Technical Note
Classification of ECG arrhythmia by a modular neural network based on Mixture of Experts and Negatively Correlated Learning

https://doi.org/10.1016/j.bspc.2012.10.005Get rights and content

Abstract

In this paper, we propose a novel ECG arrhythmia classification method using the complementary features of Mixture of Experts (ME) and Negatively Correlated Learning (NCL). Negative Correlation Learning and Mixture of Experts methods utilize different error functions for simultaneous training of negatively correlated Neural Networks. The capability of a control parameter for NCL is incorporated in the error function of ME, which enables the training algorithm of ME to establish a balance in bias-variance-covariance trade-offs. ECG records from the MIT-BIH arrhythmia database are selected as test data. It is observed that the proposed classification method not only preserves the advantages and alleviates the disadvantages of its basis approaches, but also offering significantly improved performance over the original methods.

Highlights

► Mixture of Experts (ME). ► Negatively Correlated Learning (NCL). ► Neural network ensembles. ► Classifier fusion. ► Arrhythmia classification. ► Stationary wavelet transform (SWT).

Introduction

Arrhythmia can be defined as an irregular single heartbeat (arrhythmic beat). Arrhythmias can affect the heart rate causing irregular rhythms, such as slow or fast heartbeat [1]. Arrhythmias can take place in a healthy heart and be of minimal consequence but they may also indicate a serious problem that may lead to stroke or sudden cardiac death [2]. Therefore, automatic arrhythmia detection and classification is critical in clinical cardiology, especially when performed in real time. This is achieved through the analysis of the electrocardiogram (ECG) and its extracted features [3]. In the literature, several methods have been proposed for the automatic detection and classification of ECG signals [4], [5], [6]. ECG features can be extracted in time domain [7], in frequency domain [8] or represented as statistical measures [5]. The results of the studies have demonstrated that the Wavelet Transformation (WT) is the most promising method to extract features from the ECG signals [6], [9], [10]. WT opens another category of methods that represent the signal in different positions and scales. Researchers also have demonstrated the features extractions using Fourier Transform [11], Principal Component Analysis (PCA) [12] and Independent Component Analysis (ICA) [13] which were used as an input pattern for classification task to the classifier. Artificial neural networks (ANNs) were also employed to exploit their natural ability in pattern recognition tasks for successful classification of ECG beats after dilation and translation of an analyzing wavelet [14], [15], [16]. Combining classifiers to achieve higher accuracy is an active field of research with application in the area of ECG arrhythmia classification. Essentially, the idea behind combining classifiers is based on the so-called divide-and-conquer principle that is often used to tackle a complex problem by dividing it into simpler problems whose solutions can be combined to yield a final solution. Utilizing this principle, Jacobs et al. [17] proposed a modular neural network architecture called Mixture of Experts (ME). The ME models the conditional probability density of the target output by mixing the outputs from a set of local experts, each of which separately derives a conditional probability density of the target output. The ME weights the input space by using the posterior probabilities that expert networks generated for getting the output from the input. The outputs of expert networks are combined by a gating network simultaneously trained in order to stochastically select the expert that is performing the best at solving the problem [18]. As pointed out by Jordan and Jacobs [19], the gating network performs a typical multiclass classification task [20]. In [21] the authors have used ME approach for diagnosis of ophthalmic arterial diseases and also in [22] has demonstrated that the combined eigenvector methods (RNN approach) can be useful in analyzing the ECG beats. In [23] the authors have used an ensemble of neural networks for recognition and classification of arrhythmia.

Generally, combining methods have two major components, i.e., a method for creating individual neural networks (NNs) as the experts and a method for combining them [24], [25], [26], [27]. Both theoretical and experimental studies [28] have shown that combining procedure is most effective when the experts’ estimates are negatively correlated; they are moderately effective when the experts are uncorrelated and only mildly effective when the experts are positively correlated. So, more improved generalization ability can be obtained by combining the outputs of NNs which are accurate and their errors are negatively correlated [29]. Negative Correlation Learning (NCL) [30] and ME [17], are two leading combining methods, employs special error functions to simultaneously train neural networks (NNs) and produce negatively correlated NNs. NCL and ME implicitly creates different training sets by encouraging different NNs (experts) to learn different parts or aspects of the training data [30].

