Research On Analysis Method For The Heart Sound Signal Of Congenital Heart Disease

Heart sound (HS) signal is an audible biomedical signal, which contains a lot of cardiac activity information. Domestic and international studies have shown that many heart diseases, especially Congenital Heart Disease (CHD), can be detected through cardiac auscultation. Today’s advanced electronic technologies can produce useful diagnostic information by analyzing heart sounds, extracting features and analyze them with methods of signal processing, such as pattern recognition. As the heart sound is a quasi-periodic and non-stationary signal, this analysis can be very challenging.This paper, focusing on the features of the heart sounds, the mechanisms and inherited characteristics, along with the pathological features of CHD, studies different signal processing algorithms at each stage and provides a full procedure for analyzing heart sounds. The procedure includes:First, heart sound pre-processing, which include HS fusion, HS de-noise and HS segmenting. Second, through pre-processing heart sounds are divided into two types:divisible positing HS and indivisible positing ones. For the first type, the heart sound are separated into two channels named HS dominant channel and murmur dominant channel using single-channel overlap partially sparse model; and then, six categories of features are extracted for further recognition and classification. For the second type, we introduce another procedure to analysis HS without segmenting, positioning and clustering. Finaly, the procedures are verified by processing clinical acquired data from the real world and the results are analyzed.The main characteristics and innovations of this paper are:●The application of template matching algorithm to the segmenting and locating operations during the heart sound pre-processing, and automatic determination of divisibility of the heart sound;●For the divisible HS, in order to highlight the characteristics of murmur, the algorithm of single channel overlapping partially sparse model is presented, included idea of separation and assumption. The separation algorithm based on this model can separate a single HS signal into HS dominant channel and murmur dominant channel;●Instead of experimenting with simulated data, hundreds of clinically collected HS signals are analyzed in this paper and, therefore, the conclusions are of great practical reference value. For experimenting, the HS signals of patients who suffer from atrial septal defect (ASD) or ventricular septal defect (VSD) are analyzed and them before and after surgery are compared; the method for quickly categorizing the signals is also presented.Heart sound analysis using the procedures presented in this paper is more comprehensive and detailed than other reference, and it can deal with various types of heart sound, whether it’s divisible or not. Compared with other heart sound analysis methods using220cases heart sound from real world, the method shown in this paper have higher ratio of correctly recognized normal and abnormal heart sounds, which have reached81.3%, this resdult is of great practical reference value.