Atrial Fibrillation Detection Based On Convolutional Neural Network And Recurrent Neural Network

Atrial fibrillation is the most common arrhythmia.The prevalence rate is about 0.4?1% in the overall population,and the number of people over 80 years old increases to 8%.The appearance of atrial fibrillation symptoms is also closely related to cardiovascular disease.Atrial fibrillation itself does not directly threaten the life and health of patients.But without timely treatment,atrial fibrillation can cause serious complications,such as heart failure and stroke.Therefore,early and accurate diagnosis of atrial fibrillation is essential to prevent its complications.However,due to the occasional instability of atrial fibrillation and the unreliability of single-channel ECG signals,it is difficult to improve the accuracy of the traditional single-channel atrial fibrillation recognition scheme in the past,and the real-time performance of atrial fibrillation recognition cannot be guaranteed.The two most critical factors in patient monitoring products are the reliability and realtime performance of the diagnostic function.In order to further improve the accuracy and realtime analysis performance of atrial fibrillation recognition and diagnosis,we summarize and analyze the related technology research methods of predecessors.Starting from the technical basis of convolutional neural networks,recurrent neural networks and residual networks,a new recognition method for atrial fibrillation with satisfactory recognition accuracy and realtime performance is proposed: Based on Multi-channel Attention Combine Res Ne Xt Separable CNN and Trend Attention Bi-LSTM.In this method,we first input the multichannel signals in parallel to the Res Ne Xt Separable CNN to extract the features in the frequency domain,and then use the Trend Attention Bi-LSTM to give more attention to the key waveform trends that are favorable for atrial fibrillation recognition.Finally,a multichannel attention mechanism is used to match larger weight parameters for the channel branches of greater importance for atrial fibrillation recognition.Multi-channel comprehensive analysis can improve the reliability of atrial fibrillation classification,which can extract and analyze more effective information for shorter ECG signals to improve recognition accuracy.Through theoretical and experimental analysis,this paper proposes a new atrial fibrillation recognition method Based on Multi-channel Attention Combine Res Ne Xt Separable CNN and Trend Attention Bi-LSTM.Compared with this article in the case of 3s short ECG input The other solutions have obvious advantages.The six-fold cross-validation between subjects was conducted in the official data set of the MIT-BIH Atrial Fibrillation Database.The average sensitivity,specificity and accuracy of the atrial fibrillation recognition obtained during the 3s short ECG input were: 95.57%,95.67% and 96.13%.In the comparative experiment in this paper,this method is compared with other atrial fibrillation recognition schemes,the sensitivity is 4% ? 6%,and the specificity is 2% ? 5%.It can be explained that the method of Based on Multi-channel Attention Combine Res Ne Xt Separable CNN and Trend Attention BiLSTM also has good AF recognition performance under the condition of ensuring good realtime performance.