Classification of EMG signals using PSO optimized SVM for diagnosis of neuromuscular disorders
Introduction
The human skeletal muscular system is mainly responsible for providing the required forces to perform various actions. This system consists of the nervous system and the muscular system, which together form the neuromuscular system. Neuromuscular disorders originating in the nervous system, in the neuromuscular junctions, and in the muscle fibers have different degrees of severity ranging from minor loss of strength to amputation due to neuron or muscle death. Neuromuscular disorders can be caused by several different muscle fibers or nerves, occasionally far removed from the symptoms. Hence precise localization of the disorder is of paramount importance so that more focused treatment can be administered. Electromyography (EMG) is used for diagnosing patients with neuromuscular disorders. Type of pathology, location and etiology may be investigated by using characteristic of the EMG waveform. However, these techniques assist doctors in their diagnosis but in complicated cases more invasive methods such as muscle biopsies or more sophisticated imaging techniques such as ultrasound or MRI are required. The analysis of EMG signals is generally carried out by trained and skilled neurologists who in addition to examining EMG waveforms also use techniques such as needle conduction studies and muscle acoustics. Problems occur when there are too few experts to meet the demand of patients and, hence, it is becoming more and more important to develop an automated diagnostic systems based on EMG readings. This need provides capacity for the application of machine learning techniques for the detection and classification of neuromuscular disorders based on EMG signal processing. These diagnostic systems will help medical specialists in detecting anomalies in the neuromuscular system. The purpose of diagnostic systems is to first preprocess the raw EMG signal and then extract characteristic information or features. Furthermore, it helps in diagnosis of neuromuscular disorders without using muscle biopsies or more sophisticated imaging techniques such as ultrasound or MRI. Features that can be time and frequency domain information, such as Fourier coefficients, autoregressive coefficients, wavelet coefficients, and a wide range of quantities derived from other signal processing techniques [1]. In this study, statistical features of discrete wavelet transform (DWT) have been used to characterize the EMG interference pattern in the diagnosis of neuromuscular diseases. It has been found that these statistical features give highly significant differences between healthy, myopathic, and neuropathic subjects and thus provide very useful features for disease classification. These features can then be used as input data for classifiers such as NNs and SVMs, which can classify or detect neuromuscular disorders [2]. However, design of practical and accurate automated diagnostic system remains challenging. The challenge in this area is to develop a robust and efficient classification technique that preserves important discriminatory information so as to provide a better accuracy for classification. Consequently, for an effective automated EMG signal classification, a systematic treatment of EMG signals must be carried out. For this reason, a number of computer-based quantitative EMG analysis algorithms have been developed. More recently, Hassoun et al. [3] proposed the NNERVE algorithm in 1994 in an attempt to computerize the extraction of individual EMG. Schizas et al. [4] used ANN to classify action potentials of a large group of muscles. Their work was later extended by Schizas et al. [5] to compare K-means, MLP-NN, SOMs, and genetic-based classifiers. It was shown that the simple K-means algorithm was not suitable as it gave the lowest classification accuracy, but both ANN and genetic-based models produced promising results. Pattichis et al. [6] extended the work by Schizas et al. [5] by applying ANN to motor unit action potential (MUAP) signals collected from the biceps brachii muscle. They compared the MLP network against the K-means clustering and Kohonen's SOM. Comparisons with unsupervised learning methods were undertaken by Pattichis and Elia [7] using SOM, learning vector quantization (LVQ), and statistical methods. Pattichis and Pattichis [8] have investigated the usefulness of the wavelet transform (WT), that provides a linear time-scale representation is for describing MUAP morphology and three different neural networks, the backpropagation (BP), the radial basis function network (RBF), and the self-organizing feature map (SOFM) [1]. Subasi et al. [9] investigated the practicality of using an autoregressive model with wavelet neural network to classify EMG signals. Subasi and Kiymik [10] developed the EMG decomposition system consisting of time–frequency methods and independent component analysis (ICA).
