A deep learning approach for atrial fibrillation signals classification based on convolutional and modified Elman neural network

摘要

Atrial fibrillation (AF) is one of the main causes of life-threatening heart disease. Its detection and diagnosis have been highly concerned by physicians in recent years. However, the conventional AF detection through visual inspection of electrocardiogram (ECG) data by expert cardiologists is low-efficient and time-consuming. In this work, we develop a novel approach for the automated AF detection based on an 11-layers neural network. The network structure is primarily stacked by convolutional neural network (CNN) and the modified Elman neural network (MENN), while automatically performing end-to-end signals classification. To verify the superiority of the proposed model, two relevant deep network models were specially constructed for comparison. Moreover, tenfold cross-validation was employed to evaluate the classification performance of the model on the MIT-BIH AF database. Compared with the two relevant models and several state-of-the-art methods, the model yielded excellent classification performance with the accuracy, sensitivity and specificity of 97.4%, 97.9%, and 97.1%, respectively. To the best of our knowledge, this is also the first time to stack CNN and MENN for ECG signals analysis and AF detection in particular. Its outstanding performance demonstrates that the proposed model has great potential as an efficient and robust identification system to assist physicians and reduce mortality. (C) 2019 Elsevier B.V. All rights reserved.