NNERVE: Neural Network Extraction of Repetitive Vectors forElectromyography. I. Algorithm

Artificial neural network (ANN) based signal processing methods have been shown to have significant robustness in processing complex, degraded, noisy, and unstable signals. A novel approach to automated electromyogram (EMG) signal decomposition, using an ANN processing architecture, is presented here. Due to the lack of a priori knowledge of motor unit action potential (MUAP) morphology, the EMG decomposition must be performed in an unsupervised manner. An ANN classifier, consisting of a multilayer perceptron neural network and employing a novel unsupervised training strategy, is proposed. The ANN learns repetitive appearances of MUAP waveforms from their suspected occurrences in a filtered EMG signal in an autoassociative learning task. The same training waveforms are fed into the trained ANN and the output of the ANN is fed back to its input, giving rise to a dynamic retrieval net classifier. For each waveform in the data, the network discovers a feature vector associated with that waveform. For each waveform, classification is achieved by comparing its feature vector with those of the other waveforms. Firing information of each MUAP is further used to refine the classification results of the ANN classifier. Then, individual MUAP waveform shapes are derived and their firing tables are created     

A software package for the decomposition of long-term multichannel EMG signals using wavelet coefficients

This paper presents a method to decompose multichannel long-term intramuscular electromyogram (EMG) signals. In contrast to existing decomposition methods which only support short registration periods or single-channel recordings of signals of constant muscle effort, the decomposition software EMG-LODEC (ElectroMyoGram LOng-term DEComposition) is especially designed for multichannel long-term recordings of signals of slight muscle movements. A wavelet-based, hierarchical cluster analysis algorithm estimates the number of classes [motor units (MUs)], distinguishes single MUAPs from superpositions, and sets up the shape of the template for each class. Using three channels and a weighted averaging method to track action potential (AP) shape changes improve the analysis. In the last step, nonclassified segments, i.e., segments containing superimposed APs, are decomposed into their units using class-mean signals. Based on experiments on simulated and long-term recorded EMG signals, our software is capable of providing reliable decompositions with satisfying accuracy. EMG-LODEC is suitable for the study of MU discharge patterns and recruitment order in healthy subjects and patients during long-term measurements.     

A model for the generation of synthetic intramuscular EMG signals to test decomposition algorithms

As more and more intramuscular electromyogram (EMG) decomposition programs are being developed, there is a growing need for evaluating and comparing their performances. One way to achieve this goal is to generate synthetic EMG signals having known features. Features of interest are: the number of channels acquired (number of detection surfaces), the number of detected motor unit action potential (MUAP) trains, their time-varying firing rates, the degree of shape similarity among MUAPs belonging to the same motor unit (MU) or to different MUs, the degree of MUAP superposition, the MU activation intervals, the amount and type of additive noise. A model is proposed to generate one or more channels of intramuscular EMG starting from a library of real MUAPs represented in a 16-dimensional space using their Associated Hermite expansion. The MUAP shapes, regularity of repetition rate, degree of superposition, activation intervals, etc. may be time variable and are described quantitatively by a number of parameters which define a stochastic process (the model) with known statistical features. The desired amount of noise may be added to the synthetic signal which may then be processed by the decomposition algorithm under test to evaluate its capability of recovering the signal features.     

Gesture-based control and EMG decomposition

This paper presents two probabilistic developments for the use with electromyograms (EMGs). First described is a neuroelectric interface for virtual device control based on gesture recognition. The second development is a Bayesian method for decomposing EMGs into individual motor unit action potentials (MUAPs). This Bayesian decomposition method allows for distinguishing individual muscle groups with the goal of enhancing gesture recognition. All examples presented rely upon sampling EMG data from a subject's forearm. The gesture-based recognition uses pattern recognition software that has been trained to identify gestures from among a given set of gestures. The pattern recognition software consists of hidden Markov models, which are used to recognize the gestures as they are being performed in real time from moving averages of EMGs. Two experiments were conducted to examine the feasibility of this interface technology. The first replicated a virtual joystick interface, and the second replicated a keyboard. Moving averages of EMGs do not provide an easy distinction between fine muscle groups. To better distinguish between different fine motor skill muscle groups, we present a Bayesian algorithm to separate surface EMGs into representative MUAPs. The algorithm is based on differential variable component analysis, which was originally developed for electroencephalograms. The algorithm uses a simple forward model representing a mixture of MUAPs as seen across multiple channels. The parameters of this model are iteratively optimized for each component. Results are presented on both synthetic and experimental EMG data. The synthetic case has additive white noise and is compared with known components. The experimental EMG data were obtained using a custom linear electrode array designed for this study.     