Nevertheless, in some of the studies conducted before, the EMG signals were acquired from surface EMG and the classification is realized on the basis of different feature extraction methods. Actually, there exist so many differences between intramuscular EMG (iEMG) signals and surface EMG (sEMG) signals that it cannot be guaranteed that the effective algorithms used in sEMG scenario also work well in iEMG scenario. Furthermore, the feature extraction methods are also very important in the classification of EMG signals. In this paper, particle swarm optimization (PSO) optimized SVM classifier combined with statistical features extracted from DWT are compared different machine learning methods to classify iEMG signals.
SVM which is a promising data classification technique proposed by Vapnik [11]. SVM is generated from the training process with the training data. Later on, classification is implemented based on the trained model. The main problems encountered in setting up the SVM model are how to decide on the kernel function and its parameter values. Improper parameter settings and feature-selection may lead to deprived classification results [12], [13]. In order to improve the SVM classification accuracy, the parameters that should be optimized including the penalty parameter σ and the kernel function parameter (γ) for the radial basis function (RBF) kernel must be tuned [14]. So, the tuned model parameters have an effect on the classification accuracy [15], [16]. This study attempts to increase the EMG signal classification accuracy rate by utilizing an approach based on particle swarm optimization (PSO) in SVM. PSO is proposed by Kennedy and Eberhart [17], [18] and a population-based search algorithm that is initialized with a population of random solutions, called particles. Recently, several researches on PSO theories and applications have been reported [15], [16], [17], [18], [19], [20], [21]. In this study PSO is used as an optimization technique to optimize the SVM parameters in EMG signal classification. To contribute to the quantification of the routine needle EMG examination, a methodology has been developed for EMG signal classification which consists of three steps. In the first step, the EMG signals are decomposed into different frequency bands using discrete wavelet transform (DWT). In the second step, statistical features extracted from these subband decomposed EMG signals to get better accuracy for diagnosis of neuromuscular disorder. In the last step, an unknown EMG signal is classified as normal, myopathic or neurogenic by different classification methods (Fig. 1). It is shown that PSO-SVM can obtain a high accuracy of 97.41% using 10-fold cross-validation. As a result, PSO-SVM can be used as a powerful tool for EMG signal classification.
The remaining paper is organized as follows: in the next section, we give information about the subjects and present the methods applied in each step of the EMG signal classification process. Section 3 provides a complete experimental study of the PSO-SVM model based EMG signal classification scheme, in which the impact of feature set and algorithmic issues are examined with respect to classification performance. Finally, the conclusions are summarized in Section 4.
Section snippets
Subject and data acquisition
An EMG system (Keypoint; Medtronic Functional Diagnostics, Skovlunde, Denmark) with standard settings was used. The EMG signal was acquired from the biceps brachii muscle using a concentric needle electrode (0.45 mm diameter with a recording surface area 0.07 mm2; impedance at 20 Hz below 200 kΩ). The signal was band-pass filtered at 5 Hz to 10 kHz and sampled at 20 kHz for 5 s with 12 bit resolution. All the measurements from patients and control group were done in Neurology Department of University of
Results and discussion
The fitness value is used to evaluate goodness of the particles, namely hyper-parameter combination in PSO. An ideal fitness value should reflect the generalization performance of SVM for different hyper-parameter combination. In order to define the fitness value, ten-fold cross validations used in the training set for each particle and the average correct rate is taken as the fitness value [20], [49]. In this study, PSO-SVM classifiers were used to classify EMG signals with DWT and was used to
Conclusion
This study proposed a PSO-SVM system in which PSO was used to optimize the best SVM model parameters for EMG signal classification. One of the main contributions of this work was how the kernel parameters setting of the SVMs in EMG signal classification affects the classification performance. From the obtained experimental results we can suggest the use of the SVM approach for classifying EMG signals on account of their superior generalization capability compared to traditional classification
Conflict of interest statement
The author declares that they have no conflict of interest.
