Automatic Detection of Atrial Fibrillation Using Electrocardiomatrix and Convolutional Neural Network

Abstract

Long-term electrocardiogram (ECG) monitoring is a standard clinical routine in cryptogenic stroke survivors to assess the presence of atrial fibrillation (AF). However, manual evaluation of such recordings is time consuming, in particular when brief episodes are of interest. The electrocardiomatrix (ECM) technique allows compact, two-dimensional representation of the ECG and facilitates its review. In this study, we present a convolutional neural network (CNN) approach for automatic detection of AF based on ECM images. ECG segments of only 10 beats were converted into ECM images. A CNN was implemented to classify the ECMs between non-AF and AF. The CNN was trained using the MIT-BIH-NSR and the MIT-BIH-LTAF, and tested on the MIT-BIH-AF. A total of 120088 non-AF and 108088 AF ECM images were classified with accuracy of 86.95%. This study suggests that a CNN allows automatic detection of AF episodes of only 10 beats when the ECG data is represented as an ECM image.