Classification of Heart Sounds with Re-sampled Energy Method

Abstract

Due to the high number of heart diseases in the world and the increasing number of deaths resulting from them, the studies about early diagnosis of heart diseases has also increased. Studies on cardiac signals focus especially on the classification of heart sounds. Naturally, researches are generally concerned with increasing classification success. For this purpose, many studies use the separation of heart sounds into S1 and S2 segments by methods such as shannon energy, discreat wavelet transform and Hilbert transform. In this study, the use of signal energy, which is used as a sub-method of segmentation, has been explored for directly classification. For this purpose, the energy of the heart sounds was calculated by the re-sampled energy method and the obtained data were classified by artificial neural networks. Especially in these studies, the effect of sampling time on classification success was investigated. The obtained results were compared within themselves and with classification by dataset of S1-S2 sounds segmented. It is show that if the sampling rate is reduced in the direct classification by the re-sampled energy method, the accuracy of the classification is seen to increase. In addition, accuracy of classification is higher than the classification made by the segmented data set. Here I have reached the conclusion that the energy of heart sounds used to segment heart sounds can be used directly in classification studies and a more efficient classification performance can be achieved.