Feature Extraction And Recognition Of Pulse Diagnosis Based On Information Processing Technology

Wrist pulse has been used to diagnose diseases in pulse diagnosis for thousands of years in Traditional Chinese Medicine (TCM), which, regarded by TCM, can reflect the condition of human body, especially the condition of cardiovascular system. However, in traditional pulse diagnosis, the beating of the pulse is felt at the measuring position of the radial artery through a practitioner’s fingertips, which means the diagnosis depends heavily on the practitioner’s skills and experience. To eliminate subjectivity in pulse diagnosis, which is not acceptable to many people, the development of computerized pulse signal analysis methods is necessary to standardize and objectify pulse diagnosis. Many studies have used a number of methods for pulse signal analysis such as time domain, frequency domain, and time-frequency domain analyses of pulse signals.Since most of these studies are based on the linear theory which might not analyze the pulse signal accurately because of the fact that wirst pulse is the output of the cardiovascular system which has been proved its involving a great deal of nonlinearity. In the paper, wirst pulse is analyzed by the two-elastic chambers model and by the modified nonlinear methods, such as windowed multiscale entropy, windowed recurrence quantification analysis and symbolic dynamics. Firstly, some methods are used for pulse waveform preprocessing; secondly, using the two-elastic chambers model for extracting the features of hemodynamics from wiry pulse and slippery and wiry pulse which result is agreement with the theory of TCM, and the modified nonlinear methods are using for extracting the nonlinear features and most of which exist the significant difference between different classes; thirdly, the correlation between the features of hemodynamics and the nonlinear features are analysis and obtained by using Bayes net; fourthly, some learning methods are applied to classify the pulse signal; finally, the pulse analysis system is built using C++in Visual Studio2005.The result of classification denotes that the modified methods which are applied to analysis are helpful for extracting the useful features for pulse diagnosis, especially for distinguishing the pulse signal, such as normal pulse, slippery pulse and wiry pulse, which recognized accuray reachs90.03%. Comparing with the classifying result from time domain and time-frequency domain analysis, it can be found this is a satisfactory result. Meanwhile, the correlation analysis gets the relationship between the features of hemodynamics and the nonlinear features, which also achieves our goal.