Research On Analysis And Classification Method For Heart Sound Signal

Cardiovascular disease has long been one of the diseases with serious threat to human health, and its incidence has also been improved. So improving the early diagnosis of cardiovascular disease and the diagnosis rate is a very important task. As one of the most important physiological signals of the human body, the heart sound signal contains a large number of different pathological information about the various parts of the heart such as the Atrium, Ventricle, great vessels, cardiovascular and valves function, and it can reflects the status of mechanical motion of the heart and great vessels. So heart sound detection is an important means to understand the status of the heart, with irreplaceable clinical value of ECG detection.Heart sound signal is a time-varying, non-stationary complex signal. It needs to be analyzed from three aspects that time-domain, frequency domain and time-frequency domain for a comprehensive understanding of its characteristics. To ensure the quality of the analysis of heart sound signals, in addition to the hardware filtering, the pre-treatment before the analysis is also required, because of the existed interference in the collecting process.In this paper, Auto-Regressive Model, Short-time Fourier Transform, Wavelet Transform, Hilbert-Huang Transform (HHT) and Support Vector Machine (SVM) are used to analyze heart sound signals. The principal tasks are as follows:1. De-noising. A wavelet threshold de-noising method with double variables has been introduced. Comparing with the other methods, it proves this method de-noises heart sound signal better.2. Envelope extraction. This paper uses the Empirical Mode Decomposition (EMD) envelope algorithm for the extraction of heart sound envelope. It lays a foundation for the positioning and time-domain characteristics extraction of the first and second heart sound.3. Feature extraction. In this paper, the HHT method is used to obtain the frequency distribution at any time of the heart sound signal, the mean and variance of the instantaneous frequency as frequency-domain characteristic parameters. The eigenvector is composed by all the parameters extracted from the time and frequency domain. 4. Classification. After the principles of SVM is described, it is used to the classification of normal and abnormal heart sound diagnosis with correct classification rate of 98.9% for normal and 92.3% for abnormal.