Research On Classification Of Pulse Signal And Blood Flow Signal For Pulse Diagnosis

Pulse diagnosis is one of the most distinctive diagnosis methods among the traditional Chinese medicines. In the past years, various civilizations have used arterial pulse as a guide to diagnose or to treat variety of diseases. These traditional medicines have reasonable content and rich experience. However, they did not make full use of the progress of science and technology. Therefore, it is important and necessary to apply the advantage of modern science and technology to pulse diagnosis. Among these work, pulse diagnosis digitization is the key to develop pulse diagnosis.In this paper, based on the analysis of pressure-based pulse signals and Doppler Ultrasonic blood flow signals, we have study the methods on the pulse patterns classification, pulse diagnosis by using the pressured-based pulse signal and the diagnosis method based on blood flow signals.By using the technique in time series classification, we have proposed two ERP-based pulse waveform classification methods. The experimental results shows that, comparing with the existing pulse waveform classification method, our methods have good performace and could further promote the classification accuracies. Besides, in order to solve the problem in the kernel machine-based time series classification mehtods, we propose a new kind of elastic kernel function, Gaussian elastic metric kernel (GEMK) function. This kind of kernel function inherits the elastic metric in being robust with noise and can handle the local time axis distortion problem. Experimental results show that, this kind of kernel can not only be used in pulse signal classification, but also could be used in other type of time series classification.We also have done the research in pulse diagnosis by using pressure-based pulse signals. We have extracted the time domain features and wavelet packet energy features of pressure-based pulse signals. By using the two kinds of features, we propose a fusion method of using k-nearest neighbor classifier and SVM classifier. Experimental results of classifying healthy people, cholecystitis and nephropathy show that our mehtod could be used in the diagnosis.In this paper, we have also proposed to use Doppler Ultrasonic sensor to study the blood flow signal in the radial pulse, and we have proposed a feature extraction method based on Hilbert-Huang transform. Experimental result in classifiying healthy, nephropathy, cholecystitis have proved effectiveness of this kind of method, and the effectiveness of using blood flow signal for pulse diagnosis and it also proved the blood flow singal could be used in the diagnosis directly.