| Electrocardiogram(ECG)is used to record the surface potential difference generated in the physiological process of the heart,which reflects the health of the human heart and has been widely used in the diagnosis of clinical cardiovascular diseases.The performance of traditional classification algorithms relies on manual extraction of ECG features,and then feeding them into the constructed model for identification.However,the individual ECG waveform features are quite different,resulting in that the extracted features cannot truly reflect the ECG intrinsic properties,and feature selection is difficult.Therefore,it is especially important to study an automatic classification method that can effectively detect and identify ECG signals.In this paper,we study the current situation of automatic classification of arrhythmias from three aspects:ECG signal preprocessing,QRS complex detection and ECG feature extraction and classification recognition.In addition,we summarize the existing problems in the field of automatic classification of arrhythmia,and improve the automatic classification of arrhythmia based on neural network.The main innovations are as follows:1、The ECG feature description extracted by the traditional convolutional neural network arrhythmia classification method usually does not contain time-frequency information.In order to solve this problem,wavelet transform is introduced for improvement,and an automatic arrhythmia classification method based on multi-scale wavelet convolutional neural network is proposed.The model obtains the characteristic representation of ECG signals at different scales by concatenating the features under different spatial structures,and realizes the automatic classification of N,S,V,F and Q arrhythmias recommended by the Association for Advancement of Medical Instrumentation(AAMI).In addition,we compare and analyze the classification performance of the proposed model,CNN,SVM and other models in the standard ECG database MIT-BIH,and verify the advantages and feasibility of the proposed method.2、Aiming at the problem that the traditional arrhythmia classification based on convolution neural network has not been extracted the timing information deeply,and ECG is a typical time series signal.This paper introduces the Recurrent Neural Network to improve it and proposes an automatic arrhythmia classification method based on Convolution Recurrent Neural Network(CRNN).The feature sequences extracted by CNN are sent into the Long Short Term Memory(LSTM)to deeply explore the correlation between ECG signals.The proposed method combines the memory advantages of the recurrent neural network and the sensing field advantage of convolutional neural network,and effectively realizes the automatic learning of ECG signal characteristics. |
