Open AccessAtrial Fibrillation Prediction Algorithm Based on Attention Model
Author(s) -
Meng Shen,
Luqiao Zhang,
Xue Luo,
Jing Xu
Publication year - 2020
Publication title -
journal of physics. conference series
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.21
H-Index - 85
eISSN - 1742-6596
pISSN - 1742-6588
DOI - 10.1088/1742-6596/1575/1/012122
Subject(s) - artificial intelligence , atrial fibrillation , computer science , deep learning , convolutional neural network , artificial neural network , machine learning , data set , feature (linguistics) , feature extraction , algorithm , pattern recognition (psychology) , cardiology , medicine , linguistics , philosophy
Atrial Fibrillation is a common arrhythmia. when the heart is in the state of atrial fibrillation, can not deliver enough oxygen rich blood to the body. It may be asymptomatic when AF occurs. The patient does not know that AF has occurred. If the patient is not treated in time, the consequences of AF may be fatal. The diagnosis of AF needs a doctor with rich clinical experience, because the ECG signal of AF type is similar to that of other types of heart rate, which is difficult to confirm. The purpose of this paper is to train a better model through machine learning method, to realize the automatic detection of long-term AF ECG signal, to predict AF in advance, to automatically classify, recognize and inform patients. In this way, even if the patient himself can know the physical condition in time, early treatment. Automatic prediction and diagnosis of AF by machine learning method can play a positive role in the treatment of AF, but the existing scheme is not ideal for the detection and recognition rate of AF type. This paper proposes a deep learning framework based on attention mechanism. The segmented data are sequentially input into the deep convolution neural network for feature extraction, and then input into the two-way recurrent neural network with attention layer to predict the AF signal. The accuracy of 99.1% in the existing data set is better than the existing deep learning classification algorithm, which proves the validity and feasibility of the model.