Classification of heart sound based on multipoint auscultation system

Abstract

Heart disorder can be diagnosed by listening to the heart sound that is recorded using stethoscope on the human chest. However, human interpretation and diagnosis based on auscultation is somewhat subjective and vary depending on the skill and hearing ability of the physician. Studies have been focusing on the development and evaluation of methods in detecting the various components of the heart sound at a specific auscultation point. The principle interest of this paper is, however focused towards finding the optimal auscultation point which involves placing the stethoscope at different position namely at the aortic valve and pulmonary valve which provide better quality of the second heart sound component (S2) and mitral valve and tricuspid valve where the first heart sound component (Sl) can be heard more clearly. Comparative experiments using to Mel-Frequency Cepstrum Coefficient (MFCC) property, variation of the number of Hidden markov Model (HMM) states and variation of the number of gaussian mixtures were conducted to measure the offects of these factors to the classification performance at the four locations of auscultation point. Further works was also carried out with time-frequency distribution which is known to provide information about how the spectral content of the signal evolves with time. The Extended Modified B-distribution was chosen from a number of time-frequency methods due to its ability to represent the signal in the most efficient way in term of noise and cross term elimination.