Digitalization of Pulse Force in Pulse Conditions of Traditional Chinese Medicine

Abstract:
Pulse force is considered to be a key element of pulse condition, but the numerical analysis of pulse force in previous studies was not based on objective data. In this study, the pulse data of 150 subjects are obtained using bionic pulse diagnosis equipment and examined by Traditional Chinese Medicine (TCM) experts to assess their pulse condition. We analyze 36 parameters of pulse condition. The results show that two main features, w/h1 and Maximum Angle decline (MAD) (c > 0.15, P < 0.05), can effectively reflect the changes of pulse force in different physiological states. In addition, we also use K-medoids algorithm and soft Dynamic Time Warping (soft-DTW) algorithm to identify the strength of pulse force. And the silhouette coefficient of a cycle of pulse wave clustering reaches 0.599. To further predict the corresponding pulse force value, we build a model combining Convolutional Neural Networks (CNN) with Long Short-Term Memory (LSTM) and Label Smoothing Regularization (CLLSR). And it can predict the pulse force score more accurately, and the accuracy rate can reach 97.4%, when pulse force score within ±10 by using CLLSR model. Therefore, this study provides an experimental basis for the digitalization of “Palpation”, one of the four diagnostic methods of TCM, which is “Observation, Ofaction, Inquiry, and Palpation”, and contributes to the development of more accurate and efficient disease diagnosis strategies in the future.