Hidden Markov model with heart sound signals for identification of heart diseases

Heart auscultation still remains a dominant method for the diagnosis of heart diseases caused by heart valve abnormalities. But it is very subjective and significantly relies on the interpretation or perception of well-trained physicians. Thus it would be very desirable to develop a computer-aided automated or semi-automated heart sound identification system that can provide more objective diagnositic results. Recently a hidden Markov model (HMM) has been used quite successfully for the classification of heart sounds. In this paper, we have investigated the classification performance of the MFCC-based HMM with heart sound signals by varying the model’s number of states, number of mixtures, and analysis frame size in MFCC feature extraction. We carried out the classification experiments using the 325 heart sound data made up of 10 different types of heart sound signals. From this, maximum correct classification rate of 95.08% was achieved when the HMM has 4 states, 8 mixtures with analsys frame size of 20ms for feature extraction.