Phonocardiogram segmentation by using Hidden Markov Models

Abstract

This paper is concerned to the segmentation of heart sounds by using state of art Hidden Markov Models technology. Concerning to several heart pathologies the analysis of the intervals between the first and second heart sounds is of utmost importance. Such intervals are silent for a normal subject and the presence of murmurs indicate certain cardiovascular defects and diseases. While the first heart sound can easily be detected if the ECG is available, the second heart sound is much more difficult to be detected given the low amplitude and smoothness of the T-wave. In the scope of this segmentation difficulty the well known non-stationary statistical properties of Hidden Markov Models concerned to temporal signal segmentation capabilities can be adequate to deal with this kind of segmentation problems. The feature vectors are based on a MFCC based representation obtained from a spectral normalisation procedure, which showed better performance than the MFCC representation alone in an Isolated Speech Recognition framework. Experimental results were evaluated on data collected from five different subjects, using CardioLab system and a Dash family patient monitor. The ECG leads I, II and III and an electronic stethoscope signal were sampled at 977 samples per second.