Digital Feature Analysis Of Multiple Pressure Sensors Pulse Instrument

In Traditional Chinese Medicine (TCM), doctors always rely on the so-called four diagnoses, including inspection, listening, inquiring and palpation to give the diagnosis. In Palpation, an essential and fundamental method, the doctor always uses fingers to feel the beating of the artery and get knowledge of the patient's illness. Due to the subjectivity and fuzziness of pulse diagnosis in TCM, quantitative systems or methods are needed to modernize pulse diagnosis.During the data collection process, we use three multi-points pressure sensors integration systems, produced by Harbin Institute of Technology, to collect the raw pulse data. This system can synchronously detect the movements of radial artery at the positions of Cun, Guan, Chi, as well as the vascular width, using three corresponding pressure sensors. Pulse data from 30 students and 120 patients acquired by our device are carefully studied.When we acquire the pulse data, the high frequency noise would still be introduced by the interference of electromagnetic signals. Therefore, in data preprocessing we utilize smooth filtering to eliminate noise after local deburring. A threshold filtering algorithm is introduced to detect and remove another kind of noise efficiently, and a statistics weighted approach is proposed to remove the pseudo-peaks of pulse wave, which not only overcome shortcomings of the old algorithm but preserve the details of pulse completely. Finally, it lays a solid foundation for the feature extraction and classification.The feature extraction of pulse is important too. We make full use of the reliable information of each period by a weighted trust algorithm for the static features in time domain, extracting the features more exactly. In this algorithm, we give different trust or weight to each period of pulse wave according to its changes in preprocessing, based on which we calculate the mean value of all the weighted period as an estimation of this pulse.In the classification phase, generalized dynamic time warping (GDTW) algorithm is utilized in this thesis to classify five kinds of pulse patterns, and a new algorithm is introduced using multiple templates and the statistics information of all the periods to handle the limitation of single template. Compared with the old algorithm, our experiment shows that our system obtains relatively reliable predictions of pulse types, and the predictive accuracy of DanHua pulse reaches as high as 94.67%.