Application of higher order statistics techniques to EMG signals to characterize the motor unit action potential

The electromyographic (EMG) signal provides information about the performance of muscles and nerves. At any instant, the shape of the muscle signal, motor unit action potential (MUAP), is constant unless there is movement of the position of the electrode or biochemical changes in the muscle due to changes in contraction level. The rate of neuron pulses, whose exact times of occurrence are random in nature, is related to the time duration and force of a muscle contraction. The EMG signal can be modeled as the output signal of a filtered impulse process where the neuron firing pulses are assumed to be the input of a system whose transfer function is the motor unit action potential. Representing the neuron pulses as a point process with random times of occurrence, the higher order statistics based system reconstruction algorithm can be applied to the EMG signal to characterize the motor unit action potential. In this paper, we report results from applying a cepstrum of bispectrum based system reconstruction algorithm to real wired-EMG (wEMG) and surface-EMG (sEMG) signals to estimate the appearance of MUAPs in the Rectus Femoris and Vastus Lateralis muscles while the muscles are at rest and in six other contraction positions. It is observed that the appearance of MUAPs estimated from any EMG (wEMG or sEMG) signal clearly shows evidence of motor unit recruitment and crosstalk, if any, due to activity in neighboring muscles. It is also found that the shape of MUAPs remains the same on loading.     

Recognition of Electromyographic Signals Using Cascaded Kernel Learning Machine

Electromyographic (EMG) signals recognition is a complex pattern recognition problem due to its property of large variations in signals and features. This paper proposes a novel EMG classifier called cascaded kernel learning machine (CKLM) to achieve the goal of high-accuracy EMG recognition. First, the EMG signals are acquired by three surface electrodes placed on three different muscles. Second, EMG features are extracted by autoregressive model (ARM) and EMG histogram. After the feature extraction, the CKLM is performed to classify the features. CKLM is composed of two different kinds of kernel learning machines: generalized discriminant analysis (GDA) algorithm and support vector machine (SVM). By using GDA, both the goals of the dimensionality reduction of input features and the selection of discriminating features, named kernel FisherEMG, can be reached. Then, SVM combined with one-against-one strategy is executed to classify the kernel FisherEMG. By cascading SVM with GDA, the input features will be nonlinearly mapped twice by radial-basis function (RBF). As a result, a linear optimal separating hyperplane can be found with the largest margin of separation between each pair of postures' classes in the implicit dot product feature space. In addition, we develop a digital signal processor (DSP)-based EMG classification system for the control of a multi-degrees-of-freedom prosthetic hand for the practical implementation. Based on the clinical experiments, the results show that the proposed CKLM is superior to other frequently used methods, such as k-nearest neighbor algorithm, multilayer neural network, and SVM. The best EMG recognition rate 93.54% is obtained by CKLM.     

Automated decomposition of intramuscular electromyographic signals

We present a novel method for extracting and classifying motor unit action potentials (MUAPs) from one-channel electromyographic recordings. The extraction of MUAP templates is carried out using a symbolic representation of waveforms, a common technique in signature verification applications. The assignment of MUAPs to their specific trains is achieved by means of repeated template matching passes using pseudocorrelation, a new matched-filter-based similarity measure. Identified MUAPs are peeled off and the residual signal is analyzed using shortened templates to facilitate the resolution of superimpositions. The program was tested with simulated data and with experimental signals obtained using fine-wire electrodes in the biceps brachii during isometric contractions ranging from 5% to 30% of the maximum voluntary contraction. Analyzed signals were made of up to 14 MUAP trains. Most templates were extracted automatically, but complex signals sometimes required the adjustment of 2 parameters to account for all the MUAP trains present. Classification accuracy rates for simulations ranged from an average of 96.3% +/- 0.9% (4 trains) to 75.6% +/- 11.0% (12 trains). The classification portion of the program never required user intervention. Decomposition of most 10-s-long signals required less than 10 s using a conventional desktop computer, thus showing capabilities for real-time applications.     

