The Research And Application Of Cardiorespiratory Sounds Separation Meathod Based On Blind Source Separation

In modern medicine, auscultation is an effective noninvasive medical way for examining the cardiorespiratory system. Cardiopulmonary tone signal can be obtained by auscultation, which provides useful information about the behavior of the heart and lung organ. Cardiac and respiratory sounds contain abundant heart and respiratory physiology and pathology information. However, in actual clinical examination, sound signal collected by classical stethoscope includes the heart sound signals, as well as lung sound signal and external ambient noise. Cardiac and respiratory sounds interfere with each other as well as with other sounds like snore, speech or traffic noise, which compromises the effectiveness of auscultation. The cardiac and respiratory acoustic sounds also interfere in time as well as in spectral contents, which hampers the diagnosability of the classical stethoscope. How to decompose the auscultation signal into an independent heart sound and lung sound signal has been identified as blind source separation problem.Blind Source Separation (BSS) also to be called Blind signal separation technology has been widely applied in the speech signal processing, image processing, medical signal processing, the earth signal processing, radar signal processing, communication signal processing as well as other areas. What is a blind source separation? The so-called blind source separation, which described as follows:1) the mixing process and the mixed signals are unknown; 2) how to recover or estimate the source signals from the observed mixed signals collected by sensors. Under certain assumptions, this seemingly impossible problem has made great success. At the same time, a lot of innovative and effective solutions are proposed by researchers.In this paper, we propose a new method for effective auscultation by blindly recovering the original cardiac and respiratory sounds from a single observation mixture. The method we proposed which utilizes non-negative matrix factorization, cluster analysis and time-frequency mask technology. We successfully separated the hybrid clinical auscultation sound signal into two indepent acoustic signal (heart sounds and lung sounds). The method is divided into three phases as follow:1) a decomposition phase, which is the decomposition of mixture into independent and non-redundant components based on NMF. For the mean time, we have carried on the time-frequency transform domain to mixture signal by using short-time Fourier transform (ISTFT); 2) a clustering phase, in order to group the decomposed components into original sources, a new unsupervised technique is proposed. We have successfully divided the training data provided by the second phase into two categories through our clustering algorithm: cardiac and respiratory spectrometry signal; 3) a reconstruction phase, where the original sources are recovered from the spectrogram of the mixture using time-frequency masking and inverse short time Fourier transform. This effective auscultation method is successfully applied to actual data from an open data set which collected from different subjects in different clinical settings. We also made a simulation experiment of the actual simulation data simultaneously using mathematical analysis tools-MATLAB. The results we have obtained from the experiment have proved the feasibility of our method. At the same time, in order to verify the practical application feasibility of blind source separation of the cardiopulmonary tone, we have made an Android application and achieved ideal results. We also introduce the relevant technology of the cardiorespiratory blind source separation, which introduces the classification of blind source separation techniques, independent component analysis and single-channel blind source separation problem.