A Study Of Lung Sound Extraction Based On Blind Source Separation

In modern medicine,auscultation is still an important method to diagnose respiratory diseases.However,heart beat is an unavoidable source of interference during lung sound recording.Removing heart sounds(HS)from lung sound recordings is a challenging task but of great interest.Blind Source Separation(BSS)is a technology to separate or estimate the original sources from the observations.Nowdays,plenty of researches have been done in this domain,and it is widely used in the separation of speeches,communication and biomedical signal processing, image processing and etc.In this study,two models of mixtures of Heart Sound and Lung Sound are established: instantaneous linear mixtures and convolutive mixtures.Given the fact that the Heart Sound and Lung Sound are nonstationary signals with Short-Time Stationary structure,we proposed two algorithm:Blind Separation of instantaneous mixtures based on the nonstationary and second-order statistics,Blind separation ofconvolutive mixtures in time-frequency domain.In blind separation of linear mixture,the performance is largely dependent on the proper selection of time-frequency points which are used for joint diagonalization.On the basis of former researches, we propose a criteria of time-frequency point selection using local maximum.In the same way,we improve the method of mixture deconvolution in frequency domain,which is achieved by joint diagonal ization of the time-frequency presentation of mixtures,with a new cost-function:least square function.Througth emulational experiments,they are proved to be of better performance,compared with other algorithms.Finally,by applying these algorithms to a set of data collected on the surface,it denotes that the convolutive model and its corresponding algorithm are more effective than the linear model and its algorithm.Compared with other Lung Sound extraction methods,the Heart Sound is also extracted when extracting the Lung Sound by using Blind Source Separation method which would be of great interest for both groups of researchers.Therefore,this research has duple significance.