Feature Extraction and Classification for ECG signals Processing based on Stationary Multiwavelet Transform and Artificial Neural Network

Abstract

This paper proposes an algorithm that uses mix of Stationary Multiwavelet Transform and Artificial Neural Network (ANN) algorithm for classification of Electrocardiograph (ECG) signals. The MIT-BIH arrhythmia database is used to measure the performance of the suggested method and compare the results with conventional techniques. The Stationary Multiwavelet Transform (SMWT) and the Minimum Average Maximum strategy (MAM) is suggested to calculate the useful features of the signal before utilizing ANN algorithm for classification. Since SMWT is a translation invariant, therefore, it enhances the classification performance and reduces mean square error (MSE). Repeated Row Processing exists in this scheme to make it more suitable for feature extraction compared with Stationary Wavelet Transform (SWT), Multiwavelet Transform (MWT) and Principle Component Analysis (PCA). SMWT and MAM reduce dimensional space and decrease the complexity of classification circuit. ECG signal is classified using ANN. Finally, the results of the proposed method are realistic compared with SWT-ANN, MWT-ANN, and PCA-ANN. The obtained results emphasize the excellence of the presented algorithm than the traditional techniques. The SMWT-ANN achieves classification accuracy of 100% and mean square error of 〖1.4*10〗^(-3).