Research On The Feature Extraction And Classification Of CHD Heart Sound Based On EMD Correlation Dimension

Heart sound is a common human physiological signal that can accurately reflect the overall operating state of the heart.It is usually one of the main criteria for determining whether a heart is healthy or not.Auscultation of heart sounds is the main basis for diagnosing congenital heart disease,but auscultation is susceptible to subjective factors such as the environment and the auscultation doctor.Echocardiography is the most effective method for diagnosing heart disease,but it is a device for patients in remote areas.Expensive and difficult to afford medical expenses.Therefore,the feature analysis and processing of heart sounds is of great significance in the diagnosis of congenital heart disease.It also provides new ideas for the implementation of machine-assisted auscultation and facilitates the diagnosis of congenital heart disease.Heart sound is a kind of non-linear and non-stationary random signal.This paper uses fractal theory to analyze the heart sound,fully reveals the intrinsic characteristics of this type of signal,focuses on the study of heart sound characteristic parameters,understands the heart’s operating mechanism and combines normal and abnormal heart sounds.specialty.The main research work of this paper is:1.Heart sound signal preprocessing work.Wavelet analysis is used to eliminate the noise of the heart sound.Then the Hilbert transform method is used to extract the envelope of the heart sound.Finally,the extracted envelope is segmented using a dual-threshold method to determine the cardiac cycle of the heart sound.2.Empirical Mode Decomposition(EMD)of heart sound signals.The heart sound is decomposed into a series of intrinsic mode functions(IMF),and the instantaneous frequency characteristics of each order IMF component are analyzed to accurately reflect the details of the original signal.The cross-correlation method is used to filter the IMF component.Derive the main components of IMF1~IMF5.3.Extraction of feature parameters of heart sounds.The main components of IMF1~IMF5 of the preconditioned heart sounds are extracted and their correlation dimensions are obtained as characteristic parameters.Its changes can describe in a timely manner the inherent rules and complexity of changes in heart sound signals.Finally,the experiment shows that the parameters have good results in heart sound recognition..4.BP neural network identification classification.A BP neural network model was established,and the extracted heart sound feature vectors were imported into the network for classification and recognition,and the recognition rates of normal and abnormal heart sounds were obtained.This paper analyzes heart sound signals of 366 patients with congenital heart disease and normal cases.The results show that the correlation dimension can better reflect the details of heart sounds,including the law of change and highlight the key content,using BP neural network recognition,effective classification of heart sounds,The recognition rates for normal and abnormal heart sounds were 84% and 79.2%,respectively,and the average recognition rate was 81.6%.