Study On Preprocessing And Classification Methods Of Arrhythmia ECG Signal

Electrocardiogram(ECG)is the most widely used physiological signal in clinical practice to diagnose cardiovascular diseases such as arrhythmia and myocardial infarction.For the electrocardiogram,the doctor’s manual interpretation is very time-consuming,and it is difficult to detect small changes in the long-term electrocardiogram or dynamic electrocardiogram records.Therefore,the use of computer-aided diagnosis(CAD)systems to automatically identify arrhythmia is a research hotspot,which can effectively reduce the mortality of heart disease patients.In terms of ECG signal preprocessing,ECG denoising has always been one of the main areas of research because effective denoising methods can perform good preprocessing on ECG signals.The ECG signal after denoising can extract the maximum amount of effective and meaningful pathological information,and can maximize the various indexes of the subsequent ECG classification process.This thesis proposes a new method of fusion of empirical mode decomposition(EMD)and discrete wavelet transform(DWT)to remove ECG noise by adaptive threshold,and optimize the denoising process through cultural algorithms.Based on the simulation experiment of MIT-BIH arrhythmia ECG database,MIT-BIH supraventricular arrhythmia ECG database and INCART arrhythmia ECG data,the performance evaluation index signal-to-noise ratio and mean square error are calculated.Aiming at the problem that the number of ECG data in the existing arrhythmia public database is small and the categories are unbalanced,this thesis proposes a data enhancement method based on an improved generative confrontation network.The experimental results show that the proposed method can effectively improve the problem of data imbalance.In terms of arrhythmia ECG classification,research is carried out from two directions based on traditional machine learning and deep neural network.The main contributions of this thesis are:(1)Extraction using discrete wavelet transform,empirical mode decomposition and variational mode decomposition(VMD)algorithms for the time domain and transform domain features in the ECG signal,the feature selection method is used to rank the features according to the weight of the feature,thereby reducing the computational complexity of the classification process and improving the classification accuracy.Using machine learning algorithms to classify the ranking set features into different types of arrhythmia symptoms,a new method is proposed,which includes multi-domain features based on RR interval,DWT,EMD and VMD,according to The AAMI standard classifies arrhythmia heart beats.The chi-square test and particle swarm search algorithm are used to rank the feature set,and the random forest classifier is used to classify the feature set.Finally,the classification accuracy is improved to 99.22% in the chi-square test.(2)A twostep classification method based on deep neural network is proposed to identify abnormal heart beats and detect arrhythmia.In the first step,a deep dual-channel convolutional neural network(D-CNN)is proposed to classify all heartbeat categories except for class S and normal heartbeats.In the second step,a central LSTM network(C-LSTM)is proposed to distinguish S heartbeats from normal heartbeats.The proposed C-LSTM network learns to extract the hidden time information between heartbeats and their neighboring heartbeats,and distinguish the deep difference between the two heartbeat types.(3)Aiming at the needs of wearable devices for lightweight models,low computing hardware requirements,and strong generalization capabilities,a Deep LSTM network based on the LSTM network is proposed.This sensitive and lightweight model for real-time ECG anomaly detection,combined with a specific and computationally intensive model for classification and diagnosis,is an effective method for ECG analysis in current hospital outpatient clinics.Through a large number of experiments,a comprehensive evaluation of the proposed method is provided.The results show that the study of ECG signal preprocessing and heartbeat classification in this thesis provides practical ideas and solutions for the detection of arrhythmia.