Detection Of Atrial Fibrillation Via Deep Learning

Atrial fibrillation(AF)is one of the most common forms of arrhythmia in the clinic and is a serious threat to human health.Therefore,effective and accurate diagnosis of AF from electrocardiogram(ECG)recording is of significant importance and still challenging in a clinical setting.The performance of the traditional feature-based algorithms for AF detection depends on the robustness of P-wave and R-peak detection.If the key waveform is missed or detected by mistake,their performance may be reduced significantly.Manually defined features do not accommodate a large user base,resulting in a decline in performance based on traditional feature detection algorithms in a clinical setting.This paper proposes a deep learning-based atrial fibrillation detection algorithm,which extracts sensitive features from the original data automatically and classifies them.The main research contents of this paper are as follows:(1)An ECG signal data augmentation method based on characteristic points is proposed to produce extra and potentially non-redundant training data and improve the training effect of the model obviously.(2)The atrial fibrillation detection model was established via recurrent neural network and the data augmentation,parameter adjustment and overfitting prevention of network training were studied.(3)According to the characteristics of electrocardiogram signal,an algorithm for atrial fibrillation detection,which was suitable for single lead and multiple lead electrocardiogram,was proposed to improve the proposed detection model and improve the generalization ability of the algorithm Performance of the algorithm was validated using two public datasets from the computing in cardiology challenge 2017 challenge and China physiological signal challenge 2018.The algorithm is based on recurrent neural network to learn the sensitive features in the data and used them for atrial fibrillation detection.The data augmentation was used to improve the performance of models.The algorithm achieved high performance on both public data sets.