The Study Of Heart Sound Analysis And Recognition Algorithm

Detection and analysis of heart sounds is an important and economical method that can be used to judge the state of heart and great vessels. Phonocardiogram(PCG) has its own advantages that ECG can't replace. ECG is the best way to monitor cardiac inotropism and chronotropism, but it is helpless to estimate dromotropism. Amplitude of the first heart sound is the normal measurement of cardiac contractility. To get amplitude parameter, the first step is to localize the different components of PCG. This paper, based on the summarization for prevenient research, presents some improvements for analysis and recognition of PCG, and validates them by tests and models.In the course of recording heart sound, it is inevitable that many kinds of noise will be merged in the main signal. Before further processings of phonocardiographic records, noise must be suppressed first. PCG is a highly nonstationary signal, so the paper introduces the application of adaptive filter method to the elimination of noise. This method needs a reference noise as one input, but it is unpractical to get the background noise in the original data condition. So, this paper uses mathematical morphology theory to catch heart sound envelope, and then composes the reference noise which is heavily correlative to the noise in the primary PCG signal. Pure heart sound can be attained after adaptively filtered by the compositive noise in the model created in Simulink. The aim of this course is to make the boundary of heart sounds ingredient and clearance plot more perspicuous, so we call this process preliminary denoising.After denoising, it analyzes PCG signal by Short-Time Fourier Transform(STFT). Different window-width is corresponding to different time-frequency resolution. It chooses the spectrum data with very good time resolution, and then takes the maximum element of the spectrogram at every time point through the frequency axis, so the temporal energy envelope of PCG signal is generated. From this envelope plot, heart sounds can be observed very intuitionisticly. Using the temporal energy data, and connecting traditional difference method with medicinal general knowledge, the thesis makes the recognition for heart sounds. The result proves this method is available and effective. To know cardiac contractility reserve, the amplitude of S1 must be gotten. In one record sample, the first heart sounds are related each other. With this kind of correlation, one S1 is applied to estimate the next S1 in a LMS denoising model created in Simulink. The result shows that after denoising, the amplitude of S1 is much closer to the true value than it of S1 before denoising. In actual application, the real data of S1 is impossible to know, but the amplitude after denoising can be taken as a replacement or approximation of real pure sample.