Atrial Fibrillation Classification Using Convolutional Neural Networks and Time Domain Features of ECG Sequence

Abstract

Atrial fibrillation is a serious cardiovascular disease. It is the main cause of heart disease such as myocardial infarction. ECG based atrial fibrillation detection is very important for clinical diagnosis. In this paper, a method based on one-dimensional CNN and time domain features of ECG sequence is proposed to detect atrial fibrillation. The ECG data used came from the MIT-BIH atrial fibrillation database. The first step is to filter out the noise interference in ECG. In the second step, ECG signals were segmented into seven heart beats. In the third step, 8 features are extracted based on the time domain features of ECG sequence to form the feature vector (size 1*8). In the fourth step, the one-hot label (1*2) output by the convolutional neural network was combined with the extracted time domain features (size 1*8) to obtain a total of 10 dimensional features. In the fifth step, the extracted 10-dimensional features are normalized and then put into the SVM classifier. The experimental results show that the sensitivity, specificity and total accuracy of the proposed algorithm are 99.07%, 97.05% and 98.03%, respectively. This algorithm has great potential to help doctors and reduce mortality.