Automated detection of atrial fibrillation based on DenseNet using ECG signals

Abstract

Atrial fibrillation (AF) is the most common cardiac arrhythmia, and it can cause a variety of cardiovascular diseases. Nonetheless, the early stage of AF is usually paroxysmal, with strong concealment. Electrocardiogram (ECG) is one of the most important noninvasive diagnostic tools for heart disease. However, in order to interpret ECG accurately, clinicians need to have well-trained professional knowledge and skills. It is valuable to develop an efficient, accurate and stable automatic AF detection algorithm in clinical settings. In this paper, we propose a novel network architecture, named DenseNet-BLSTM network model, for automatically AF detection using the ECG signals. The proposed model is constructed integrating the DenseNet module, the BLSTM module, two fully connected layers and one SoftMax layer. In this paper, the DenseNet module is utilized for further capturing local feature maps, whereas the BLSTM module is used to obtain the long-term dependencies in ECG signals. The datasets used to validate and test the proposed model are from the MIT-BIH Atrial Fibrillation Database (MIT-AF). The experimental results show that our proposed model achieved 99.07% and 98.15% accuracy in training and validation set, and achieved 97.78% accuracy in the testing set which is unseen dataset. The proposed DenseNet-BLSTM has shown excellent robustness and accuracy in automatic AF detection.