Automated fundamental heart sound detection using spectral clustering technique

For automated analysis of heart sound the first and essential step is detection of the position of fundamental heart sounds (S1 and S2) within a Phonocardiogram (PCG) signal. In this study we propose an acoustic feature based technique for efficient localization of S1-S2 and non S1-S2 segments of a PCG signal. While envelop extraction based methods have been shown moderate successful outcomes, we have proposed a spectral clustering based technique which uses the power spectral density (PSD) as the input feature. The clustering algorithm has been utilized to group the signal into two clusters. Proposed method has been obtained a high true positive rate of 99.45% and low false alarm of 0.14%.