Research On Classification Method Of ECG Signal Based On Deep Learning

Cardiovascular disease is the biggest threat to human health.Because of the sudden and high risk of cardiovascular disease,it is very important to monitor the ECG signal in real time.Computer-aided automatic classification of ECG signal is an effective solution to this problem,but the existing classification algorithm in feature extraction ability,data imbalance and classification accuracy is still the main bottleneck restricting its wide application.In view of the above problems,combining the advantages of deep learning model,this paper studies the classification methods of ECG signals,the main research contents are as follows:1)The ECG signal is denoised and reconstructed by wavelet transform.After detecting the position of R peak,the ECG signal is segmented,and the ECG data is enhanced by an over sampling method based on interpolation to reduce the impact of data imbalance.2)The ECG signal classification model based on one-dimensional convolution neural network is designed,and the beat feature is extracted by using the local receptive field characteristic of convolution neural network.The final classification accuracy of this model is 98.43% and comprehensive index of Macro-F1 is 91.95%.3)A combined classification model based on convolution neural network and long-short term memory network is designed.The model uses the memory ability of long-short term memory network in time,overcomes the shortcomings of convolution neural network in time feature acquisition ability,and effectively extracts the time feature and spatial feature of ECG signal.The final classification accuracy of the model is 98.69% and Macro-F1 is 93.07%.4)Referring to the algorithms in natural language processing,a sequence to sequence model based on bidirectional long-short term memory network is designed,and attention mechanism is added into the model.The model can learn the heart beat features in a deeper level,and give corresponding weight according to its importance,so as to improve the feature extraction ability and screening ability of heart beat.The final classification accuracy of the model is 99.28% and Macro-F1 is 95.70%.The test of MIT-BIH arrhythmia database shows that the three kinds of ECG classification algorithms based on deep learning designed in this paper have achieved good results in the aspects of ECG feature extraction ability,data imbalance processing and classification recognition accuracy.These methods are helpful to improve the effectiveness and clinical reference value of computer-aided ECG automatic classification diagnosis.