Heart Abnormality Classification Using Phonocardiogram (PCG) Signals

Heart abnormality or disease is one of the leading causes of mortality worldwide. Sound signal produced by the mechanical activity of heart, known as phono-cardiogram (PCG), provides useful information about the heart's health. To increase discriminability among PCG signals of different normal and abnormal persons, an appropriate combination of signal features and classifiers is important. The segmentation of PCG signal, which requires corresponding ECG signal, is typically used for better prediction. But using ECG is generally expensive and time consuming. 781039In this paper, we therefore propose a segmentation free method to extract information from PCG signal. The signal is first preprocessed for DC removal and to limit the frequency to the required range. Four features (i.e. WPS, PS, FD, and SF) and four classifiers (i.e. LDA, ESVM, DT, and KNN) are then considered for the classification of heart murmur sound from PCG signals. A preliminary experiment with 56 signals showed the highest classification accuracy of 82.6%, obtained by simple statistical feature (SF) with ESVM classifier. On average, the best performing classifier was ESVM (accuracy: 77.17%), while the best feature was PS (accuracy: 75%). In addition, the PS feature showed stable and consistent performance irrespective of the classifiers used. Results also indicate the importance of combining multiple features and classifiers for better accuracy and reliability.