Research On Detection Method Of Coronary Artery Stenbsis Based On Multi-modal Features

In recent years,the incidence and mortality of coronary artery disease have been increasing gradually,and its prevalence in young people is also elevated,bringing great harm to society and country.Noninvasive and nondestructive accurate detection of coronary artery disease is an effective means to prevent coronary artery disease at an early stage and reduce the disease hazards,which is also a major challenge in the field of biomedical engineering.Coronary atherosclerotic heart disease is an important disease of coronary artery disease（accounted for about 95%to 99%）and coronary atherosclerosis causes narrowing or obstruction of the vascular lumen.Therefore,it is feasible to accurately detect the degree of coronary artery stenosis based on the joint analysis of multi-modal features.The combined analysis of electrocardiogram,phonocardiogram signals,with Holter,echocardiography,biochemical indexes,and other commonly used clinical examination results can describe the functional status and changes of the cardiovascular system in a more comprehensive way.However,there are few researches on this topic.In addition,the cardiac electro-mechanical coupling features contain information that can characterize the functional status of the heart,and studies of such coupling features in coronary artery stenosis are also rare.In this dissertation,electrocardiogram,phonocardiogram signal,and Holter,echocardiogram,and biochemical indexes results of patients with varying degrees of coronary artery stenosis were collected synchronously in the clinic to explore the performance of bimodal physiological signals and multiple examination results on the detection of coronary artery disease.On this basis,the combined analysis of electrocardiogram and phonocardiogram signals was systematically investigated at feature level and time series level with the aim of exploring the potential application value of multi-modal feature fusion in detecting the degree of coronary artery stenosis.The main works and innovations were as follows:（1）In view of the difficulty in distinguishing patients with coronary artery disease and patients with chest pain and normal coronary angiograms,a multi-modal information fusion model and hybrid feature selection algorithm were proposed.The recognition ability of electrocardiogram and phonocardiogram signals combined with Holter,echocardiography,and biochemical indicators was studied.Results showed that the Holter model had the best classification result among single-modal feature models;The current optimal feature model was all formed by adding new modal features to the previous optimal model.Compared with the non-physiological signal feature models,the classification result based on physiological signal feature model was better.The classification accuracy can reach 96.79%by using support vector machine.Compared with existing studies,it was found that multi-modal features fusion and hybrid feature selection algorithm can effectively obtain multi-source information for the identification between patients with coronary artery disease and patients with chest pain and normal coronary angiograms.（2）In view of the difficulty in identifying patients with different degrees of coronary stenosis,the influence of the coronary stenosis degree on the cardiac electrical and mechanical interaction function was systematically studied,and an analysis method based on entropy combination was proposed.It was confirmed that the multi-domain features of electrocardiogram and phonocardiogram signals contained important information to distinguish patients with varying degrees of coronary stenosis.It was found that the cardiac electro-mechanical coupling intensity decreased with the increase of the degree of coronary artery stenosis,which further indicated that coronary artery disease would lead to the synchronization disorder of cardiac electrical and mechanical activities of the patients.Clinical data studies indicated that the classification performance of electrocardiogram signals was superior to that of phonocardiogram signals in the detection of patients with varying degrees of coronary artery;The joint analysis based on cross entropy was significantly superior to the single coupling algorithm and the joint analysis based on non-entropy in identifying patients with varying degrees of coronary artery stenosis.Combining the cardiac electrical and mechanical features to detect patients with varying degrees of coronary artery stenosis,a classification accuracy of 86.90%.After the introduction of the cardiac electrical-mechanical coupling features,the classification accuracy reached 88.97%.The results confirmed that the combined analysis of cardiac electrical and mechanical features could distinguish patients with varying degrees of coronary artery stenosis.The cardiac electro-mechanical coupling features play an irreplaceable role in the detection of coronary artery disease.（3）In view of the limited ability of traditional coupling analysis methods to extract the cardiac electro-mechanical coupling characteristics of patients,an refined joint recurrence plot algorithm was proposed.Combined with the local binary pattern,the cardiac electro-mechanical coupling characteristics of patients was deeply mined,and the differences between patients with coronary stenosis were better captured.Clinical data showed that compared with the joint recurrence plot algorithm,the refined joint recurrence plot algorithm showed excellent performance in analyzing the electro-mechanical coupling characteristics of patients with varying degrees of coronary artery stenosis.The coupling features of the QT-systolic interval sequence extracted from the refined joint recurrence were the most effective in distinguishing patients with varying degrees of coronary artery stenosis.The multiple classification accuracy of patients with varying degrees of coronary stenosis was 86.90%only based on the refined joint recurrence plot algorithm,and the classification accuracy was improved to 90.34%when the cardiac electrical features,mechanical features,and electro-mechanical coupling features were added.The results showed that the refined joint recurrence plot algorithm can fully explore the potential pathological information in patients’ body surface signals,and provide a valuable tool for accurate detection of coronary artery disease,assessment of disease progression,and evaluation of treatment effect.