Study Of Heart Sound Automatic Analysis And Recognition Algorithm

Heart sound is one of the most important medical signals which contains important physical and pathological information. Heart sound analysis is the key tache in cardiac reserve noninvasive monitoring system based on the relationship between heart sound and cardiac contractility. A complete cardiac cycle can be divided into four parts: First heart sound (S1), Systole (S), Second heart sound (S2), Diastole (D). There are very few studies about S3 and S4 because they are so weak. Although the ECG analysis is the best monitoring method in analysis of cardiac chronotropic and dromotropic, it can not be used for detection of cardiac inotropic. The first heart sound's amplitude is the standard measure of the cardiac contractility, so PCG can be used as an assessment of cardiac contractility. If a certain part of the heart organ lesions occurred, the corresponding components of heart sound would response accordingly, such as coronary artery stenosis may cause diastolic cacophony in heart sound. It's necessary to subsection heart sound before further analyzing and processing the heart sound signal. Although heart sound has characteristics of a certain law, for various factors, its clinical manifestation is in the form of a highly complex, it's very difficult to locate heart sound accurately.In the course of recording heart sound, inevitably, there will be noises merging in the main signal. Before further processing the phonocardiographic records, noise must be suppressed first. Considering the nonstationarity of PCG, we analyzed the application of time-frequency,wavelet transform and mathematical morphology in heart sound denoising, and eventually adopted the method of lifting wavelet transform, which proved to have a good performance. We got the envelope of the denoised heart sound, and then finished the heart sound auto-recognition according to the physiological knowledge and some large sample clinical statistic results. After that, S1, S2, Systole and Diastole can be auto-distinguished which is very favorable for further processing and analyzing.In this paper, we use special methods to process different types of heart sound signals. By using the amplitude characteristics of S1 and S2, and the specificities of systole and diastole, we analyzed two kinds of abnormal heart sound which are arrhythmia and bigeminal rhythm, and specifically designed recognition methods for them. We used a large number of heart sounds to verify our arithmetic, and simulated the algorithm by using MATLAB-GUIDE toolbox. 212 heart sound samples, collected in the project supported by National Natural Science Foundation of China, were tested using this algorithm. The accuracy of recognition is up to 98%, specially, in the abnormal heart sound, the accuracy reaches 95%. The result shows that the algorithm proposed in this paper has a pretty good performance. It provides a supporting technology for further cardiac contractility variability analysis and clinical diagnosis.