Decision support system for arrhythmia prediction using convolutional neural network structure without preprocessing

Abstract

Arrhythmia is a disease-influencing heart and is manifested by an irregular heartbeat. Atrial fibrillation (A(fib)), atrial flutter (A(fl)), and ventricular fibrillation (V-fib) are heart arrhythmias affecting predominantly senior citizens. An electrocardiogram (ECG) is a device serving to measure the ECG signal and diagnosis of an abnormal pattern which represents a heartbeat defects. Though it is possible to analyze these signals manually, in some cases it is a difficult task due to the often signal distortion by noise. Furthermore, manual analyzation of patterns is subjective and can lead to an inaccurate diagnosis. An automated computer-aided diagnosis (CAD) is a technique to eliminate these shortcomings. In this work, we present an 6-layer deep convolutional neural network (CNN) for automatic ECG pattern classification of the normal (N-r), A(fib), A(fl), and V-fib classes. This proposed CNN model requires simple feature extraction and no pre-processing of ECG signals. For two seconds ECG segments, the model obtained the accuracy of 97.78%, specificity and sensitivity of 98.82% and 99.76% respectively. This proposed system can be used as an assistant automatic tool in a clinical environment as a decision support system.