gestures with an artificial neural network (ANN), a support vector machine (SVM), a random forest (RF), and a logistic regression (LR). The ANN, SVM, RF, and LR achieved mean accuracies of 0.940, 0.876, 0.831, and 0.539, respectively. One-way analyses of variance and F-tests showed that the ANN achieved the highest mean accuracy and the lowest inter-subject variance in the accuracy, respectively, suggesting that it was the least affected by individual variability in EMG signals. Using only TD features, we achieved a higher ratio of gestures to channels than other similar studies, suggesting that the proposed method can improve the system usability and reduce the computational burden.

Using electromyogram (EMG) signals as a human-machine interface to control the myoelectric prostheses is consummately advanced. Nevertheless, reported studies present high classification performance from the EMG signal captured in laboratory conditions. There is a very challenging suffix regarding untrained condition. Limb position change causes high variation in EMG signals resulting in high misclassification rate in the untrained position. The goal of this work is to indicate the most relevance EMG channel, which is robust to the effect of a limb position change. The EMG signals from four different object shapes with five different object placement positions were examined using linear discriminant analysis. To determine the best EMG channel subset of all possible combination of selective channels, we reduced the number of EMG channels from 12 to 3. The results have shown that the minimum classification error rate (2.09%) across five subjects can be achieved by use of eight optimal EMG channels. Additionally, our results suggest the EMG channel number 3 (Triceps brachii), 7 (Extensor digitorum communis), and 9 (Brachioradialis) as the most three emphasized channels that contain the rich information of hand grasping and generalize limb position change.

A three-electrode human-computer interaction system, based on digital processing of the Electromyogram (EMG) signal, is presented. This system can effectively help disabled individuals paralyzed from the neck down to interact with computers or communicate with people through computers using point-and-click graphic interfaces. The three electrodes are placed on the right frontalis, the left temporalis and the right temporalis muscles in the head, respectively. The signal processing algorithm used translates the EMG signals during five kinds of facial movements (left jaw clenching, right jaw clenching, eyebrows up, eyebrows down, simultaneous left & right jaw clenching) into five corresponding types of cursor movements (left, right, up, down and left-click), to provide basic mouse control. The classification strategy is based on three principles: the EMG energy of one channel is typically larger than the others during one specific muscle contraction; the spectral characteristics of the EMG signals produced by the frontalis and temporalis muscles during different movements are different; the EMG signals from adjacent channels typically have correlated energy profiles. The algorithm is evaluated on 20 pre-recorded EMG signal sets, using Matlab simulations. The results show that this method provides improvements and is more robust than other previous approaches.

A discriminant bispectrum feature for surface electromyogram signal classification

The aim of this study is to analyze Electromyogram (EMG) signals in Rapid Eye Movement (REM) sleep using different techniques to detect the level of normality and abnormality of normal and abnormal (patients with a lack of REM sleep atonia) subjects and predict the development of Parkinson’s disease in abnormal subjects. Quantitative elctromyogram (EMG) signal analysis in the frequency domain using classical power spectrum analysis techniques have been well documented over the past decade. Yet none of these [sic] work have been done on EMG during Rapid Eye Movement (REM) Stage of sleep. In this work three techniques for classifying chin movement via EMG signals during sleep is presented. Three methods (Autoregressive modeling, Cepstrum Analysis and Wavelet Analysis) for extracting features from EMG signal during sleep and a classification algorithm (Linear Discriminant Analysis (LDA)) were analyzed and compared. EMG data are used to detect and describe different disease processes affecting sleep. Rapid Eye Movement Behavior Disorder (RBD) is an example of EMG abnormality in which patients lose their muscle control while in REM stage of sleep resulting in physically acting out their dreams. An adaptive segmentation based on Recursive Least Square (RLS) algorithm was analyzed. This algorithm was used to segment the non-stationary EMG signal into locally stationary components, which were then autoregressive modeled using the Burg-Lattice method. The cepstral measurements described was used and applied to modify the coefficients computed from the autoregressive (AR) model. Yet due to the nature of the EMG, frequency analysis cannot be used to approximate a signal whose properties change over time. To address this problem a time varying feature representation is necessary for analysis to extract useful infomration from the signal. As a consequence Wavelet coefficients were computed using discrete and continuous wavelet transforms. Furthermore, the classification performance of the above three feature sets were investigated for the two classes (Normal and Abnormal). Results showed wavelet analysis compared to AR modeling and cepstrum analysis is a better assessment in finding EMG abnormalities during sleep. However, these methods may be useful in distinguishing EMG patterns that predict the emergence of Parkinson disease in humans.

The texture of foods is affected by concurrent auditory sensations. To improve the texture of nursing care foods, we developed a pseudo-mastication sound presentation device based on electromyogram (EMG) signals from the muscles of mastication. EMG signals have enabled us to promptly present care recipients with pseudo-mastication sounds. However, actual mastication sounds vary in intensity and duration more than EMG signals. Here, we investigated changes in EMG signals and actual mastication sounds during the mastication of two food types (rice crackers and Japanese pickles) to improve our device. We found that the intensity and duration of mastication sounds decreased as the number of mastication strokes increased. Furthermore, the intensity and duration of mastication sounds and the latency between the onset of EMG signals and the onset of mastication sounds also varied by food type. For EMG signals, only the intensity varied by food type. Based on our findings, we modified our pseudo-mastication sound presentation device to enable control of the intensity and duration of pseudo-mastication sounds based on the number of mastication strokes and food type. Reproducing more natural pseudo-mastication sounds can improve care recipients' motivation for ingesting nursing care foods, thus preventing malnutrition and frailty.

