Classification of electrocardiogram signals with support vector machines and genetic algorithms using power spectral features

doi:10.1016/j.bspc.2010.07.006

Biomedical Signal Processing and Control

Volume 5, Issue 4, October 2010, Pages 252-263

https://doi.org/10.1016/j.bspc.2010.07.006 Get rights and content

Abstract

This paper proposes a new power spectral-based hybrid genetic algorithm-support vector machines (SVMGA) technique to classify five types of electrocardiogram (ECG) beats, namely normal beats and four manifestations of heart arrhythmia. This method employs three modules: a feature extraction module, a classification module and an optimization module. Feature extraction module extracts electrocardiogram's spectral and three timing interval features. Non-parametric power spectral density (PSD) estimation methods are used to extract spectral features. Support vector machine (SVM) is employed as a classifier to recognize the ECG beats. We investigate and compare two such classification approaches. First they are specified experimentally by the trial and error method. In the second technique the approach optimizes the relevant parameters through an intelligent algorithm. These parameters are: Gaussian radial basis function (GRBF) kernel parameter σ and C penalty parameter of SVM classifier. Then their performances in classification of ECG signals are evaluated for eight files obtained from the MIT–BIH arrhythmia database. Classification accuracy of the SVMGA approach proves superior to that of the SVM which has constant and manually extracted parameter.

Introduction

An arrhythmia is any abnormal cardiac rhythm [1]. Heart arrhythmias result from any disturbance in the rate, regularity, and site of origin or conduction of the cardiac electric impulse [2]. Classification of arrhythmia is an important step in developing devices for monitoring the health of individuals. The sequence of electrical signals of heart provides symptomatic information for classifying cardiac arrhythmias. Classification of normal and abnormal beats requires offline analysis of the ECG record data. This paper investigates the detection and classification of ECG arrhythmias.

In the literature, several methods have been proposed for the automatic classification of ECG signals. Among the most recently published work are those presented in [3], [4], [5], [6], [7], [8], [9], [10], [11], [12], [13], [14], [15], [16], [17], [18], [19], [20].

These works are a clear indication of research maturation in the field of automatic ECG classification. However there are still some open issues in the design of an ECG classification system which may lead to the development of more robust and efficient classifiers. One of these issues is related to the choice of the classification approach. In particular, the SVM approach does not seem to have received the attention it deserves in the ECG classification literature despite its great potential. Indeed, the SVM classifier exhibits a promising generalization capability, thanks to the maximal margin principle (MMP) it is based upon [21]. Another important property is that it is less sensitive to the curse of dimensionality than traditional classification approaches. This is explained by the fact that the MMP makes it unnecessary to estimate explicitly the statistical distributions of classes in the hyper-dimensional feature space in order to carry out the classification task. Thanks to these interesting properties, the SVM classifier has proved successful in a number of different application fields. Turning back to ECG classification, another issue that need to be addressed is that, the selection of the best free parameters of the adopted classifier is generally done empirically (model selection issue).

In this paper, we propose an automated method for differentiating normal heartbeats (N) from left bundle branch blocks (LBBB or L), right bundle branch blocks (RBBB or R), atrial premature contractions (APC or A) and premature ventricular contractions (PVC or V) heartbeats [1]. The spectral feature extraction in combination with temporal features is used in this study. As mentioned, SVM classifier is used due to its popularity in various classification problems in recent years. One of the strengths of this study is the use of search capability of genetic algorithms for finding optimum values of parameters of SVM (model selection). The value of soft margin constant C penalty parameter of support vector machines which is a positive integer number and the value of Gaussian radial basis function (GRBF) kernel parameter σ which is a positive real number must be optimized. In our proposed power spectral-based hybrid genetic algorithm-support vector machine (SVMGA) method, the values of C and σ parameters of SVM classifier are specified by genetic algorithms.

