Research On Early Detection Of Atrial Fibrillation Based On Deep Learning

Atrial fibrillation,abbreviated as AF,is a common arrhythmia disease,which can easily lead to complications such as stroke and cerebral infarction,thus seriously endangering the life safety of patients.However,the early onset of atrial fibrillation is relatively hidden,and patients cannot be diagnosed in time,making it a serious and potentially harmful disease.Therefore,the early diagnosis of atrial fibrillation based on portable mobile medical equipment is helpful for early detection and treatment of patients,and has become a research direction with great medical health significance and value.With the continuous development of deep learning technology,more and more studies apply deep neural networks to the detection of cardiac arrhythmia diseases,and the deep learning-based atrial fibrillation detection algorithm simplifies the cumbersome feature-extraction processs compared to the original traditional detection algorithm.Through data feature learning,the deep learning-based atrial fibrillation detection algorithm constructs an end-to-end neural network to realize atrial fibrillation detection,improves the algorithm’s anti-interference ability and generalization ability,and has great potential for use in portable mobile medical devices for atrial fibrillation detection.The main goal of this paper is to study the early rapid detection algorithm of atrial fibrillation based on deep learning technology,which can be applied to mobile medical equipment and lay the foundation for the algorithm research for the early screening of atrial fibrillation.This paper is based on the ECG signal data set published in the 2018 China Physiological Signal Analysis Challenge.The main research content includes two parts:(1)Based on the Densely Connected Convolutional Networks(Dense Net)originally used for two-dimensional image detection classification research,accomplishing the construction of a one-dimensional Dense Net network for single-lead atrial fibrillation detection,achieving up to 97.10% accuracy,92.44% sensitivity and 98.00% specificity,the network is connected through its fully connected interlayer.In this way,the full use of features is achieved,in addition,through reasonable network structure settings,the features are fully integrated and the amount of calculation is also reasonably controlled,making the network detection efficient and accurate.(2)Based on the advantages of convolutional recurrent neural network to process time series signals,combining one-dimensional Dense Net network with bidirectional recurrent neural network,a Densely Connected Convolutional Recurrent Neural Network(DBRNN)is proposed for atrial fibrillation Detection,the accuracy rate obtained is 97.61%,the corresponding sensitivity and specificity are 92.98% and 98.60%,respectively,compared with the one-dimensional Dense Net network has been further improved.The early detection algorithm for atrial fibrillation based on deep learning proposed in this paper has a high detection accuracy and an efficient network structure,which provides a higher possibility for its future application in portable single-lead ECG signal analysis scenarios.