Research On Analysis For Heart Sounds

The heart is one of the key organs of human bodies and each component of heart sounds reflects important information about the cardiac status. Nowadays, cardiovascular disease has seriously affected the human's health and sometimes even causes death. With the development of modern community health care center and in order to help people to know the status of heart more quickly and conveniently, the computer-aided diagnosis for heart sound disease is becoming more and more desirable.In this paper, different kinds of method are used for analyzing heart sounds to extract some useful features both in time domain and frequency domain. Some basic works are listed as follows:1, De-noising. A novel de-noising method based on two parameters using new thresholding function in wavelet domain was used to remove background noise and white noise. According to the results, this function is better than others. The main components of heart sound will remain after de-noising step.2, Envelope Extraction. Original envelope was got according to Hilbert Transfer and Shannon Energy. Then the smoothed envelope was extracted using Cubic polynomial interpolation based on original Hilbert Envelope. Compare to Shannon Energy, the smoothed Hilbert Envelope is more accurate because time resolution does not reduced. At last, a morphological transform action that amplifies the sharp peaks and attenuates the broad ones is used for distinguish heart sounds and noises.3, Segmentation and feature extraction in time domain. Segmentation for heart sounds is the basic and key step time domain analysis. Some strategies were taken to get the correct locations of main components of heart sounds in time domain, especially a new method based on window period and heart sound characteristics for recovering lost peaks was discussed. In order to recover the error detection for boundaries of heart sound in some situation like large noise at the beginning of systolic period, the boundary detection method was improved by taking heart sound split and murmurs into account. Based on the segmentation result, some features in time domain is extracted like S1 duration, S2 duration, systole, diastole, S1S2 interval, S2S1 interval and heart rate.4, Power Spectrum Estimation and feature extraction in frequency domain. Classical PSD (Power Spectral Density) estimation based on Welch method is used in normal and abnormal heart sounds for extracting some features in frequency domain like maximum of PSD, the frequency of maximum of PSD and three frequency band energy proportions.