A stacked LSTM for atrial fibrillation prediction based on multivariate ECGs.

Abstract

Atrial fibrillation (AF) is an irregular and rapid heart rate that can increase the risk of various heart-related complications, such as the stroke and the heart failure. Electrocardiography (ECG) is widely used to monitor the health of heart disease patients. It can dramatically improve the health and the survival rate of heart disease patients by accurately predicting the AFs in an ECG. Most of the existing researches focus on the AF detection, but few of them explore the AF prediction. In this paper, we develop a recurrent neural network (RNN) composed of stacked LSTMs for AF prediction, which called SLAP. This model can effectively avoid the gradient explosion and gradient explosion of ordinary RNN and learn the features better. We conduct comprehensive experiments based on two public datasets. Our experiment results show 92% accuracy and 92% f-score of the AF prediction, which are better than the state-of-the-art AF detection architectures like the RNN and the LSTM.