The paper is organized as follows. Section 2 describes non-parametric power spectral density (PSD) estimation and the feature extraction module. Sections 3 Support vector machine (SVM), 4 Genetic algorithms explain the support vector machines (SVM) and genetic algorithms (GA), respectively. Section 5 presents our proposed SVMGA method. Section 6 describes the database and performance metrics. Section 7 shows some simulation results. Section 8 discusses the results and finally Section 9 concludes the paper.

Section snippets

Feature extraction

Power spectrum estimation is perhaps the most widely used method of signal analysis. The power spectrum is related to the correlation function through the Fourier transform. The power spectrum reveals the repetitive and correlated patterns of a signal, which are important in detection, estimation, data forecasting and decision-making systems. The goal of spectral estimation is to describe the distribution (over frequency) of the power contained in a signal, based on a finite set of data. The

Support vector machine (SVM)

SVM is a supervised machine learning method. SVM uses structural risk minimization (SRM) principle whereas in ANN, empirical risk minimization (ERM) is used to minimize training data error [25], [26].

SVM performs classification tasks by constructing optimal separating hyper-planes (OSH). OSH maximizes the margin between the two nearest data points belonging to two separate classes (Fig. 1).

Suppose the training set, (x_i, y_i), i = 1, 2, …, l, x ∈ R^d, y ∈{− 1, + 1} can be separated by the hyper-plane $w^{T} x +$

Genetic algorithms

Fig. 2 illustrates the operation of a general genetic algorithm. In GA, a candidate solution for a specific problem is called an individual or a chromosome and consists of a linear list of genes. Each individual represents a point in the search space, and hence a possible solution to the problem. A population consists of a finite number of individuals. Each individual is decided by an evaluating mechanism to obtain its fitness value. Based on this fitness value and undergoing genetic operators,

Proposed method

The free parameters C and σ greatly affect the classification accuracy of SVM model. However, it is not known beforehand what values of the parameters are appropriate. Therefore, GA is used to search for better combinations of the parameters in SVM. Based on the Darwinian principle of ‘survival of the fittest’, GA can obtain the optimal solution after a series of iterative computations. Fig. 3 presents the whole process of ECG beat classification method implemented in the paper. The obtaining

MIT–BIH arrhythmia database

The MIT–BIH arrhythmia database [30] was used as the data source in this study. The database contains 48 recordings. Each has a duration of 30 min and includes two leads; the modified limb lead II and one of the modified leads V1, V2, V4 or V5. The sampling frequency is 360 Hz, the data are bandpass filtered at 0.1–100 Hz and the resolution is 200 samples per mV. Twenty-three of the recordings are intended to serve as a representative sample of routine clinical recordings and 25 recordings contain

Results

We randomly selected 100 beats from each class, and used these 500 beats for training of classifiers. Total number of beats in our database was 18,290. Clearly, the number of our training set is less than 3% of all beats. Thus, our study is well generalized. We conducted six experiments in the evaluation of our algorithm. In the first experiment we try to find best feature extraction method among four different non-parametric PSD estimation methods that introduced in Section 2.1. Since

Discussion

As seen in Table 2, the multitaper PSD estimation method achieved best classification accuracy of 93.97% among four non-parametric PSD estimation methods (as bolded). The next best method is modified periodogram. Periodogram and Welch methods are in the next ranks.

Values of C and σ that used in Table 2 were experimentally predicted. Multiple execution of the program under a multitude of variables resulted in C = 10,000 and σ = 0.1 to show better outcome through trial and error.

One of superiorities

Conclusion

In this study, SVMGA approach is proposed for an automatic ECG beat classification. The SVMGA approach optimizes the value of GRBF kernel function parameter σ and the value of C parameter for SVM classifier, simultaneously.