In this paper, we have proposed a new combination method for classifying normal heartbeats, premature ventricular contraction (PVC) arrhythmias and other abnormalities. Among the various abnormalities related with functioning of the human heart, PVC is one the most important arrhythmias. PVC is the contraction of the lower chambers of the heart (the ventricles) that occur earlier than usual, because of abnormal electrical activity of the ventricles [2]. In preprocessing module, an un-decimated wavelet transform is used to provide an informative representation that is both robust to noise and tuned to the morphological characteristics of the waveform features. For feature extraction module, we have used a suitable set of features that consist of both morphological and temporal features. This way we can include both shaping and timing information of signals. In this work, we propose a new method based on the complementary features of Mixture of Experts (ME) and Negative Correlation Learning (NCL) methods to constitute the best classification system with high accuracy rate for ECG beats. The paper is organized as follows. Section 2 explains the ECG signals and spectral analysis of signals using UWT in order to extract features characterizing the behavior of the signals under study. Section 3 presents the back ground on ME and NCL. Section 3.3 presents the Incorporation of NCL Training Algorithm into ME. It is followed then by the experimental results of ECG signals classification in Section 4. Finally, Section 5 draws conclusion and summarizes the paper.

Section snippets

Data preparation and feature extraction

An ECG consists of three basic waves: the P, QRS, and T. These waves correspond to the far field induced by specific electrical phenomena on the cardiac surface, namely, a trial depolarization (P wave), ventricular depolarization (QRS complex), and ventricular repolarization (T wave). One of the most important ECG components is the QRS complex [5]. Fig. 1 shows a waveform of normal signal. Among the various abnormalities related with functioning of the human heart, premature ventricular

Negative Correlation Learning (NCL)

Neural network ensembles are effective techniques to improve the generalization of a neural network system. Most of them always train ensemble individual networks independently. In this situation, the interaction among the individual networks in the ensemble may not be fully exploited. Negative Correlation Learning (NCL) was firstly proposed by Liu and Yao [30], which emphasizes interaction among individual networks in the ensemble during the learning process. It introduces a correlation

Experimental results

This experiment is designed to show how Mixture of Experts based on Negative Correlation Learning s method works better than Mixture of Experts and Negative Correlation Learning methods by comparing the performances against each other.

After that the improvement of our proposed method via error analysis are shown using an evaluation metrics to determine the effectiveness of Mixture of Experts based on Negative Correlation Learning. The MIT–BIH arrhythmia database [40] was used as the data source

Conclusion

This paper first introduces a new combining approach to classifying ECG arrhythmia. Then it analyses and compares mixture of expert and negative correlation learning as well as proposed method, Mixture of Experts based on Negative Correlation Learning s, in term of classification accuracy on MIT-BIH arrhythmia database. Taking advantage of the stationary wavelet transform, which also serves as a tool for noise reduction, we extracted 10 ECG morphological, as well as one timing interval

Acknowledgment

The authors wish to thank Islamic Azad University, South Tehran Branch for funding this project.

References (41)

  • E. Sandoe et al.

    Arrhythmia: a Guide to Clinical Electrocardiology

    (1991)
  • D.A. Sideris

    Primary Cardiology

    (1991)
  • A.S. Al-Fahoum et al.

    Combined wavelet transformation and radial basis neural networks for classifying life-threatening cardiac arrhythmias

    Medical and Biological Engineering and Computing

    (1999)
  • S. Osowski et al.

    ECG beat recognition using fuzzy hybrid neural network

    IEEE Transactions on Biomedical Engineering

    (2001)
  • Y.H. Hu et al.

    A patient adaptable ECG beat classifier using a mixture of experts approach

    IEEE Transactions on Biomedical Engineering

    (1997)
  • K. Minami et al.

    Real-time discrimination of ventricular tachyarrhythmia with Fourier-transform neural network

    IEEE Transactions on Biomedical Engineering

    (1999)
  • D. Donoho

    De-noising by soft-thresholding

    IEEE Transactions on Information Theory

    (1995)
  • A. Ebrahimzadeh et al.

    Detection of premature ventricular contractions using MLP neural networks: a comparative study

    Measurement

    (2009)
  • Z. Dokur et al.

    Comparison of discrete wavelet and Fourier transforms for ECG beat classification

    Electronics Letters

    (1999)
  • A. Hyvärinen

    Fast and robust fixed-point algorithms for independent component analysis

    IEEE Transactions on Neural Networks

    (1999)
  • Cited by (103)

    • Image based deep learning in 12-lead ECG diagnosis

      2023, Frontiers in Artificial Intelligence
    View all citing articles on Scopus
    View full text