Exploring the Stationary Wavelet Transform detail coefficients for detection and identification of the S1 and S2 heart sounds

Most work done in Heart Sound Segmentation approaches use a threshold-based approach to correctly identify S1 and S2 segments in a given signal. We propose a new method that uses the Stationary Wavelet Transform to segment the signal and hierarchical clustering to distinguish the S1 and S2 heart sound from noise. This approach was tested in the Classifying Heart Sounds PASCAL Challenge datasets and achieved better results than the winning approach of this contest, with a total error redcution of 21% and 43% for Digiscope and iStethoscope in test sets, respectively.