Characterizing EMG data using machine-learning tools
Introduction
Electromyography, the study of the electrical currents generated in a muscle during its contraction, provides data describing both neuromuscular activity, as well as muscular morphology [1], [2].
Over the last twenty years, electromyography (EMG) has been widely used by researchers and clinicians as a valuable tool for an accurate diagnosis of neuromuscular disorders [3], [4]. Neuromuscular disorder is a general term that refers to diseases that affect any part of the nerve or muscle including motor neurons, neuromuscular junctions, and muscle tissue. Myopathy and neuropathy are two critical neuromuscular disease types, and discerning between these, as well as between a disease and non-disease state, are typical objectives of a classifier based EMG characterization systems.
Myopathy describes a group of diseases that affect skeletal muscle tissue directly, and are independent from any disorder of the nervous system. Neuropathy, conversely, refers to any of a number of diseases that cause damage to the nerves involved in muscular control, or in sensation [2], [4], [5]. Accurate and correct characterization of these two types of diseases becomes an important first step in the diagnostic process.
While historically EMG data has been approached qualitatively [3], [4], in recent years a great deal of interest has been found in quantitative EMG analysis, called QEMG [6], [7], [8], [9], [10], [11], in which a series of quantitative measures of the EMG signal are analyzed for their diagnostic information, as described in Section 2.3.
To characterize a muscle using QEMG data, the acquired signals must be analyzed, decomposed and classified. In diagnosing neuromuscular disorders, the classification of EMG signal into different groups is used in the detection of abnormalities. This paper presents several classification techniques used for EMG signal classification for diagnosis of neuromuscular disorders, in particular, myopathy and neuropathy types.
In Section 2, we briefly review EMG signals and their attributes. In Section 3, we present a review of EMG classification methods. Discussion is presented in Section 4.
Table 1, Table 2 provide a summary of all the techniques discussed, broken down by the type of classifier used. This is divided into two tables due to page size constraints. Through these tables and the accompanying discussion, both newer results and important historical context are provided, in order to better understand recent trends in EMG classification.
Section snippets
EMG signal analysis
An EMG signal is a biological signal obtained by measuring voltages associated with the electrical currents generated in a muscle during its contraction, providing a measure of neuromuscular activity [1].
EMG classification approaches
Several studies have employed classification methods including Bayesian techniques [26], [27], [28], [29], [30], neural networks [31], [32], multilayer perceptrons [33], fuzzy approaches [34], support vector machines [12], and neuro-fuzzy systems [35]. In the following section, we discuss some of the works which explore the EMG classification problem in the domain of neuromuscular pathology.
All of the techniques discussed in this and later sections are summarized in Table 1, Table 2 which
Discussion
In this paper, a detailed review of the common methods of classifying EMG signals for the diagnosis of neuromuscular disorders have been presented. All of the papers cited are presented in Table 1, Table 2, grouped by classifier type, and annotated by their classification performance and year of publication.
The survey shows that using an ANN is the most popular classification method (measured by the number of implementations), for the classification of EMG signals [12], [31], [33], [36], [37],
Conclusions
The review demonstrates that ANNs and hybrid neural networks play important roles in EMG classification and yield high accuracy results. Back-propagation generally leads to high accuracy results, but it suffers from poor learning-discrimination performance in EMG classification due to the various noise signals. WNNs with sufficient accuracy and speed were found to be one of the most ideal techniques to provide clinically useful information.
One may also note, based on the materials reviewed,
Conflict of interest statement
None declared.
Acknowledgements
The authors gratefully acknowledge the support of NSERC, the National Sciences and Engineering Research Council of Canada, for support through grant 341486.
References (83)
- et al.
Within-subject reliability of motor unit number estimates and quantitative motor unit analysis in a distal and proximal upper limb muscle
Clin. Neurophysiol.
(2006) - et al.
Automatic decomposition of multichannel intramuscular EMG signals
J. Electromyogr. Kinesiol.
(2009) EMG signal decompositionhow can it be accomplished and used?
J. Electromyogr. Kinesiol.
(2001)Decomposition and quantitative analysis of clinical electromyographic signals
Med. Eng. Phys.
(1999)- et al.
Automatic analysis of the EMG interference pattern
Electroencephalogr. Clin. Neurophysiol.
(1983) - et al.
Discriminant classification of motor unit potentials (MUPs) successfully separates neurogenic and myopathic conditions
Electroencephalogr. Clin. Neurophysiol.
(1995) - et al.
Multiply sectioned Bayesian networks for neuromuscular diagnosis
Artif. Intell. Med.
(1993) - et al.
Neural network analysis of the EMG interference pattern
Med. Eng. Phys.
(1996) - et al.
An integrated intelligent computing model for the interpretation of EMG based neuromuscular diseases
Expert Syst. Appl.
(2009) - et al.
A new method for the extraction and classification of single motor unit action potentials from surface EMG signals
J. Neurosci. Methods
(2004)
Classification of EMG signals using wavelet neural network
J. Neurosci. Methods
A two-stage method for MUAP classification based on EMG decomposition
Comput. Biol. Med.
Adaptive fuzzy k-NN classifier for EMG signal decomposition
Med. Eng. Phys.
Multiresolution MUAPs decomposition and SVM-based analysis in the classification of neuromuscular disorders
Comput. Methods Prog. Biomed.
Medical decision support system for diagnosis of neuromuscular disorders using DWT and fuzzy support vector machines
Comput. Biol. Med.
Classification of EMG signals using PSO optimized SVM for diagnosis of neuromuscular disorders
Comput. Biol. Med.
A two-stage method for MUAP classification based on EMG decomposition
Artif. Intell. Med.
Techniques of EMG signal analysis: detection, classification and applications
Biol. Proced. Online
Electromyography in Clinical Practice: Electrodiagnostic Aspects of Neuromuscular Disease
Muscles Alive: Their Functions Revealed by Electromyography
Principles of Neural Science
Decomposition-based quantitative electromyography: methods and initial normative data in five muscles
Muscle Nerve
Quantitative EMG and motor unit recruitment threshold using a concentric needle with quadrifilar electrode
Muscle Nerve
Comparison of reliability of different algorithms used for quantitative computer assisted motor unit action potential analysis
Electroencephalogr. Clin. Neurophysiol.
Bioelectromagnetism
Automatic analysis of the electromyographic interference pattern. Part I. Development of quantitative features
Muscle Nerve
Automatic analysis of the electromyographic interference pattern. Part IIfindings in control subjects and in some neuromuscular diseases
Muscle Nerve
New attempts to quantify concentric needle electromyography
Muscle Nerve
The diagnostic power of motor unit potential analysisan objective Bayesian approach
Muscle Nerve
Unsupervised pattern recognition for the classification of EMG signals
IEEE Trans. Biomed. Eng.
Cited by (126)
Optimizing electrode positions on forearm to increase SNR and myoelectric pattern recognition performance
2023, Engineering Applications of Artificial IntelligenceSwallowing disorders analysis using surface EMG biomarkers and classification models
2023, Digital Signal Processing: A Review JournalOn the detection of activity patterns in electromyographic signals via decision trees
2024, Evolutionary IntelligenceStudies on biomedical signal processing: A review
2024, AIP Conference ProceedingsHuman-machine interface improving quality of patient care
2023, Human Machine Interface: Making Healthcare DigitalHand medical monitoring system based on machine learning and optimal EMG feature set
2023, Personal and Ubiquitous Computing