A multi-wavelet optimization approach using similarity measures for electrocardiogram signal classification. (July 2015)

• Record Type: Journal Article
• Authors: Nazarahari, Milad; Ghorbanpour Namin, Sahand; Davaie Markazi, Amir Hossein; Kabir Anaraki, Amin
• Is Part Of: Biomedical signal processing and control. Volume 20(2015)
• Page Start: 142
• Page End: 151
• Publication Date: 2015-07
• Subjects: ECG classification -- Wavelet design -- Discrete wavelet transform -- Similarity measure -- Hybrid GA-PSO; Signal processing -- Periodicals; Biomedical engineering -- Periodicals; Signal Processing, Computer-Assisted -- Periodicals; Image Processing, Computer-Assisted -- Periodicals; Biomedical Engineering -- Periodicals; 610.28
• DOI: 10.1016/j.bspc.2015.04.010
• Languages: English
• ISSNs: 1746-8094

Abstract

Highlights: This research paper presents a novel ECG classification method using new wavelets. New wavelet functions (WFs) have been designed using polyphase representation. The approach for generating WFs relies on the similarity between WFs and ECGs. Feature vector is obtained by applying all designed WFs to every single beat. ECG decomposition, application of PCA, and MLP provides the classification scheme. Abstract: One of the main approaches for classifying the ECG signals is the use of wavelet transform. In this paper, a method has been presented for classifying the ECG signals by means of new wavelet functions (WFs). The considered approach for generating the new WFs relies on the degree of similarity between the shapes of the WFs and ECG signals. Thus, by formulating the wavelet design problem in the hybrid GA-PSO framework, and using Euclidean, Dynamic Time Warping, Signed Correlation Index, and Adaptive Signed Correlation Index similarity measures as wavelet design criterion, six WFs corresponding to six common arrhythmias have been designed. Decomposition of ECG signal using designed WFs, and thereafter, application of PCA, and multilayer perceptron classifier provides a classification scheme for ECG signals. Feature vector is obtained by applying all designed WFs to every single beat; so, the main advantage of this method is that the set of WFs used to decompose the beats, always includes a WF similar to those beats. Therefore, the generated features better resolve the various classes. Also, the effects of the number of neurons in the hidden layer and the different training methods of the MLP have been investigated. By performing some tests on the benchmark MIT-BIH arrhythmia database using the proposed method and also the common WFs, the superiority of the proposed approach in the overall accuracy as well as the accuracy of each class has been demonstrated.