In the first experiment periodogram, modified periodogram, Welch and multitaper non-parametric PSD estimation methods are compared for feature subset. As the result, the MTM method was selected to obtain a compact set of spectral features. Three timing features are extracted

References (31)

A. Ebrahimzadeh et al.
Detection of premature ventricular contractions using MLP neural networks: a comparative study
Measurement
(2010)
S.N. Yu et al.
Selection of significant for ECG beat classification
Expert Syst. Appl.
(2009)
U.R. Acharya
Automatic identification of cardiac health using modeling techniques: a comparative study
Inform. Sci.
(2008)
C.H. Lin
Frequency-domain features for ECG beat discrimination using grey relational analysis-based classifier
Comput. Math. Appl.
(2008)
S. Osowski et al.
Recognition and classification system of arrhythmia using ensemble of neural networks
Measurement
(2008)
G.D. Clifford et al.
Advanced Methods and Tools for ECG Data Analysis
(2006)
M.S. Thaler
The Only EKG Book You’ll Ever Need
(1999)
L.Y. Shyu et al.
Using wavelet transform and fuzzy neural network for VPC detection from the Holter ECG
IEEE Trans. Biomed. Eng.
(2004)
O.T. Inan et al.
Robust neural-network-based classification of premature ventricular contractions using wavelet transform and timing interval features
IEEE Trans. Biomed. Eng.
(2006)
T. Ince et al.
A generic and robust system for automated patient-specific classification of electrocardiogram signals
IEEE Trans. Biomed. Eng.
(2009)

R.R. Sarvestani et al.

VT and VF classification using trajectory analysis

Nonlinear Anal.

(2009)

S. Osowski et al.

ECG beat recognition using fuzzy hybrid neural network

IEEE Trans. Biomed. Eng.

(2001)

E.D. Ubeyli

Recurrent neural networks employing Lyapunov exponents for analysis of ECG signals

Expert Syst. Appl.

(2010)

P. Chazal et al.

Automatic classification of heartbeats using ECG morphology and heartbeat interval features

IEEE Trans. Biomed. Eng.

(2004)

M. Lagerholm

Clustering ECG complexes using Hermite functions and self-organizing maps

IEEE Trans. Biomed. Eng.

(2000)

Cited by (138)

