Research Of Detection Algorithm Of Atrial Fibrillation Based On Multi Feature Extraction And Convolution Neural Network

In recent years,the incidence of atrial fibrillation(AF)is increasing with the increasing aging population.Meanwhile,the complications caused by AF and cardiovascular disease are a serious threat to human health.Therefore,the development of AF detection system to reduce the incidence and mortality of critical illness is of great significance through the detection and treatment of AF as soon as possible.At present,the existing AF detection algorithms failed to extract the robust,discriminate features and had poor generalization.The Recurrence Complex Network and the Magnitude-squared coherence have been shown to be reliable feature extractors of ECG signal.And this advantage of these two methods is applied to the AF detection algorithm.In view of the atrial activity,during AF,the P-wave is replaced by fibrillatory waves.This phenomenon can reflect the essential characteristics of AF.Based on the characteristic of atrial activity,this paper designs an algorithm for AF detection based on multi feature extraction and Convolution Neural Network.The contents of this thesis mainly include:(1)An atrial fibrillation detection algorithm is proposed based on Recurrence Complex Network and Convolution Neural Network.There are many defects such as unstable feature extraction and poor robustness when the atrial activity signal is used as the input of the network,but the extraction of low level feature can solve this problem.In this paper,the intrinsic feature of each heartbeat of the atrial activity signal is constructed by a recurrence complex network.Then,a convolution neural network is used to extract high level feature and classify by analyzing the low level feature extracted by the recurrence complex network.(2)An atrial fibrillation detection algorithm is proposed based on Magnitude-squared coherence and Convolution Neural Network.There are significant differences between adjacent heart beats during atrial fibrillation.Therefore,the coherence spectrum between adjacent atrial activity signals is used as the low level feature,and then the input features arestudied and classified by convolution neural network.(3)By using the method of decision level fusion,the above two algorithms are fused to obtain a new algorithm for atrial fibrillation detection based on multi feature extraction and convolution neural network.The low level features extracted by the recurrence matrix and coherence spectrum are respectively based on single heart beat and the adjacent beats.The integration of these two types of features can better show the characteristics of atrial fibrillation.In this paper,the AF detection algorithm is completely rate-independent.The MIT-BIH AF database is used to evaluate the performance of the proposed method.The results show that the proposed algorithm has high performance and with excellent generalization ability.