Automatic Recognition Method Of Arrhythmia Based On Deep Learning

Automatic recognition of arrhythmia is of great significance for timely diagnosis and treatment of arrhythmia.The ECG abnormalities are acknowledged to be the most important evidence for the recognition of arrhythmia.However,because there are many types of arrhythmias,and the ECG abnormalities are very similar between them,the existing machine learning-based methods don’t work well in the recognition of multiple arrhythmias.This paper proposes two novel automatic recognition methods of arrhythmia based on deep learning.Firstly,the heart beat segments and corresponding RR intervals of ECG are extracted respectively,which are fed into the deep network model.Secondly,combining with convolution neural network and bidirectional long-term and short-term memory network,a convolution-bidirectional long-term and short-term memory network(CNN-bi LSTM)model is designed.Thirdly,in order to overcome the large computational cost of the bi LSTM module in the CNN-bi LSTM model,a one-dimensional multi-scale convolution depth neural network(1DMS-CNN)model is proposed by the multi-scale technique.Finally,performances of the proposed method are verified on the MIT-BIH ECG data.The results of numerical experiments show that the accuracy of CNN-bi LSTM model in automatic recognition of arrhythmia is 98.93%.The accuracy of 1DMS-CNN model in automatic recognition of arrhythmia is 99.26%.