Analysis of Noise Perturbation on Neural Network Based ECG Classification

Abstract

The heart diseases are diagnosed by analysing the ECG signals. However, during the acquisition process, the ECG signals are affected by different noises. Therefore, it is crucial to realize a pretreatment of the ECG signals before extracting the features. This article aims to study the effects of the Empirical Mode Decomposition filtering and Butterworth filtering on arrhythmia classification based on the convolutional neural network. Five classes of arrhythmia are concerned, including the sino-auricular node dysfunction, the supra-ventricular tachycardia, the ventricular tachycardia, the auricular flutter and the auricular Fibrillation. The proposed approach is evaluated with the MIT-BIH Arrhythmia database.