Acknowledgment
The author thank Dr. Mustafa Yilmaz at University of Gaziantep, Neurology Department for providing the EMG data utilized in this research. This research has been supported by International Burch University (IBU Project no: IBU2010-PRD001).
References (60)
Classification of EMG signals using combined features and soft computing techniques
Appl. Soft Comput.
(2012)- et al.
Autoregressive and cepstral analyses of motor unit action potentials
Med. Eng. Phys.
(1999) - et al.
Classification of EMG signals using wavelet neural network
J. Neurosci. Methods
(2006) - et al.
Particle swarm optimization for parameter determination and feature selection of support vector machines
Expert Syst. Appl.
(2008) - et al.
Preoperative prediction of malignancy of ovarian tumors using least squares support vector machines
Artif. Intell. Med.
(2003) - et al.
A GA-based attribute selection and parameter optimization for support vector machine
Expert Syst. Appl.
(2006) - et al.
A distributed PSO–SVM hybrid system with feature selection and parameter optimization
Appl. Soft Comput.
(2008) - et al.
A novel LS-SVMs hyper-parameter selection based on particle swarm optimization
Neurocomputing
(2008) - et al.
A combination of modified particle swarm optimization algorithm and support vector machine for gene selection and tumor classification
Talanta
(2007) EEG signal classification using wavelet feature extraction and a mixture of expert model
Expert Syst. Appl.
(2007)
Medical decision support system for diagnosis of neuromuscular disorders using DWT and fuzzy support vector machines
Comput. Biol. Med.
Neural classification of lung sounds using wavelet coefficients
Comput. Biol. Med.
Nuttle HLW. A neural network model with bounded-weights for pattern classification
Comput. Oper. Res.
Classification of electronic nose data with support vector machines
Sensors Actuators B Chem.
Classification of brain tumours using short echo time H MR spectra
J. Magn. Reson.
Evolutionary tuning of SVM parameter values in multiclass problems
Neurocomputing
Computational Intelligence in Biomedical Engineering
NNERVE: neural network extraction of repetitive vectors for electromyography. II. Performance analysis
IEEE Trans. Biomed. Eng.
Artificial neural nets in computer-aided macro motor unit potential classification
IEEE Eng. Med. Biol. Mag.
Neural network models in EMG diagnosis
IEEE Trans. Biomed. Eng.
Time-scale analysis of motor unit action potentials
IEEE Trans. Biomed. Eng.
Muscle fatigue detection in EMG using time–frequency methods, ICA and neural networks
J. Med. Syst.
The Nature of Statistical Learning Theory
Asymptotic behaviors of support vector machines with gaussian kernel
Neural Comput.
Swarm Intelligence
Classification of electrocardiogram signals with support vector machines and particle swarm optimization
IEEE Trans. Info. Technol. Biomed.
Introduction to the special issue on wavelet transforms and multiresolution signal analysis
IEEE Trans. Info. Theory
Wavelets and filter banks: theory and design
IEEE Trans. Signal Process.
Cited by (435)
Comparison of machine learning algorithms and feature extraction techniques for the automatic detection of surface EMG activation timing
2024, Biomedical Signal Processing and ControlMultiscale entropy with electrocardiograph, electromyography, electroencephalography, and photoplethysmography signals in healthcare: A twelve-year systematic review
2024, Biomedical Signal Processing and ControlDeep neural networks for corporate misconduct prediction in construction industry using data from social networks
2024, Automation in ConstructionArtificial intelligence for automatic classification of needle EMG signals: A scoping review
2024, Clinical NeurophysiologySurface EMG feature disentanglement for robust pattern recognition
2024, Expert Systems with ApplicationsHydrogen leakage faults classification diagnosis based on data-driven in hydrogen supply system of fuel cell trucks
2024, International Journal of Hydrogen Energy