Modeling And Classification Of Human Body Surface ECG Signals Based On Deterministic Learning And Deep Learning

Myocardial ischemia is a common cardiovascular disease,which seriously threatens people's life and health.Continuing myocardial ischemia can cause myocardial cell necrosis and myocardial infarction.Therefore,how to use simple and convenient methods in the early detection and early diagnosis of myocardial ischemia in the population is important.Compared with other invasive examination methods,body surface electrocardiogram has the advantages of non-invasiveness,simplicity,convenience,and economy.On one hand,the electrocardiogram is a comprehensive manifestation of the human heart electrical activity.The use of the electrocardiogram does not harm patients,and the operation is simple.On the other hand,ECG covers a wide area and deploy in almost every primary hospital.In the past ten years,many valuable works have emerged in the intelligent detection of myocardial ischemia based on ECG signals.Because of its important theoretical research background and practical application value,this paper will conduct an in-depth study on this subject.Based on the previous work,the following content is studied for the problem of feature extraction and classification of ECG signal:(1)An ECG modeling method based on deterministic learning is proposed.Accurately model the inherent dynamic characteristics of the state trajectory of the ECG signal based on the deterministic learning theory,and an effective ECG pattern recognition method is proposed.In this paper,the standard 12 lead ECG signal is first converted into low-dimensional signal,and then dynamic RBF neural network is constructed to accurately model the nonlinear dynamic pattern in ECG signals locally,and the obtained dynamic information is saved in the form of a constant RBF network weight matrix.In the classification stage,the similarity between the constructed estimator and the extracted dynamic information,the ECG signal classification is realized.This method has been experimentally verified on the PTB database.(2)A method for classification myocardial ischemia based on deep learning and deterministic learning is proposed.In order to improve the accuracy of early diagnosis and automatic classification of myocardial ischemia,this paper uses deep learning to perform deeper feature learning on the extracted cardiac dynamics data,and two types of neural network structures are proposed: convolutional recurrent neural network(CRNN)and parallel convolutional recurrent neural network(PRCNN),which has the advantages of convolutional neural network and recurrent neural network.The convolutional neural network extracts local features in the ECG signal,and the recurrent neural network learns the time dependence of the ECG signal sequence.The two neural network structures are used to train and learn the cardiac dynamics data,and achieve the classification of myocardial ischemia.A high recognition rate was obtained in the ECG data sets under the actual clinical environment,which verified the effectiveness of the proposed method.(3)A classification method for myocardial ischemia based on deep learning and transfer learning is proposed.In order to solve the problem of insufficient number of training data sets,this paper uses the CRNN and PRCNN network models designed in the previous chapter to pre-train on the source domain ECG signal data sets,migrate the pre-trained network model parameters as initialization parameters,and fine-tune some networks through the target domain ECG signal data sets,the network training speed is accelerated and the network classification performance is improved.Due to the great similarity between the target domain ECG signal data sets and the source domain ECG signal data sets,this paper proposes model-based transfer learning,conducts transfer learning on the proposed CRNN and PRCNN,and obtains a new fine-tuning network model.The performance of the network model is tested under the data sets of ecg signals in the target domain.