Network Biology (Dec 2023)
Pattern classification of human body's acupoints based on functional similarity in Traditional Chinese Medicine
Abstract
Based on previously constructed data tables of acupoint functions and meridians, we performed a pattern classification of acupoints based on functional similarity in this study. Unsupervised pattern classification methods include k-means clustering, hierarchical clustering, BP neural network, self-organizing feature map network; supervised pattern classification methods include BP neural network, and LVQ network. Supervised pattern classification helps to determine the correspondence between acupoint functions and meridians, and to further explore the meridian properties of acupoints. Among them, hierarchical clustering gives two categories of pattern classifications, that is, fixed number classes of pattern classification and cluster tree, and other methods give pattern classification of fixed number classes. The cluster tree of acupoints reflects hierarchical relationship of functional similarity between acupoints, which can determine the functional relationship between acupoints at different levels. In unsupervised pattern classification, k-means clustering gave the most reasonable pattern classification, followed by BP neural network, and the results of hierarchical clustering and self-organizing feature map network were poor. In supervised pattern classification, LVQ network is superior to BP neural network, and self-organizing feature map network is poor. Compared with 15 categories of meridians, the LVQ network divided 13 classes of acupoints, the BP neural network divided 6 classes of acupoints, and the self-organizing feature mapping network divided 83 classes of acupoints. The pattern classification results obtained have guiding significance for selection, collocation, diagnosis and treatment of acupoints, and can be used for further in-depth mining and analysis.
