Computerized Wrist pulse signal Diagnosis using Gradient Boosting Decision Tree

In traditional Chinese medicine (TCM), pulse diagnosis is an important diagnostic method that has a long history and has been widely applied. Wrist pulse signals can be used to analyze a person’s health status, reflecting the pathologic changes of the person’s body condition. With regard to TCM pulse diagnosis, the However, the traditional diagnostic approach has been mainly based on the feel of the doctor, which is non-quantitative and subjective. This paper aims to present a new classification method is proposed for analyzing wrist pulse signals, to provide an automatic and quantitative approach for the diagnosis of TCM based on the pulse. Methods: First, the time domain analysis and hemodynamics method were used to extract and analyze pulse parameters. Then the filtering method was used to select all features. Furthermore, GBDT was used to classify and identify the pulse, and establish a model. Results: The wave peaks, wave valleys and time periods, pulse wave velocity and reflection factors are extracted by time domain analysis and hemodynamic analysis. Then, four important features, including h3/h1, h4/h1, w/t and Rf, were selected using the filter feature selection method. Then, the GBDT classification method was used to classify the pulse image of TCM. The middle GBDT classification method exhibited the best effect. The recognition accuracy of the sliding vein, chord vein and chord pulse was 90.33%, 83.52%, 97.74% and 78.60%, respectively, and the overall recognition accuracy was 90.51%. Conclusion: The parameters of the pulse map were optimized and the classification and recognition model of the pulse image was established to realize the automatic recognition of characteristics of pulse diagnosis in TCM. Based on the GBDT classification recognition method, a more accurate classification and recognition model of TCM was established.