Atrial Fibrillation Detection Based On Atrial Activity Feature And Convolution Neural Network

In recent years, with the development of society and the phenomenon of population aging is becoming more and more prominent, the incidence of cardiovascular disease in the global range shows a significant growth trend. At the same time, the incidence of atrial fibrillation is increasing year by year. Not only that, the complications of atrial fibrillation such as cerebral infarction, hypertension and so on have serious harm to human life and healthy. Therefore, the development of atrial fibrillation detection system to prevent and detect atrial fibrillation early has important clinical and practical social significance to improve the patient’s quality of medical care and economic burden.Although atrial fibrillation detection draws intensive attention to many scholars, it also exists deficiencies on data feature extraction and detection of classification. Based on this, this paper proposes a atrial fibrillation detection based on atrial activity feature and convolution neural network algorithm, the main research work is as follows:(1) Extracting the underlying feature of the atrial activity. Firstly, the heart beats are became into a plurality of ecg signal segments, and then these signal segments are whited to remove the redundancy of information between the data. Secondly, selecting the part of the whitening data construct the sparse dictionary and solve sparse coefficient according to the dictionary. Finally, because there are multiple ecg fragments make the obtained sparse feature dimensions too high. Therefore, in order to avoid over-fitting and be able to get a good test results, we deal with them pooling in this paper and obtain the underlying feature of the atrial activity.(2) Selecting the convolutional neural network detection atrial fibrillation. Convolution neural network is a kind of deep network which combines efficient learning feature and automatic classification. Therefore, this article put the underlying feature into the convolution neural network for learning feature again. According to the characteristics of the input data constantly adjust the parameters of the network to achieve the best. Finally, the learned characteristics are combined and classified, and then achieve the purpose of detecting atrial fibrillation.Using the MIT-BIH atrial fibrillation database is tested to evaluate the feasibility of the algorithm and compared with the detection results of the other kinds of algorithm. The experimental results show that the algorithm has good detection performance.