A sub-band energy tracking algorithm for heart sound segmentation

The objective of this paper is to present an algorithm for automatic segmentation of the heart sound. The algorithm utilises an autoregressive (AR) model to estimate the power spectral density (PSD) of the signal as well as the energy in certain frequency bands for consecutive overlapping frames. The starting and end points of each event are then calculated by filtering the tracking level using a morphological transform and estimating the boundary of its dominant peaks. The algorithm was tested for 960 cycles of heart sound recorded front all four popular auscaltatory areas of 30 patients. Results indicate the capability of this algorithm to isolate desired events in subjects with various pathological conditions.