Separation Of Cardiopulmonary Sounds And Its Application

The normality of heart sounds and lung sounds can directly reflect the health of the human heart and lungs.The use of stethoscope for auscultation is a traditional,non-invasive,low-cost diagnostic method,but in real life,the doctor directly listens to the sound signal obtained through the stethoscope,which is often a mixed signal of heart sound and lung sound,and is also mixed with environmental noise,which seriously interferes with the doctor’s diagnosis.Therefore,finding a reliable method to separate the cardiopulmonary signals can help doctors to make more accurate diagnosis of the patient’s lesions,greatly improving the efficiency of treatment.How to separate the cardiopulmonary mixed signal into independent heart sound signals and lung sound signals is a blind signal separation problem,that is,separate signals are separated without knowing the aliasing condition of the signals.This paper provides a method for the separation of cardiopulmonary sounds based on the existing cardiopulmonary sound separation method.By combining the existing blind source separation method with the non-negative matrix decomposition method with label constraints,The a priori information of the cardiopulmonary sound is added to the algorithm of cardiopulmonary separation in the form of a label,so that the decomposed mixed signal has been initially clustered,which improves the accuracy of the separation.In the implementation of this method,short-time Fourier transform,non-negative matrix factorization(NMF),non-negative matrix factorization with label constraints(CNMF),clustering,time-frequency mask reconstruction and other algorithms are successively used.This method is mainly divided into four stages: 1.Signal transformation stage—The short-time Fourier transform is used to transform the time-domain mixed signal obtained by auscultation to obtain a sparse signal in the time-frequency domain.2.Decompositionstage-Decompose the signals in the time-frequency domain with non-negative matrix decomposition and non-negative matrix decomposition with label constraints,respectively.3.Clustering stage--A clustering algorithm is used to cluster the decomposed base matrix and the reference base matrix.4.Reconstruction stage—The independent heart sound and lung sound time-frequency signals are obtained by time-frequency mask technique,and then the time-frequency domain signal is transformed into time-domain signal by short-time Fourier inverse transform,and finally the cleaned heart sound and lung sound time domain signal are obtained.In addition,in the experimental part,different data sets were selected.Under the same conditions,several different separation methods were simulated,and the relevant waveforms and corresponding separation accuracy indexes were obtained to verify the feasibility of the method and the accuracy of the separation.