Fractal Theory In The Analysis And Identification Of Heart Sound Signals

Heart sound is one of the most important voice signals of human body. It includes physiology and pathologic information of each part of heart, and reflects the mechanical movement condition of the heart and the cardiovascular system. When the cardiovascular diseases haven't developed to be good enough to produce clinical and pathologic change, murmur and aberration of heart sound is important information for diagnosis. Therefore, correct analysis and recognition of heart sound is significant to the diagnosis of cardiovascular diseases.This paper introduces fractal theory to the study of heart sound, and makes analysis and recognition of heart sound on the basis of fractal dimension, and provides a new noninvasive diagnostic for cardiovascular diseases. Firstly, this paper studies two kinds of fractal dimension estimation methods: variance fractal dimension and binary box-counting dimension, and verifies and compares the two methods. Next, it studies the fractal characteristic of heart sound, and draws a parallel between fractal dimension of normal and morbid heart sound. Then, this paper presents construction method of fractal dimension trajectory of heart sound, and discusses how time window parameters affect 'fractal dimension trajectory of heart sound, and analyses the characteristics of fractal dimension trajectory. Finally, a new automatic segmentation algorithm of heart sound is put forward by combining fractal dimension trajectory and time characteristics of heart sound, and the strategies and steps of segmentation are given, and the algorithm is tested with a large number of normal and abnormal heart sound data.The results show heart sound has non-scaling span and obvious fractal characteristics. Fractal dimension can quantitatively express the complexity of heart sound, and evidently distinguish normal and morbid heart sound. Automatic segmentation algorithm of heart sound based on fractal dimension trajectory is of high accuracy and robust without other signal as reference, and segment results are good.