Automatic decomposition of selective needle-detected myoelectricsignals

A procedure for the storage and documentation of myoelectric signals has been developed that consists of a selective needle signal detection protocol, a data collection-compression routine, an adaptive signal decomposition algorithm, and an error filter. The collection-compression routine stores only fixed-length signal epochs that contain motor unit action potentials (MUAPs) detected during individual motor unit firings. The decomposition algorithm assigns the collected MUAPs to candidate motor units, based on template matching using power-spectrum domain features and firing-time criteria calculated from the motor units' firing statistics. Power spectrum features allow the use of Nyquist sampling rates and remove the need for template alignment. The algorithm is adaptive and attempts to minimize dependent errors. The error filter, using firing statistics, accounts for unresolved superpositions and other decomposition errors. Using a standard TECA single-fiber needle electrode, signal recorded during isometric, constant, or slow force-varying contractions of up to 50% of the maximal voluntary contraction level, have been successfully analyzed     

Selectivity of spatial filters for surface EMG detection from the tibialis anterior muscle

Many spatial filters have been proposed for surface electromyographic (EMG) signal detection. Although theoretical and modeling predictions on spatial selectivity are available, there are no extensive experimental validations of these techniques based on single motor unit (MU) activity detection. The aim of this study was to compare spatial selectivity of one- and two-dimensional (1-D and 2-D) spatial filters for EMG signal detection. Intramuscular and surface EMG signals were recorded from the tibialis anterior muscle of ten subjects. The simultaneous use of intramuscular wire and surface recordings (with the spike triggered averaging technique) allowed investigation of the activity of single MUs at the skin surface. The surface EMG signals were recorded with a grid of point electrodes (3 x 3 electrodes) and a ring electrode system at 15 locations over the muscle, with the wires detecting signals from the same intramuscular location. For most subjects, it was possible to classify, from the intramuscular recordings, the activity of the same MUs for all the contractions. The surface EMG signals were averaged with the intramuscularly detected MU action potentials as triggers. In this way, eight spatial filters--longitudinal and transversal, single and double differential (LSD, TSD, LDD, TDD), Laplacian (NDD), inverse binomial filter of the second order (IB2), inverse rectangle filter (IR), and differential ring system (C1)--could be compared on the basis of their spatial selectivity. The distance from the source (transversal with respect to the muscle fiber orientation) after which the surface detected potential did not exceed +/- 5% of the maximal peak-to-peak amplitude (detection distance) was statistically smaller for the 2-D systems and TDD than for the other filters. The MU action potential duration was significantly shorter with LDD and with the 2-D systems than with the other filters. The 2-D filters investigated (including C1) showed very similar performance and were, thus, considered equivalent from the point of view of spatial selectivity.     

A genetic algorithm for the resolution of superimposed motor unit action potentials

This paper presents a novel method, which aims at resolving difficult superimpositions of motor unit action potentials (MUAPs) obtained from single-channel intramuscular electromyographic recordings. Resolution is achieved by means of a genetic algorithm (GA) combined with a gradient descent method. This dual optimization scheme has been tested by means of simulations of isolated superimpositions involving two to six MUAPs, along with simulated extended signals of 10-s duration where the density reached 300 MUAPs/s. Of the hundreds of isolated superimpositions tested, more than 90% of the MUAPs were positively identified. With extended signals, identification rates of better than 85% were obtained. The GA alone accounted for up to an 8% improvement over the decomposition conducted using only template matching.     

Intelligent hybrid approaches for human ECG signals identification

This paper presents hybrid approaches for human identification based on electrocardiogram (ECG). The proposed approaches consist of four phases, namely data acquisition, preprocessing, feature extraction and classification. In the first phase, data acquisition phase, data sets are collected from two different databases, ECG-ID and MIT-BIH Arrhythmia database. In the second phase, noise reduction of ECG signals is performed by using wavelet transform and a series of filters used for de-noising. In the third phase, features are obtained by using three different intelligent approaches: a non-fiducial, fiducial and a fusion approach between them. In the last phase, the classification approach, three classifiers are developed to classify subjects. The first classifier is based on artificial neural network (ANN). The second classifier is based on K-nearest neighbor (KNN), relying on Euclidean distance. The last classifier is support vector machine (SVM) classification accuracy of 95% is obtained for ANN, 98 % for KNN and 99% for SVM on the ECG-ID database, while 100% is obtained for ANN, KNN, and SVM on MIT-BIH Arrhythmia database. The results show that the proposed approaches are robust and effective compared with other recent works.     