Fragment-level classification of ECG arrhythmia using wavelet scattering transform
2023, Expert Systems with Applications
Cardiovascular disease detection and its prevention are among the most demanding tasks in the healthcare system nowadays, as around 50 million people worldwide are at risk of being affected by heart disease. The heart’s electrical activity recorded by an electrocardiogram (ECG) provides vital pathological information about cardiac abnormalities such as arrhythmia. However, the complexity and non-linearity observed in ECG signals make disease anticipation difficult. In this work, we proposed a new approach to classify 17-classes of cardiac arrhythmia using wavelet scattering transform (WST). The WST can provide translation-invariant and deformation-stable representations of ECG by using a series of wavelet convolutions with non-linear modulus and averaging operators. Scattering coefficients from four-time windows of WST for fixed-duration ECG fragments are taken as input features to the SVM classifier. We achieved an overall classification accuracy of 98.90% in categorizing 17 arrhythmia classes taken from the MIT-BIH arrhythmia database, having 1000 ECG fragments of 45 subjects. The proposed method categorizes a 10-second ECG fragment with an average classification time of 0.007 s on a computing platform of a 2.5 GHz processor with 8 GB RAM. Our results outperform existing state-of-the-art solutions and can be deployed in real-world applications.
Health warning based on 3R ECG Sample's combined features and LSTM
2023, Computers in Biology and Medicine
Most researches use the fixed-length sample to identify ECG abnormalities based on MIT ECG dataset, which leads to information loss. To address this problem, this paper proposes a method for ECG abnormality detection and health warning based on ECG Holter of PHIA and 3R-TSH-L method. The 3R-TSH-L method is implemented by:(1) getting 3R ECG samples using Pan-Tompkins method and using volatility to obtain high-quality raw ECG data; (2) extracting combination features including time-domain features, frequency domain features and time-frequency domain features; (3) using LSTM for classification, training and testing the algorithm based on the MIT-BIH dataset, and obtaining relatively optimal features as spliced normalized fusion features including kurtosis, skewness and RR interval time domain features, STFT-based sub-band spectrum features, and harmonic ratio features. The ECG data were collected using the self-developed ECG Holter (PHIA) on 14 subjects, aged between 24 and 75 including both male and female, to build the ECG dataset (ECG-H). The algorithm was transferred to the ECG-H dataset, and a health warning assessment model based on abnormal ECG rate and heart rate variability weighting was proposed. Experiments show that 3R-TSH-L method proposed in the paper has a high accuracy of 98.28% for the detection of ECG abnormalities of MIT-BIH dataset and a good transfer learning ability of 95.66% accuracy for ECG-H. The health warning model was also testified to be reasonable. The key technique of the ECG Holter of PHIA and the method 3R-TSH-L proposed in this paper is expected to be widely used in family-oriented healthcare.
ECG segmentation algorithm based on bidirectional hidden semi-Markov model
2022, Computers in Biology and Medicine
Accurate segmentation of electrocardiogram (ECG) waves is crucial for cardiovascular diseases (CVDs). In this study, a bidirectional hidden semi-Markov model (BI-HSMM) based on the probability distributions of ECG waveform duration was proposed for ECG wave segmentation. Four feature-vectors of ECG signals were extracted as the observation sequence of the hidden Markov model (HMM), and the statistical probability distribution of each waveform duration was counted. Logistic regression (LR) was used to train model parameters. The starting and ending positions of the QRS wave were first detected, and thereafter, bidirectional prediction was employed for the other waves. Forwardly, ST segment, T wave, and TP segment were predicted. Backwardly, P wave and PQ segments were detected. The Viterbi algorithm was improved by integrating the recursive formula of the forward prediction and backward backtracking algorithms. In the QT database, the proposed method demonstrated excellent performance (Acc = 97.98%, F₁ score of P wave = 98.37%, F₁ score of QRS wave = 97.60%, F₁ score of T wave = 97.79%). For the wearable dynamic electrocardiography (DCG) signals collected by the Shandong Provincial Hospital (SPH), the detection accuracy was 99.71% and the F₁ of each waveform was above 99%. The experimental results and real DCG signal validation confirmed that the proposed new BI-HSMM method exhibits significant ability to segment the resting and DCG signals; this is conducive to the detection and monitoring of CVDs.
ECG-based expert-knowledge attention network to tachyarrhythmia recognition
2022, Biomedical Signal Processing and Control
An expert-knowledge attention network (EKANet) was designed to improve the accuracy of arrhythmia diagnosis and reduce the recheck time. This network classifies four tachyarrhythmia on electrocardiogram (ECG) signals, encompassing most arrhythmia diseases. In the EKANet, two attention modules based on the knowledge of cardiology can rapidly capture the ECG rhythm and P waves in multiple leads without any training. This mechanism is performed to reduce the computational time of re-building a model. The EKANet integrates a six-layer convolutional neural network (CNN) and a gated recurrent unit (GRU) as the classifier to realise the tachyarrhythmia classification. The EKANet outperformed 1D CNN and ArrhythmiaNet on the MIT-BIH datasets by 3.1% on average accuracy. Furthermore, the EKANet achieved approximately 8.5% and 3.9% average F1-score increases on the dataset of China ECG challenge contest compared with time-incremental CNN (TI-CNN) and attention-based TI-CNN, respectively. Meanwhile, the EKANet has a much lower complexity than that of the other typical models with a competitive accuracy.
Hardness prediction of high entropy alloys with machine learning and material descriptors selection by improved genetic algorithm
2022, Computational Materials Science
Citation Excerpt :
However, GA is a stochastic global optimization algorithm, and thus the result of GA is strongly depended on the quality of initial population which is normally generated randomly, as well as the iteration times for reaching steady state. There are lots of improved GAs proposed for various specific problems [39–42], in which some GAs [34,37,43–46], such as svmGA [43], rfGA [44] and GARS [34] are designed to focus on feature selection. To the author’s knowledge, however, these methods [34,37,43–46] are mainly aimed to classify datasets.
With the coming of the age of artificial intelligence and big data, machine learning (ML) has been showing powerful potentials for properties prediction of materials. For achieving satisfying prediction performance, rational feature selection plays a key role along with a suitable ML model itself. In the present work, the traditional genetic algorithm (GA) has been further improved to serve as a feature selection method for the hardness prediction problem of high entropy alloys (HEAs). The concepts of feature importance and gene manipulation were introduced into the improved GA to make it more comprehensible. Comparative analysis demonstrated that the improved GA is superior to the traditional GA in the aspects of accuracy, stability and efficiency obviously. A comparison with other typical feature selection methods was also made. In addition, ML model selection was discussed with the composition feature or the optimal physical feature combination selected by the improved GA. Finally, in order to elevate the prediction ability of ML model, the stacking method as an ensemble learning strategy was proposed in Al-Co-Cr-Cu-Fe-Ni HEAs hardness prediction. It was shown that the prediction errors are successfully lowered. This ML framework could be regarded as a method with general applicability to select suitable ML model and material descriptors, for designing various materials with excellent properties and complex composition.
Imbalanced heartbeat classification using EasyEnsemble technique and global heartbeat information
2022, Biomedical Signal Processing and Control
Heartbeat classification, also known as arrhythmia detection, is essential for early prevention of cardiovascular diseases (CVD). However, in clinical, the physician checks the ECG signal beat-by-beat for diagnosis, which is time-consuming and laborious. Recently, some scholars have proposed computer-aided heartbeat classification methods, while these methods mainly focus on using the local information of heartbeat and ignore the role of the global information where the heartbeat is located. Meanwhile, the heartbeat is highly imbalanced, resulting in poor performance in the minority categories of existing methods. Based on these issues, in this study we propose a new method for classifying imbalanced heartbeat using EasyEnsemble technique with global heartbeat information. By testing on the MIT-BIH arrhythmia database using the inter-patient scheme, the experimental results show that the global heartbeat information is useful for heartbeat classification. Meanwhile, compared with the existing methods, our method can not only achieve the best overall performance, but also can significantly improve the performance of minority categories while maintaining the good performance of majority categories.

