Heart Sound Signal Processing Methods Based On Theory Of Cyclostationary Signal

Heart sounds, which are the mechanical vibrations produced by interaction between the heart and heart hemodynamics, reflect the operation states of heart chambers, valves and large blood vessels, and have important application value for the assessment of the cardiovascular system function. Heart sounds (including murmurs) repeat in each cardiac cycle, which is called "cyclostationary characteristics" in signal processing theory. Based on the theory of cyclostationary signal, this dissertation systematically studied the quantitative evaluation of heart sound signal cyclostationarity, instantaneous cycle frequency of the heart sound signal, automatic heart sound segmentation and their applications. The main innovation points are as follows.(1) Quantitative evaluation of heart sound signal cyclostationarity and its applications were studied. The heart sound signal is not a strictly cyclostationary signal, but a quasi-cyclostationary one. According to the theory of cyclostationary signal, the statistic and calculation method of cyclostationary degree were proposed to quantitatively evaluate the cyclostationary strength of heart sound signals at any time. When contaminated by noise and interference, the cyclostationarity of heart sound signals were weakened. Thus, the degree of cyclostationarity of the heart sound signal, which reflected its quality, can determine how much noise or interference the heart sound signal had. The subsequence of a heart sound signal with no noise (or minimum noise) (i.e., optimum subsequence) can be automatically found with the statistic by eliminating those parts with more noise or interference. The method operated automatically, which solved a practical engineering problem in automatic analysis of heart sound signals by computer. Another noise reduction method of heart sound signals was put forward through enhancing the cyclostationary characteristics of heart sound signals. As long as the heart sound and the noise had no intersection in cycle frequency domain, the noise can be removed from the heart sound completely, no matter the noise was non-Gaussian, colored, or nonstationary.(2) Instantaneous cycle frequency of the heart sound signal and its applications were studied. Based on the theory of cyclostationary signal, the calculation method of instantaneous cycle frequency of the heart sound signal which reflected the instantaneous repetition frequency of the heart mechanical vibration was proposed. In this thesis, the instantaneous cycle frequency had three aspects of application. One was automatic detection of the cardiac cycle. Typically, the R waves of ECG signals were needed as reference signals while segmenting the cardiac cycle. This thesis estimated the cardiac cycle accurately with the instantaneous cycle frequency and the instantaneous phase which were obtained from the heart sound signals. One was to separate the heart sound signal from the mixed heart-lung sound signals. In the acquisition of heart sound signals, the breathing sound (lung sound) was one of the common interferences. In order to obtain pure heart sound signals, it was necessary to remove the lung sound. In view of the fact that the heart sound and lung sound were overlapped both in time domain and frequency domain, the application of time frequency domain separation methods were theoretically limited. However, the heart sound and lung sound were significantly different in cyclic domain. That is to say, they can be effectively separated based on the differences between their instantaneous cycle frequencies. Another was estimation of fetal heart rate (FHR). The widespread diagnostic tool to record FHR was ultrasonographic cardiotocography (CTG). There was no strong evidence if long applications of ultrasound radiations could be taken as absolutely harmless for the fetus. FHR was estimated by instantaneous cycle frequency of the fetal heart sound, which is accurate and no radiation. Experiments showed that these applications only need one signal source, i.e. the heart sound signal. No reference signal was required, which helped automatic analysis of heart sound signals on computer.(3) The automatic segmentation methods of heart sound signals were studied. In the analysis of heart sounds, the position of the heart sound often needed to be marked in the signal, namely segmenting the heart sound signal. Two automatic heart sound segmentation methods were presented in this thesis based on their cyclostationary characteristics. One was the cyclic envelope of heart sound signal based segmentation method. The other was the heart sound segmentation in a single cardiac cycle by using dynamic clustering. The experiment results showed that the segmentation accuracy rate reached more than 96%.Under the guidance of cyclostationary signal processing theory, the cyclostationary characteristics of heart sound signals and their applications were studied systematically and deeply. These researches and applications explored the cyclostationary characteristics of heart sound signals effectively and obtained valuable results, providing a powerful technical way for the automatic analysis of heart sound signals by computer.