Sirinee Thongpanja

The Usefulness of Mean and Median Frequencies in Electromyography Analysis

Rich useful information can be obtained from the muscles and researchers can use such information in a wide class of clinical and engineering applications by measuring surface electromyography (EMG) signals (Merletti & Parker, 2004). Normally, EMG signals are acquired by surface electrodes that are placed on the skin superimposed on the targeted muscle. In order to use the EMG signal as a diagnosis signal or a control signal, a feature is often extracted before performing analysis or classification stage (Phinyomark et al., 2012a) because a lot of information, both useful information and noise (Phinyomark et al., 2012b), is contained in the raw EMG data. An EMG feature is a distinct characteristic of the signal that can be described or observed quantitatively, such as being large or small, spiky or smooth, and fast or slow. Generally, EMG features can be computed in numerical form from a finite length time interval and can change as a function of time, i.e. a voltage or a frequency. They can be computed in several domains, such as time domain, frequency domain, timefrequency and time-scale representations (Boostani & Moradi, 2003). However, frequencydomain features show the better performance than other-domain features in case of the assessing muscle fatigue (Al-Mulla et al., 2012). Mean frequency (MNF) and median frequency (MDF) are the most useful and popular frequency-domain features (Phinyomark et al., 2009) and frequently used for the assessment of muscle fatigue in surface EMG signals (Cifrek et al., 2009).

EMG amplitude estimators based on probability distribution for muscle-computer interface

To develop an advanced muscle–computer interface (MCI) based on surface electromyography (EMG) signal, the amplitude estimations of muscle activities, i.e., root mean square (RMS) and mean absolute value (MAV) are widely used as a convenient and accurate input for a recognition system. Their classification performance is comparable to advanced and high computational time-scale methods, i.e., the wavelet transform. However, the signal-to-noise-ratio (SNR) performance of RMS and MAV depends on a probability density function (PDF) of EMG signals, i.e., Gaussian or Laplacian. The PDF of upper-limb motions associated with EMG signals is still not clear, especially for dynamic muscle contraction. In this paper, the EMG PDF is investigated based on surface EMG recorded during finger, hand, wrist and forearm motions. The results show that on average the experimental EMG PDF is closer to a Laplacian density, particularly for male subject and flexor muscle. For the amplitude estimation, MAV has a higher SNR, defined as the mean feature divided by its fluctuation, than RMS. Due to a same discrimination of RMS and MAV in feature space, MAV is recommended to be used as a suitable EMG amplitude estimator for EMG-based MCIs.

Time-Dependent EMG Power Spectrum Parameters of Biceps Brachii during Cyclic Dynamic Contraction

Mean frequency and median frequency (MNF and MDF) features are global used parameters of EMG power spectrum to determine muscle fatigue. A major problem of these parameters is a non-linear relationship between muscle load and feature value, especially in large muscle and in cyclic dynamic contraction. To analyze the EMG signal in both of muscle fatigue and muscle load, we have been proposed time dependence of the MNF and MDF (TD-MNF and TD-MDF) of a time-sequential data. Moreover, the surface EMG signals that were used in this study were acquired from the biceps brachii muscle during round-trip dynamic contraction. After that TD-MNF and TD-MDF were calculated and compared with the standard MNF and MDF features which were calcu- lated based on the whole data. Three statistical parameters including mean, median, and variance were used to apply with the selected efficient range of TD-MNF and TD-MDF in order to easily observe and use in an application. The result shows that mean parameter of selected TD-MNF have a better linear relationship with muscle load compared to the others and have a significant difference (p<0.005) between feature values for different loading conditions. In addition, it was found that there was a certain pattern of TD-MNF and TD-MDF for each data and each subject that has not been found in traditional MNF and MDF features. The optimal method of TD-MNF and TD-MDF was success when overlapping consecutive window was performed with 512-sample window size and 64-sample window increment. In conclusion, mean of the selected TD- MNF band can be used as both muscle load and muscle fatigue indexes.

