A method for automatic detection of atrial fibrillation: based on CNN combined with BLSTM (Preprint)
BACKGROUND Atrial fibrillation (AF) is the most common arrhythmia harmful to human health. The morbidity and mortality of AF increase with age. However, due to its small amplitude and short duration, as well as its complexity and nonlinearity, it is very difficult to carry out accurate analysis only through expert experience. Therefore, the automatic detection of AF becomes extremely important. OBJECTIVE The purpose of this paper is to develop a method of automatic detection of AF, more accurate and efficient automatic detection of the occurrence of AF. METHODS In this paper, a deep learning model consisting of a convolutional neural network (CNN) and a bi-directional long short-term memory (BLSTM) network is proposed to detect AF automatically. The model is mainly composed of six convolutional layers, three pooling layers, a BLSTM layer and a fully connected layer. In addition, the data of MIT-BIH AFDB database is divided into different lengths of ECG segments, which are used as input to the network to verify the effect of different lengths of ECG segments on the final result. RESULTS The method proposed also achieved excellent results on ECG signals of 1 second. Most importantly, our proposed method achieved the best performances on ECG signals of 10 seconds, with an accuracy of 99.57%,a sensitivity of 99.65%,and a specificity of 99.49%. CONCLUSIONS This method requires no complex pretreatment and retains the characteristics of the original ECG to the greatest extent. Compared with the existing studies, the proposed method has higher accuracy and provides an effective solution for the automatic detection of AF.