Automatic heart sounds segmentation based on the correlation coefficients matrix for similar cardiac cycles identification

This paper proposes a novel automatic heart sounds segmentation method for deployment in heart valve defect diagnosis. The method is based on the correlation coefficients matrix, calculated between all the heart cycles for similarity identification. Firstly, fundamental heart sounds (S1 and S2) in the presence of extra gallop sounds such as S3 and/or S4 and murmurs are localized with more accuracy. Secondly, two similarity-based filtering approaches (using time and time-frequency domains, respectively) for correlated heart cycles identification are proposed and evaluated in the context of professional clinical auscultated heart sounds of adult patients. Results show the superiority of the novel time-frequency method proposed here particularly in the presence of extra gallop sounds.