Characterization of Electromyogram (EMG) signals is important for identifying neuromuscular diseases. Although various techniques were implemented for classification, none of them were implemented in the complexity domain. In this study, characterization of EMG signals recorded from healthy, neuropathic and myopathic subjects has been performed in complexity domain based on Multiscale Entropy (MSE) method. To do this, the multiscale entropy method has been applied to an EMG database publicly available in PhysioNet. The complexity profile curves obtained with the MSE approach have shown promising classification among healthy subject, neuropathic and myopathic patients in terms of complexity. One way ANOVA test has shown statistically significant differences (p<0.01) among these three classes. Moreover, Support Vector Machine (SVM) has demonstrated a classification accuracy of 86.1% for characterizing EMG signals of neuromuscular disorders. Furthermore, myopathic and neuropathic patients have been recognized with 66.7% sensitivity at 95.8% specificity and 100% sensitivity at 100% specificity respectively.

The electromyogram (EMG) signal has been used as the command input to myoelectric prostheses. A common control scheme is based on classifying the EMG signals from multiple electrodes into one of several distinct classes of user intent/function. In this work, we investigated the use of EMG whitening as a preprocessing step to EMG pattern recognition. Whitening is known to decorrelate the EMG signal, with improved performance shown in the related applications of EMG amplitude estimation and EMG-torque processing. We reanalyzed the EMG signals recorded from 10 electrodes placed circumferentially around the forearm of 10 intact subjects and 5 amputees. The coefficient of variation of two time-domain features--mean absolute value and signal length--was significantly reduced after whitening. Pre-whitened classification models using these features, along with autoregressive power spectrum coefficients, added approximately five percentage points to their classification accuracy. Improvement was best using smaller window durations (<100 ms).

Electromyogram (EMG) signal is often processed offline, after its acquisition, using digital signal processing algorithms to extract muscle anatomical and physiological information. As most of the signal processing algorithms work on an adequate quality of the signals, thus quality checking of the EMG in real-time during its acquisition is of immense importance. In multi-channel sEMG signals, usually there are some noisy or bad channels. If the noise is of low level, it is of little concern but high level of noise can limit the usefulness of the EMG. To make sure acquisition of a good quality EMG signal in terms of SNR, one way to detect noisy channels is through visual inspection by an expert human operator, however visual inspection of multiple electrodes in real-time is not possible and is also expensive both in terms of time and cost. In this research study, we propose a novel method for automatic detection of noisy channels in multi-channel surface EMG signals based on statistical thresholding of several parameters. The results of the proposed method are in perfect agreement with the ground truth for simulated EMG signals, with an accuracy of 98.6%.

to the editor: The electromyogram (EMG) as an indirect measure of motor unit activity has proved invaluable in motor control studies. Increasingly, the application of frequency domain techniques to EMG signals has provided important insight into the structure and function of the human motor system.

Many people have been exposed to lower extremity function losses due to neurological, pathological or traffic accidents. In the physical therapy and rehabilitation of these patients, treatment programs based on robotic systems have started to be preferred instead of conventional methods. In robotic gait rehabilitation, mobilized lower extremity exoskeletons such as Rewalk or un-mobilized lower extremity exoskeletons such as RoboGait are used. It is important to evaluate the rehabilitation process in patients with lower extremity problems. Measurement of surface electromyogram (EMG) signals during the treatment process give information about the functional activities of the muscles. Obtained information plays an important role in determining the intention of patient motion in musculoskeletal design and musculoskeletal activities of the musculoskeletal. Changes in muscle activation timing and amplitude during the use of lower extremity exoskeleton can be determined by analysis of EMG. In this study, muscles involved in walking movement during robotic rehabilitation were examined. The examined iliopsoas, gluteus maximus, gluteus medius muscles provide flexion, extension and abduction movements of the hip, while the medial gastrocnemius and tibialis anterior muscles perform flexion and dorsiflexion movements of the foot. During the gait, the knee joint patency is controlled by the Vastus Medialis and Biceps Femoris muscles. In this study, while 6 patients with lower limb dysfunction were walking on the RoboGait device, the muscle activation potentials obtained from 7 different muscle groups were transferred to the computer simultaneously and wirelessly and displayed in the Matlab environment. The EMG signals measured with the MicroCor Lab device are shaped according to the activation of the muscles during walking. The electrode placement plan is critical for the analysis of EMG signals, and an appropriate electrode placement plan was obtained as a result of the study. Examined measured signals by following with the electrode placement plan, the maximum gluteus and iliopsoas muscles responsible for the extension and flexion movements of the hips are more effective during walking. Gletous maximum muscle was found to be the most effective muscle in walking while the iliopsoas muscle group was involved in the first movement of the leg. As a result of this study, these findings will help to follow the development of the treatment process and to develop EMG controlled mobilized lower extremity exoskeletons.

The objective of this study is to design a meal assistance system operated by electromyogram (EMG) signals. The system consists of four parts, EMG data acquisition, onset analysis, control circuit and operation of meal assistance robot. An algorithm based on EMG power with adaptive threshold is proposed for onset analysis, and applied to the operation of robot via control circuit. The performance of EMG-control system is tested in rate of success and comfort by both subjective and objective indices. The results show EMG signals acquired from active muscles in a human body can be used as control signals for the apparatus, by the proposed onset detection algorithm with adaptive threshold. The high rate of success and comfort of meal assistance system operated by EMG signals guarantee its use for patients with disabled upper limbs.