Session S44.2

Stationary Wavelet-Domain Phonocardiogram Segmentation Based on a Dissimilarity Measure

C Guarnizo-Lemus*, E Delgado-Trejos

Instituto Tecnológico Metropolitano
Medellin, Columbia

Murmurs in a phonocardiogram (PCG) represent some mechanical cardiac abnormality. Cardiac pathologies can be diagnosed from the location of a murmur with respect to the cardiac cycle. The location of a murmur in a PCG over a cardiac beats varies depending of the type of cardiac abnormality. Locating murmurs is a difficult task, due to they present a wide variation in their time and spectral features and duration across different patients. A method for PCG segmentation (murmurs and heart sounds) using a dissimilarity measure on stationary wavelet domain is presented. Each sub band obtained from stationary wavelet transform has the same length as input signal; this allows the realization of change point detection on different frequency bands without losing temporal information. The differences in the distributions of the variance over the principal components from wavelet coefficients are used as a dissimilarity measure for the abrupt change detection method. Maximum values which exceed a threshold are selected as change points. The database of PCG records which was used belongs to the National University of Colombia. Taking into account 360 PCG beats, where a set of 180 beats were strongly disturbed by different types of cardiac murmurs, segmentation yielded an accuracy result of 94%.

(Abstract Control Number: 238)