| Heart sound, as one of the most important physiological signal of human body, has great significance for the diagnosis of the cardiovascular disease. However, the recognition of the heart sound is still unsatisfactory because of its complexity and non-stationarity, which is one of the factors preventing it from extensive clinical application. This paper firstly introduced the physiologic basis of the heart sound. Then its features were analyzed for recognition of the heart sound and the diagnosis indicators for cardiac diseases were extracted. Finally, the heart sound database was designed and the software system of the analysis and recognition of the heart sound was developed.The feature analysis of heart sound was implemented from the time-domain, frequency-domain and the time-frequency domain, respectively. And the Welch method to estimate the power spectrum and STFT (Short Time Fourier Transform) of heart sound and the murmur were conducted. The results indicated that the frequency range of normal heart sound components and that of the murmur was different, which provided basis for murmur separation from the frequency perspective. In addition, the feature in time domain of the heart sound had certain regularity, and the information could be got for segmentation and location of the heart sound according to the time orders and amplitudes of its components.The murmur was isolated on basis of the principle of wavelet threshold method considering the time-frequency characteristics of the heart murmur, because all of the pathological signal samples were heart murmur signals in this study. In order to keep the pathological information integrity, heart sound analysis is on the basis of each cardiac cycle. The Mel-frequency cepstral coefficient and hidden markov model, which is commonly used in voice signal recognition, were used for heart sound recognition. Heart sound is not simply divided into normal or abnormal, but classified according to the type of various heart diseases. The result shows that this method is able to recognize the heart sound efficiently, the recognition rate as high as 94.2 percent, and superior to BP neural network (92.2% vs 82.8%) in heart murmur recognition. And then both the cardiac reserve indicators and indicators, i.e., the first heart sound energy fraction (S1EF) and murmur energy fraction (MEF) of heart murmur, were extracted for the heart disease auxiliary diagnosis. Statistical analysis of the above mentioned indicators, we discovered that cardiac contractility reserve and heart rate reserve were mobilized in the stress state, result in D/S ratio decreasing, S1/S2 ratio rising and HR higher; Statistical result showed that MEF is higher and S1EF is significantly lower (p< 0.01) in the abnormal group (heart with reflux murmur) compared with the normal group. So the findings suggest that the described indicators may have the potential to be used to evaluate cardiac energy and assist doctors for a more objective diagnosis.Finally, the heart sound database was built based on the SQL Server with the interface techniques—ADO(Active Data Objects)in Visual C++6.0, which can realize the modern management of the basic information and the heart sounds of the patients. And the software system of the heart sound analysis and recognition was developed based on the heart sound database with MATLAB hybrid programming technology. It had the functions of spectral analysis, wavelet transform, envelope extraction based on the normalized average Shannon energy and extraction of cardiac reserve indicators of the heart sound. |
PDF Full Text Request