Research On Segmentation Algorithm Of Heart Sound Based On Envelope Extraction

Cardiovascular and cerebrovascular disease has been one of the high prevalence in recent years, with the prevention to be difficult and the death rate to be high, Cardiovascular and cerebrovascular diseases has been the number one leading cause of deaths, especially for old people. Heart disease is one of the relatively common diseases in cardiovascular and cerebrovascular diseases. There are quite a lot means to detect heart disease, such as ECG、B ultrasound and Color Doppler Ultrasound,and so on. With all the aforementioned detection means wouldn’t give a clue until significant lesions happen in the heart, it’s not conducive to the early detection and treatment of disease. Another detection mean is cardiac auscultation, it can detect pathological information of heart disease before obvious lesions appear. However, there is no uniform and objective evaluation standard for cardiac auscultation, subjective experience of the doctor has to be relied. Combined with the physiological characteristics of heart sound, envelope extraction methods and segmentation methods of heart sound signal are used in this paper to extract relevant medical parameters, some objective basis for evaluating the clinical are also tried to be given.In this paper, different kinds of envelope extraction methods and segmentation methods of heart sound are studied, some medical parameters are extracted, and the GUI of segmentation of heart sound is developed. The main contents are as follows:1. The heart sound signals are preprocessed. Heart sound signal is resampled to reduce the amount of data and the pressure of data processing; five order Butterworth band pass filter is used to filter out the high frequency and low frequency noise; adaptive wavelet de-noising method is used to filter out noises whose frequency are same with heart sound signal; the strength of heart sound signal is uniformed with normalization method.2. A new envelope extraction method of heart sound which combining the normalized Shannon energy method and Hilbert Huang transform method is proposed. First, two common envelope extraction methods, normalized Shannon energy method and Hilbert Huang transform method are introduced, and their advantages and disadvantages are compared through experiments. In order to make up for deficiencies of the aforementioned two common methods, a new envelope extraction method of heart sound which combining the normalized Shannon energy method and Hilbert Huang transform method is proposed, in this method, by which mirror closed extension method is used to solve the problem of end effects. Experiments show that the new method gets better envelope.3. Heart sound signals are segmented and medical parameters are extracted. First, the principles along with the advantages and disadvantages of single threshold segmentation method are introduced. In order to make up for deficiencies of single threshold segmentation method, the double threshold segmentation method is studied. Experiments show that, compared with single threshold segmentation method, double threshold segmentation method gets better segmentation results. Second, several medical parameters are introduced and extracted based on the heart sound signals which have been segmented.4. The GUI of heart sound segmentation is developed. The general design principles and production steps of the GUI are introduced, and the GUI of heart sound segmentation is developed.Using the three envelope extraction methods and double threshold segmentation method described aforementioned,30 cases of heart sound signals which include 309 first heart sound and 304 second heart sound are analyzed in this paper. Segment on normalized Shannon energy envelope, the detection rate of S1 and S2 are 84.47% and 84.54%; segment on Hilbert envelope, the detection rate of S1 and S2 are 74.43% and 72.04%; segment on the envelope extracted by new envelope extraction method, the detection rate of S1 and S2 are 91.59% and 90.79%. Results show that the new envelope extraction method gets the best segmentation results.