Research Information System
1st International Symposium on Multidisciplinary Studies and Innovative Technologies, Tokat, Turkey, 2 - 04 November 2017, pp.155
Aim of the study: The heart sound segmentation is the first step of heart sound examination and plays in highly essential role in definition of pathological situation of heart diseases using heart sound. Besides, heart sound segmentation is very important to research similarities between ECG signals and heart sounds. In this study, different heart sound segmentation features have been comparatively analyzed on different heart sounds to extract fundamental heart sounds as S1 and S2. By doing this process, parts of heart sound have been prepared to reduce dimensions. It is planned that dimensions reduced parts is going to be used in artificial or convolution neural network for classification with parts of ECG with the aim of figure out similarities heart sounds and ECG.
Material and Methods: The five segmentation features that mostly used are selected by researching on literature. These features are as follows; Shannon Energy, Wavelet Coefficients, Spectral Features, Regression Coefficients and Dynamic Thresholding and each of which have been applied on the different heart sounds type like normal, abnormal, extra systole from different open heart sound database. Firstly, the fundamental heart sounds in different heart sound databases have been separated to S1 and S2 part physically and these parts have been marked by experts. Secondly, the heart sounds have been pre-processed for re-sampling and normalization. The Shannon Energy of heart sounds has been obtained for finding the exact boundary samples of each heart sounds. After obtained Shannon Energy features, Dynamic Thresholding has been applied in order to find the location of the S1. The Mel-Spaced filter banks has utilized for extracting spectral characteristics from the heart sound signal. The Regression Coefficients have been operated to obtain the changes in each features.
Results: The segmentation results show that if these segmentation features are applied stand-alone basis on heart sounds, the meaningful segmentation won’t be actualized. But some of these features is used together, it has been shown that more meaningful segmentation will be obtained.