The Method Of Cardiac Sound Localization And Cardiopulmonary Sound Separation For The Electronic Stethoscope

Cardiac and respiratory diseases are risk factors that threaten human health,and they causes a large number of deaths around the world every year.Therefore,early detection and intervention of cardiopulmonary diseases are of great significance.Auscultation is a non-invasive examination method for physicians to judge the cardiac and respiratory conditions of patients in the first time.In recent years,with the development of Internet of things technology,the development of electronic stethoscope makes the remote diagnosis and artificial intelligence diagnosis of cardiac and respiratory diseases possible.The cardiopulmonary sound signals recorded by the electronic stethoscope reflect the physiological information of the heart and respiratory tract,which are overlapped with each other and accompanied by noise and interference.Therefore,if the stethoscope algorithm can be used to output "purified" heart or lung sound,the efficiency of physicians’remote diagnosis or artificial intelligence diagnosis can be greatly improved.From the practical requirements of electronic stethoscope,the task of this thesis aims at the ultimate separation of cardiac sound signal and respiratory sound signal,and jobs in the following three aspects are carried out:Firstly,the cardiopulmonary sound signals preprocessing algorithm was studied.Cardiopulmonary sound signals collected by electronic stethoscope are mixed with electrical noise and various possible interferences,including background noise interference,compression clipping distortion and frictional sound.Wavelet denoising can be used for the elimination of electrical noise,and the adaptive filtering can be used for the elimination of background interference.These two methods are relatively mature,but there still lacks a feasible scheme for the automatic detection and elimination of addressing clipping distortions and frictional noise,which is the focus of this research point.In order to eliminate the distortion caused by compression clipping,a method involving difference matching to quickly locate the distortion area and Hermite interpolation to repair the distortion area is proposed.For the elimination of frictional noise,the frictional noise region is located by using the combination of Mel frequency cepstrum coefficient(MFCC)and support vector machine(SVM).Repairing the frictional noise interference region involves empirical mode decomposition(EMD)and component and region reserving based on correlation coefficient.The experimental results show that the proposed method can automatically detect and repair clipping distortion and frictional noise interference effectively.Secondly,the localization method of cardiac sounds in cardiopulmonary sound signal was studied.The traditional method of cardiac sound localization was based on entropy spectrum,utilizing the property that the Shannon entropy of heart sound segment is larger than that of lung sound segment.The entropy spectrum based method calculates the Shannon entropy of each data segment of cardiopulmonary sound signal,and compares it with preset threshold value to locate cardiac sound component.On the basis of this theory,a method is further proposed in this thesis to detect and identify the cardiac sound components that may be mislocated or missed based on the power ratio of the low frequency to high frequency components of each cardiac sound component,and relocalize the cardiac sound components that may be mislocated or missed in each cardiac sound cycle.Experimental results show that the performance of cardiac sound localization and segmentation under various signal acquisition environments can be effectively improved by the proposed algorithm.Finally,on the basis of the aforementioned cardiopulmonary sound preprocessing and cardiac sound localization,the cardiopulmonary sound separation method was studied.In this thesis,a method of cardiopulmonary sound separation by combined diagonalization is proposed based on the property that there are a large number of similar components corresponding to heart features between adjacent cardiac sound cycles.In the proposed method,the eigenvalues and eigenvectors of covariance matrices of several adjacent cardiac sound segments corresponding to the original signal and the reference signal are calculated by joint diagonalization.The correlation between eigenvectors of the cardiac sound segment and the reference signal is used to extract the feature vector that belongs to cardiac sound components.The separated cardiac sound signal and respiratory sound signal are ultimately obtained by using a constructed projection matrix.The experimental results show that the proposed method can not only separate the cardiac sound and respiratory sound for normal subjects,but also perform well in the cardiopulmonary sound separation for patients with respiratory diseases.In this thesis,the work was carried out in three aspects:cardiopulmonary sound preprocessing,cardiac sound localization and cardiopulmonary sound separation.The corresponding algorithms were designed progressively,and the separated cardiac sound and respiratory sound signals with the interference removed can be finally obtained.The work in this thesis has fulfilled the practical requirements of electronic stethoscope that can output separated cardiac and respiratory sound signals.The high-quality and low-interference cardiac and respiratory sound signals will guarantee the reliable implementation of subsequent remote diagnosis or artificial intelligence diagnosis.