View all citing articles on Scopus

View full text

Classification of electrocardiogram signals with support vector machines and genetic algorithms using power spectral features

Abstract

Introduction

Section snippets

Feature extraction

Support vector machine (SVM)

Genetic algorithms

Proposed method

MIT–BIH arrhythmia database

Results

Discussion

Conclusion

Measurement

Expert Syst. Appl.

Inform. Sci.

Comput. Math. Appl.

Measurement

Advanced Methods and Tools for ECG Data Analysis

The Only EKG Book You’ll Ever Need

Using wavelet transform and fuzzy neural network for VPC detection from the Holter ECG

IEEE Trans. Biomed. Eng.

Robust neural-network-based classification of premature ventricular contractions using wavelet transform and timing interval features

IEEE Trans. Biomed. Eng.

A generic and robust system for automated patient-specific classification of electrocardiogram signals

IEEE Trans. Biomed. Eng.

VT and VF classification using trajectory analysis

Nonlinear Anal.

ECG beat recognition using fuzzy hybrid neural network

IEEE Trans. Biomed. Eng.

Recurrent neural networks employing Lyapunov exponents for analysis of ECG signals

Expert Syst. Appl.

Automatic classification of heartbeats using ECG morphology and heartbeat interval features

IEEE Trans. Biomed. Eng.

Clustering ECG complexes using Hermite functions and self-organizing maps

IEEE Trans. Biomed. Eng.