Research On Classification And Recognition Method Of Cardiovascular Diseases Based On Deep Learning

The incidence of cardiovascular diseases(CVDs)has been increased with the acceleration of population aging.Electrocardiogram(ECG)is the most common and basic method in the detection approaches to CVDs because of its non-invasive,economical and convenient.However,due to the large amount of ECG data,the traditional methods based on manual detection are time-consuming and subjective.Thus,automatical detection for abnormal ECG signals is highly desirable.In this paper,two ECG automatic recognition models are proposed to detect and recognize myocardial infarction(MI)and congestive heart failure(CHF),respectively,based on deep learning models.The main research contents are as follows:Aiming at the problem of noise disturbing,the wavelet transform algorithm is used to suppress the ECG noise.According to the characteristics of ECG signal,firstly the ECG signal is decomposed into 10 layers by Coiflets4 wavelet function.Secondly,the decomposed wavelet coefficient is processed by soft threshold function.Next,the inverse wavelet transformation is employed to restructure the wavelet coefficients to obtain the de-nosing ECG signal.Finally,the de-noising ECG is located by R wave to intercept the desired beat signal.A classification model based on multi branch convolution neural network is constructed for MI detection.Firstly,the multi-branch convolutional neural network is employed to extrct features.After that,the attention module is employed to fuse these features.At the same time,in order to solve the problem of data imbalance,smote algorithm is used to equalize the samples.Finally,the proposed model is tested by cross validation in the PTB diagnostic ECG database.Experiments show that the model achieves 96.16% accuracy in MI signal recognition task.An 1-D UNet++ network based on RR interval signals is proposed for CHF recognition.First of all,the residual network and inception model with Squeeze-and-Excitation(SE)module are introduced as the backbones of UNet++model to improve the feature extracting ability of the model.After that,the model with different backbones are tested by cross validation in the database of normal sinus rhythm RR database and congestive heart failure RR database.The experiments show that the UNet++ networks proposed have achieved good results in the detection of CHF.Among them,the residual network with SE module achieves the best accuracy87.79%.These experiment results indicate that the proposed methods are accurate and short time-consuming.Theses methods can achieve fast and accurate detection and identification of congestive heart failure and myocardial infarction signals,which is helpful to assist clinicians in diagnosis or decision-making.This thesis contains 55 figures,11 tables and 88 references.