Probability Density Functions of Stationary Surface EMG Signals in Noisy Environments

The probability density function (pdf) of an electromyography (EMG) signal provides useful information for choosing an appropriate feature extraction technique. The pdf is influenced by many factors, including the level of contraction force, muscle type, and noise. In this paper, we investigated the pdfs of noisy EMG signals artificially contaminated with five different noise types: 1) Electrocardiography (ECG) interference; 2) many spurious background spikes; 3) white Gaussian noise; 4) motion artifact; and 5) power line interference at various levels of signal-to-noise ratio (SNR). In addition, we evaluated a set of statistical descriptors for identifying a noisy EMG signal from its pdf, specifically kurtosis, negentropy, L-kurtosis, and robust measures of kurtosis (KR1 and KR2). The results show that at low SNR (<;5 dB), all noise types affect the statistical descriptors for the pdf of a noisy EMG signal. In addition, KR2 performs the best among these descriptors in identifying a noisy EMG signal from its pdf, because it is computed based on the quantiles of the data. As a result, it can avoid the effects of outliers resulting in the correct identification of pdf shape of noisy EMGs with all contamination types and all levels of SNR.

Application of Mean and Median Frequency Methods for Identification of Human Joint Angles Using EMG Signal

The analysis of surface electromyography (EMG) signals is generally based on three major issues, i.e., the detection of muscle force, muscle geometry, and muscle fatigue. Recently, there are no any techniques that can analyse all the issues. Mean frequency (MNF) and median frequency (MDF) have been successfully applied to be used as muscle force and fatigue indices in previous studies. However, there is the lack of consensus upon the effect of muscle geometry on the basis of varying joint angles. In this paper, the modification of MNF and MDF using a min-max normalization technique was proposed to provide a consistent relationship between feature value and joint angle across subjects. The results show that MNF and MDF extracted from normalized EMG showed a stronger linear relationship with elbow joint angle compared to traditional MNF and MDF methods. Modified MNF and MDF features increased with increasing elbow angle during isometric flexion. As a result of the proposed technique, modified MNF and MDF features could be used as a universal index to determine all the issues involving muscle fatigue, muscle force, and also muscle geometry.

A robust measure of probability density function of various noises in electromyography (EMG) signal acquisition

Statistical methods for estimating a probability density function (PDF) of surface electromyography (EMG) signals during upper-limb motions have been investigated in previous studies to select the suitable feature extraction methods for multifunction myoelectric control systems. While these methods have achieved a good performance in estimating PDF of EMG signals from different motions and muscles, no prior studies have evaluated the performance of these methods to estimate the PDF of noises in EMG signal acquisition. The utility of these methods consisting of bicoherence, kurtosis, negentropy, and L-kurtosis, was investigated in estimating the PDF of five different noise types: the single and many spurious background spikes, white Gaussian noise, motion artifact, and power line interference. The results show that the L-kurtosis can identify the PDF of all studied noises in EMG signal acquisition correctly. In contrast, other estimating methods are inaccuracy in at least one noise type.

Analysis of Electromyography in Dynamic Hand Motions Using L-Kurtosis

A statistical measure is needed to estimate a probability density function (PDF) of EMG signals to choose the suitable feature extraction methods for EMG pattern recognition system. The utility of L-kurtosis was investigated in estimating the PDFs of three different dynamic EMG involving transient and steady-state signals during four hand motions measured from two forearm muscles, and was compared with the kurtosis. The results show that the L-kurtosis can identify the PDF of EMG for all cases. In contrast, the kurtosis is inaccuracy and less robust when measured EMG signals have a higher amplitude and are more non-stationary during a transient period.

Evaluation of feature extraction techniques and classifiers for finger movement recognition using surface electromyography signal

AbstractElectromyography (EMG) in a bio-driven system is used as a control signal, for driving a hand prosthesis or other wearable assistive devices. Processing to get informative drive signals involves three main modules: preprocessing, dimensionality reduction, and classification. This paper proposes a system for classifying a six-channel EMG signal from 14 finger movements. A feature vector of 66 elements was determined from the six-channel EMG signal for each finger movement. Subsequently, various feature extraction techniques and classifiers were tested and evaluated. We compared the performance of six feature extraction techniques, namely principal component analysis (PCA), linear discriminant analysis (LDA), uncorrelated linear discriminant analysis (ULDA), orthogonal fuzzy neighborhood discriminant analysis (OFNDA), spectral regression linear discriminant analysis (SRLDA), and spectral regression extreme learning machine (SRELM). In addition, we also evaluated the performance of seven classifiers consisting of support vector machine (SVM), linear classifier (LC), naive Bayes (NB), k-nearest neighbors (KNN), radial basis function extreme learning machine (RBF-ELM), adaptive wavelet extreme learning machine (AW-ELM), and neural network (NN). The results showed that the combination of SRELM as the feature extraction technique and NN as the classifier yielded the best classification accuracy of 99%, which was significantly higher than those from the other combinations tested. Graphical abstractMean of classification accuracies for 14 finger movements obtained with various pairs of SRELM and classifier