Cardiac Signal Analysis Based On Deep Convolutional Neural Network

The diagnosis of heart disease often relies on the subjective judgment and personal experience of doctors,which is prone to deviation,thus delaying the treatment of patients.Therefore,computer-assisted diagnosis of cardiovascular diseases has become a focus area of current researchers.In order to focus on targeted research,this thesis mainly analyzes two kinds of heart signals and designs algorithms to realize the diagnosis of cardiovascular diseases.Convolutional neural network(CNN)and recurrent neural network(RNN)are used to analyze heart sound signals to realize the classification of cardiovascular diseases.Deep convolutional neural network is used to analyze and classify ECG signals.This thesis proposes a novel heart sound analysis method based on one-dimensional CNN and RNN,using convolution neural network learning characterization,then recurrent neural network processing timing information of heart sound signals,can improve heart sound classification accuracy at the same time reduce the network parameters,suitable for portable devices.Under the condition that the parameter is only0.05 M,the accuracy of this method is 92.96% on the adult heart sound dataset,97.48%on the infant heart sound dataset,and 96.76% on the infant heart sound dataset for ventricular septal defect disease.At the same time,end-to-end class activation map analysis is carried out for ventricular septal defect diseases,which proves that the heart sound signal characteristics learned in the diagnosis of ventricular septal defect diseases by the method in this thesis are in line with doctors’ experience in clinical heart sound diagnosis.Electrocardiogram(ECG)signal is a kind of temporal physiological signal,which mainly shows the process of depolarization and repolarization of cardiomyocytes.There is still strong subjectivity in clinical diagnosis of ECG signals.This thesis proposes a multi-scale ECG diagnosis method based on the ECG data provided by the PhysioNet/Cinc 2020 Cardiac Diagnostic Challenge.This thesis uses this method to participate in the PhysioNet/ Cinc 2020 Cardiac Diagnostic Challenge and ranks the 8th among more than 100 teams worldwide.