A novel multi-scale convolutional network with attention-based bidirectional gated recurrent unit for atrial fibrillation discrimination

Abstract

Atrial fibrillation (AF) is a major cardiovascular disease that has affected thousands of individuals worldwide. The electrocardiogram (ECG) is the most extensively applied approach to detect AF at present, while the traditional detection strategy based on the visual observation of ECG data is often laborious and inefficient. In this work, we specially designed an intelligent recognition system based on a novel convolutional neural network that utilizes the multi-scale convolution kernel and bidirectional gated recurrent unit with attention mechanism for AF detection. Also, two standard control groups using 10-fold cross-validation were performed to assess the validity of the proposed model. The empirical results not only demonstrate the high efficiency of multi-scale convolution kernel, but also show that the model has a more superior classification performance to several state-of-the-art methods with an accuracy of 98.3% and 97.7% on two public databases, respectively. Due to its high performance, we plan to develop the model into portable devices to benefit more individuals such as the elderly and athletes.