Blind separation of heart sounds

Abstract

This paper presents blind separation of heart sounds by de-convoluting the directly measured mixed signals using the temporal Green’s function or the point source separation (PPS) method (Wu and Zhu, JASA, 2012). The objective of this study is to examine the feasibility of using a practical and cost-effective method to separate the aortic, pulmonic, mitral and tricuspid sounds that are involved in the first and second heart sounds, respectively. It is emphasized that because many parameters such as the locations from which the heart sounds are originated, as well as the speed at which heart sounds travel inside a human body are unknown a priori, it is unrealistic to expect a perfect separation of the heart sounds. To begin with, we conduct a numerical simulation test that uses an iteration scheme to locate sources and determine the speed of sound. This is done by applying PPS algorithm repeatedly under different source locations and speeds of sound. Results show that this iteration always leads to a convergent range of source locations and speeds of sound, from which approximate values of source locations and speed of sound can be determined. Once this is completed, the signals from individual sources are separated by de-convoluting the mixed signals with respect to individual source locations. This blind source separation methodology is then applied to the heart sounds measured on volunteers to separate the aortic, pulmonic, mitral, and tricuspid sounds.