The Acquisition And Analysis Of Pulse Signal

Pulse signal is a vital physiological signal of human body, it contains rich diagnosis information of the circulation system. Pulse signal is an external indicators that can be felt directly from skin, thus the pulse signal acquisit ion is pain free, easy to gather and at low cost, thus, it had been adopted by the medical society since ancient time. Pulse diagnosis plays an important role in the traditional C hinese medicine(TCM), Ayurveda and traditional Egyptian medicine. In pulse diagnosis, practitioners put three fingers on the wrist of the patient to adaptively feel the fluctuations in the radial pulse at the styloid processes and analys is the health condit ion of the patient. However pulse diagnosis is a subjective skill which needs years of training and practice to master. Moreover, for different practitioners, the diagnosis results may be inconsistent. To overcome these limitations, computerized pulse diagnosis where sensors are developed to acquire pulse signals and machine learning techniques are exploited to analyze health conditions based on the acquired pulse signals has recently been studied to make pulse diagnosis objective and quantitative.The acquisit ion and analysis of pulse signa l includes the acquis ition, preprocessing, feature extraction and classification. O ur work also follows the four parts.The pulse signal acquisit ion is the basis of the pulse signal analysis. Recently, kinds of pulse signal acquisition syste m was reported, however, there are still some deficiencies to be improved. First, most current systems are lack of positioning system and requires the user manually place the probe to the appropriate posit ion based on the user’s experience, thus the acquisition process procedure is subjective and time consuming. Second, most of the current systems are lack of automatic pressure control system, the setting of the pressure is subjective and time consuming. Moreover, the current systems cannot measure the pulse width or sample from C un, Guan and C hi simultaneously. To overcome these limitations we develop our own pulse acquisit ion system and using this system we can locate the key position fast and set the hold down pressure to the preconfigured value automatically which make the acquisit ion process objective and fast. Moreover, our system can also measure of pulse width and achieves multi-channel signal acquisition.Pulse signal preprocessing process is to recover pulse signal with interference and to improve the quality of pulse signal. Thus the accuracy of feature extraction and classification will be improved. The existing preprocessing process usually focus on the denoising or baseline drift removal, however, some of the corruptions is hard to remove due to the loss of information or the overlapped frequency band of the interference and pulse signal such as saturation and artifact. These two interferences are not being processed in the current preprocessing process, which makes outliers imevitable in the pulse signal dataset. In this work we present a new pulse signal preprocessing framework and provide a saturation detection algorithm based on differential operator and artifact detection algorithm based on network connectivity.Feature extraction is another important problem in pulse analysis. The exist ing feature extraction methods which can be roughly grouped into two sub-categories: single cycle-based and whole signal-based ones. However, both single cycle- based and whole signal- based methods are limited in characterizing the inter-cycle variations of pulse signal which are useful in disease diagnosis. In this paper we presents three feature extraction method s to characterize the inter- and inner-cycle variance. Experimental result shows that the inter-cycle variations are important in the diagnosis of some disease may cause some changes in the inter-cycle variations. By taking both inter- and intra-cycle variations into account, the proposed methods achieve higher classification accuracy than the competing methods.For classification, we did comparative experiment on classification performance of different types of pulse signal and we present composite kernel based pulse signal classification method for the classification fusion. The acquis ition principle and their relationship of the three kind of pulse signal was analyzed and we found that different types of pulse signals are correlate with each other but have different sensitive factors. We think the diagnosis performance of different types of sign als should be different. If some disease is connected with changes in some factor, the sensor sensit ive to that factor may have advantage in the diagnosis of that disease and by combining all kinds of pulse signal we can obtain more useful diagnosis information. The experimental results prove our point of view. Ultrasonic signals can achieve higher classification accuracy than the other two types of signa ls in diabetes classification, while in the diagnosis of arteriosclerosis, the pressure sensor outperforms the others. To further increase the diagnosis performance we present the composite kernel based pulse signal classification method and by combining different types of pulse signal the improved accuracy was obtained.