Detection of premature ventricular contractions using MLP neural networks: A comparative study Ataollah Ebrahimzadeh * , Ali Khazaee Faculty of Electrical and Computer Engineering, Babol Noushirvani University of Technology, Babol, Iran article info Article history: Received 30 March 2009 Received in revised form 26 June 2009 Accepted 13 July 2009 Available online 16 July 2009 Keywords: ECG beat classification Premature ventricular contraction MLP neural network Training algorithms Wavelet transform abstract This paper proposes a three stage technique for detection of premature ventricular contrac- tion (PVC) from normal beats and other heart diseases. This method includes a denoising module, a feature extraction module and a classification module. In the first module we investigate the application of stationary wavelet transform (SWT) for noise reduction of the electrocardiogram (ECG) signals. The feature extraction module extracts 10 ECG mor- phological features and one timing interval feature. Then a number of multilayer percep- tron (MLP) neural networks with different number of layers and nine training algorithms are designed. The performances of the networks for speed of convergence and accuracy classifications are evaluated for seven files from the MIT–BIH arrhythmia database. Among the different training algorithms, the resilient back-propagation (RP) algorithm illustrated the best convergence rate and the Levenberg–Marquardt (LM) algorithm achieved the best overall detection accuracy. Ó 2009 Elsevier Ltd. All rights reserved. 1. Introduction The development of accurate and quick methods for automatic electrocardiogram (ECG) classification is vital for clinical diagnosis of heart disease [1]. An arrhythmia is any abnormal cardiac rhythm [2]. Heart arrhythmias re- sult from any disturbance in the rate, regularity, and site of origin or conduction of the cardiac electric impulse [1]. Among the various abnormalities related with functioning of the human heart, premature ventricular contraction (PVC) is one the most important arrhythmias. PVC is the contraction of the lower chambers of the heart (the ventri- cles) that occur earlier than usual, because of abnormal electrical activity of the ventricles [2]. This paper investi- gates the detection and classification of PVC arrhythmias. In the literature, several methods have been proposed for the automatic classification of ECG signals. Among the most recently published works are those presented in [3–20]. In [3], the authors used the discrete wavelet trans- form as the feature extractor and linear discriminants as the classifier. In [4], the authors designed a local and global classifier and combined them with a mixture of experts (MOE) approach. In [5] the authors used a feed forward neural network as classifier. They derived five features including the QRS width and offset, amplitude of R seg- ment, the T segment slope and the RR-interval duration. In [6], the authors used wavelet feature extraction in tan- dem with fuzzy neural network classification for PVC beat classification. In [7], the authors used a multilayer percep- tron (MLP) neural network classifier and achieved an accu- racy of 88.3% in their testing set. In [8], the authors used morphological information as the features and a neural- network classifier for differentiating the ECG beats. In [9], the author used independent components analysis (ICA) for ECG detection. In [10,11], the authors used the power spectral density (PSD) of ECG signals. In [12], the authors have modeled the ECG signals using the MME (modified mixture of experts) network structure with diverse fea- tures. In [13], an image-based technique is presented which extracts discriminative information from the trajec- tories of ECG signals in the state space. The method 0263-2241/$ - see front matter Ó 2009 Elsevier Ltd. All rights reserved. doi:10.1016/j.measurement.2009.07.002 * Corresponding author. E-mail address: ataebrahim@yahoo.com (A. Ebrahimzadeh). Measurement 43 (2010) 103–112 Contents lists available at ScienceDirect Measurement journal homepage: www.elsevier.com/locate/measurement