Separation of cardiorespiratory sounds using time-frequency masking and sparsity

Abstract

Listening to cardiac and respiratory sounds called as auscultation is a non-invasive medical procedure, which provides useful information about the behavior of the heart and the lung. 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. This paper addresses the problem of auscultation in complex auditory environments, inspired by the coincidence detection model which suggests sound localization via estimating interaural level difference and interaural time difference. The proposed method, exploits the sparsity of cardiac and respiratory sounds and makes use of a degenerate unmixing estimation technique (DUET), which uses only two observations to recover an arbitrary number of sources, which suits well in scenarios where the number of sources can vary. The DUET approach uses time-frequency analysis to produce a two dimensional histogram of attenuation-delay estimates, where peaks in the histogram indicate the sources in a mixture. A mask is computed using attenuation-delay mixing parameters to recover the original sources. It is shown that excellent time-frequency masks exist for cardiac and respiratory sounds. The performance of the proposed method is demonstrated through a series of experiments using real data, exhibiting superior source recovery than previous techniques.