A mathematical morphology approach to Euclidean distancetransformation

A distance transformation technique for a binary digital image using a gray-scale mathematical morphology approach is presented. Applying well-developed decomposition properties of mathematical morphology, one can significantly reduce the tremendous cost of global operations to that of small neighborhood operations suitable for parallel pipelined computers. First, the distance transformation using mathematical morphology is developed. Then several approximations of the Euclidean distance are discussed. The decomposition of the Euclidean distance structuring element is presented. The decomposition technique employs a set of 3 by 3 gray scale morphological erosions with suitable weighted structuring elements and combines the outputs using the minimum operator. Real-valued distance transformations are considered during the processes and the result is approximated to the closest integer in the final output image.     

Nonlinear Spatial Filtering of Multichannel Surface Electromyogram Signals During Low Force Contractions

This study introduces the application of nonlinear spatial filters to help identify single motor unit discharge from multiple channel surface electromyogram (EMG) signals during low force contractions. The nonlinear spatial filters simultaneously take into account the instantaneous amplitude and frequency information of a signal. This property was used to enhance motor unit action potentials (MUAPs) in the surface EMG record. The advantages of nonlinear spatial filtering for surface MUAP enhancement were investigated using both simulation and experimental approaches. The simulation results indicate that when compared with various linear spatial filters, nonlinear spatial filtering achieved higher SNR and higher kurtosis of the surface EMG distribution. Over a broad range of SNR and kurtosis levels for the input signal, nonlinear spatial filters achieved at least 32 times greater SNR and 11% higher kurtosis for correlated noise, and at least 15 times greater SNR and 1.7 times higher kurtosis for independent noise, across electrode array channels. The improvements offered by nonlinear spatial filters were further documented by applying them to experimental surface EMG array recordings. Compared with linear spatial filters, nonlinear spatial filters achieved at least nine times greater SNR and 25% higher kurtosis. It follows that nonlinear spatial filters represent a potentially useful supplement to linear spatial filters for detection of motor unit activity in surface EMG at low force contractions.     

Decomposition of multiunit electromyographic signals

We have developed a comprehensive technique to identify single motor unit (SMU) potentials and to decompose overlapped electromyographic (EMG) signals into their constituent SMU potentials. This technique is based on one-channel EMG recordings and is easily implemented for many clinical EMG tests. There are several distinct features of our technique: 1) it measures waveform similarity of SMU potentials in the wavelet domain, which gives this technique significant advantages over other techniques; 2) it classifies spikes based on the nearest neighboring algorithm, which is less sensitive to waveform variation; 3) it can effectively separate compound potentials based on a maximum signal energy deduction algorithm, which is fast and relatively reliable; and 4) it also utilizes the information on discharge regularities of SMU's to help correct possible decomposition errors. The performance of this technique has been evaluated by using simulated EMG signals composed of up to eight different discharging SMU's corrupted with white noise, and also by using real EMG signals recorded at levels up to 50% maximum voluntary contraction. We believe that it is a very useful technique to study SMU discharge patterns and recruitment of motor units in patients with neuromuscular disorders in clinical EMG laboratories.     

加权联合降维的深度特征提取与分类识别算法

为了降低卷积神经网络计算的复杂度,改善特征提取过程中的过拟合现象,解决经典网络模型不能有效处理大尺寸图片的问题,采用了加权联合降维的特征融合与分类识别算法,根据两特征的识别贡献率对主成分分析法（PCA）降维处理和随机投影（RP）处理结果进行加权融合,然后将结果提供给卷积神经网络进行处理,提取图像分类的高层特征,使用欧氏距离分类器对识别对象进行分类,并进行了理论分析和实验验证。结果表明,经过加权联合降维对数据进行预处理,PCA矩阵与RP降维矩阵之比重达到6:4,识别率高达96%以上。该算法有效提高了准确率,使大尺寸图片在深度学习网络中有良好的识别效果,改善了网络的适应性。     

Singularity characteristics of needle EMG IP signals

Clinical electromyography (EMG) interference pattern (IP) signals can reveal more diagnostic information than their constituents, the motor unit action potentials (MUAPs). Singularities and irregular structures typically characterize the mathematically defined content of information in signals. In this paper, a wavelet transform method is used to detect and quantify the singularity characteristics of EMG IP signals using the Lipschitz exponent (LE) and measures derived from it. The performance of the method is assessed in terms of its ability to discriminate healthy, myopathic and neuropathic subjects and how it compares with traditionally used Turns Analysis (TA) methods and a method recently developed by the authors, interscale wavelet maximum (ISWM). Highly significant intergroup differences were found using the LE method. Most of the singularity measures have a performance similar to that of ISWM and considerably better than that of TA. Some measures such as the ratio of the mean LE value to the number of singular points in the signal have considerably superior performance to both methods. These findings add weight to the view that wavelet analysis methods offer an effective way forward in the quantitative analysis of EMG IP signal to assist the clinician in the diagnosis of neuromuscular disorders.     

