Algorithm of Heart Sound Segmentation and Feature Extraction

In this paper, an algorithm about heart sound signals segmentation was described, in which only heart sound signals were used. The algorithm was based on the normalized average Shannon energy of heart sound signals. 20 cases of normal and 20 cases of abnormal heart sound signals of heart valve disease selected randomly from the database of the Institute of Research Clinique of Montreal were used for testing. The features of each component, such as AR spectrum energy in systolic period were extracted after segmentation. The energy of high frequency is generally less than -95db to normal signal. Reversely, the energy of high frequency is generally more than -95db for abnormal signal. This feature is valuable in diagnosis of congenital heart diseases and other cardiac diseases.