One Dimensional Convolution Neural Network Model for ECG Arrhythmia Classification

Abstract

Electrocardiograms (ECG) are one of the most effective and significant tools to diagnose and predict cardiovascular diseases (CVDs) such as arrhythmia. An ECG provides essential information related to the cardiovascular system and the primary functions of the human heart. It is a frequently used tool for non-invasively observing different CVDs. Accurate identification of arrhythmias is critical to patient well-being in clinical settings, as both acute and chronic heart conditions are typically reflected in these readings. We propose a deep one-dimensional convolutional neural network (1D-CNN) that can accurately classify five types of ECG waves, namely: normal, ventricular premature contraction, left bundle branch block, atrial premature contraction, and right bundle branch block. Optimization of the proposed CNN classifier results in three convolutional layers, three down sampling layers and two fully connected layers, which extract best features from the given data and automatically classify these based on the extracted features. Labeled ECG recordings from the publicly available MIT-BIH arrhythmia database were used for the classification. Results have shown that our CNN classifier attained 97.8% classification accuracy, which is better than other recently reported ECG signal classification algorithms.