Late potential recognition by artificial neural networks

Ventricular late potentials (LPs) are high-frequency low-amplitude signals obtained from signal-averaged electrocardiograms (ECGs) [SAECGs]. LPs are useful in identifying patients prone to ventricular tachycardia (VT), spontaneous or inducible during electrophysiology testing. A combination of self-organizing and supervised artificial neural network (ANN) models was developed to identify patients with a positive electrophysiology (PEP) test for inducible ventricular tachycardia from patients with a negative electrophysiology (NEP) test using LPs. We have added morphology information of vector magnitude waveform to an original set of three time-domain features of LPs, which are total QRS duration (TQRSD), high-frequency low-amplitude signal duration (HFLAD), and root-mean-square voltage (RMSV). Pattern recognition results from an ANN model with this combination feature set are superior to the results from Bayesian classification model based on conventional three time-domain features of SAECG. In order to increase the robustness of the recognition, a filtered QRS offset point is randomly shifted ±8 ms to form a fuzzy training set, which was to simulate the possible error in detecting QRS offset point of filtered SAECG. We also found that nonlinear transformation through the hidden layer of developed ANN model could increase Euclidean distance between PEP and NEP patterns     

基于标签重构的弹载毫米波距离像识别算法

毫米波引信通过发射宽带信号获得目标的精细结构信息,然而在非合作目标的探测与识别过程中,由于缺乏目标的类别信息,大量的目标样本无法得到充分的利用。针对这一问题,将基于拉普拉斯得分(LS)的监督特征选择算法推广到半监督情况,得到基于标签重构的拉普拉斯得分算法(LRLS),并应用到非合作目标的识别中。LRLS的理论框架与LS相同,并利用标签重构技术获得半监督情况下的图拉普拉斯矩阵。为了更好地描述高维目标样本的相似性,在标签重构的过程中使用测地距离代替欧氏距离。实验结果表明,相对于传统的特征选择算法,LRLS能够得到更好的识别效果。     

基于模板匹配的雷达辐射源信号识别研究

从统计学理论出发,对基于模板匹配的雷达辐射源信号识别进行研究,通过设定容差测试了4种不同的距离计算方法的识别效果,得出了在雷达特征参数测量精度较高的情况下,使用City Block距离和Euclidean距离进行距离计算识别效果较好,当测量精度不高时,需要增大容差范围且使用M距离进行距离计算时识别效果较好。建立的数据库和雷达辐射源识别模型可用于电子战的模拟训练。     

The application of cepstral coefficients and maximum likelihoodmethod in EMG pattern recognition [movements classification]

A new technique for classifying patterns of movement via electromyographic (EMG) signals is presented. Two methods (conventional autoregressive (AR) coefficients and cepstral coefficients) for extracting features from EMG signals and three classification algorithms (Euclidean Distance Measure (EDM), Weighted Distance Measure (WDM), and Maximum Likelihood Method (MLM)) for discriminating signals representative of broad classes of movements are described and compared. These three classifiers are derived from Bayes classifier with some assumptions, the relationship among them is discussed. The conventional MLM is modified to avoid heavy matrix inversion. Six able-bodied subjects with two pairs of surface electrodes located on bilateral sternocleidomastoid and upper trapezius muscles were studied in the experiment. The EMG signals of 20 repetitions of 10 motions were analyzed for each subject. Experimental results showed that mean recognition rate of the cepstral coefficients was at least 5% superior to that of the AR coefficients. The improvement achieved by the cepstral method was statistically significant for all the three classifiers. Reasons for the superiority of cepstral features were investigated from the feature space and frequency domain, respectively. The cepstral coefficients owned better cluster separability in feature space and they emphasized the more informative part in the frequency domain. The discrimination rate of the MLM was the highest among three classifiers. Incorporation of the cepstral features with the MLM could reduce the misclassification rate by 10.6% when compared with the combination of AR coefficients and EDM. Proper choice of five of ten motions could further raise the recognition rate to more than 95%