基于人工智能的癌痛中医辨证模型构建研究  

Study on Construction of TCM Syndrome Differentiation Model of Cancer Pain Based on Artificial Intelligence

作　　者：吴冰[1] 张柳柳[1] 郭永安[2] 钱琪杰 吴静 WU Bing;ZHANG Liuliu;GUO Yong’an;QIAN Qijie;WU Jing(Jiangsu Cancer Hospital,Jiangsu Institute of Cancer Research,The Affiliated Cancer Hospital of Nanjing Medical University,Nanjing 210009,Jiangsu,China;Nanjing University of Posts and Telecommunications,Nanjing 210023,Jiangsu,China;Jiangsu Nursing Society,Nanjing 210008,Jiangsu,China)

机构地区：[1]江苏省肿瘤医院,江苏省肿瘤防治研究所,南京医科大学附属肿瘤医院,江苏南京210009 [2]南京邮电大学,江苏南京210023 [3]江苏省护理学会,江苏南京210008

出　　处：《中华中医药学刊》2025年第3期19-23,I0002,共6页Chinese Archives of Traditional Chinese Medicine

基　　金：国家自然科学基金青年科学基金项目(82203171);江苏省中医药科技发展计划项目(MS2021097);江苏省卫生健康委员会干部保健课题项目(BJ23023);江苏省卫生健康委科研项目(M2021114)。

摘　　要：目的建立基于人工智能的癌痛中医辨证模型,为癌痛中医智能辨证工作的开展提供依据。方法检索中国知网、维普中文科技期刊数据库、万方数据库、中国生物医学文献数据库、古今医案云平台建库至2023年9月收录的癌痛中医病案数据,采用经过超参数调优的线性支持向量分类(Linear support vector classification,LSVC)、最近质心(Nearest Centroid,NC)、多项式朴素贝叶斯(Multinomial Naive Bayes,MNB)、随机森林(Random Forest,RF)、随机梯度下降(Stochastic Gradient Descent,SGD)、多层感知机(Multilayer Perceptron,MLP)等机器学习算法对数据集进行量化分析,建立癌痛的诊断辨证模型。运用五折交叉验证对模型进行评价,评价指标包括Accuracy、Precision、Recall、F1 score及AUC值。结果癌痛中医四诊信息为输入变量共72项,输出变量为癌痛中医证型共6项。6种模型的拟合效果较好,Accuracy、Precision、Recall、F1 score均在0.89以上,AUC值达0.94以上;其中NC模型的准确率、查准率与查全率最高,Accuracy、Precision、Recall、F1score分别为0.978、0.980、0.978、0.977,AUC值达0.986,均高于其他算法模型,其参数的中医解释基本符合中医诊断原理。结论基于NC算法模型建立的癌痛中医辨证模型具有较高的诊断及预测能力,可为临床医务工作者开展癌痛中医智能辨证提供借鉴。Objective To establish the TCM syndrome differentiation model of cancer pain based on artificial intelligence,so as to provide basis for the development of TCM syndrome differentiation of cancer pain.Methods The TCM medical record data of cancer pain was collected from CNKI,VIP Chinese Science and Technology periodical Database,Wanfang database,China Biology Medicine disc,and Ancient and Modern Medical Record Cloud Platform from the embellishment time up to September 2023.Ma-chine learning algorithms,such as linear support vector classification(LSVC),nearest centroid(NC),multinomial Na?ve Bayes(MNB),random forest(RF),stochastic gradient descent(SGD)and multilayer perceptron(MLP)were used to quantitatively an-alyze the data set,and the syndrome differentiation model of cancer pain was established.The model was evaluated by 5-fold cross validation.The evaluation indexes included Accuracy,Precision,Recall,F1 score and AUC value.Results There were 72 input variables and 6 output variables for TCM syndromes of cancer pain.The Accuracy,Precision,Recall and F1 score of the six models were all above 0.89,and the AUC value was above 0.94.Among them,the Accuracy,Precision and Recall of the nearest centroid model were the highest.The Accuracy,precision,recall and F1 score were 0.978,0.980,0.978 and 0.977,respective-ly,and the AUC value was 0.986,which was higher than other algorithm models.The TCM interpretation of the parameters basi-cally accorded with the principle of TCM diagnosis.Conclusion The TCM syndrome differentiation model of cancer pain based on the nearest centroid algorithm model has high diagnosis and prediction ability,which can provide reference for clinical workers to carry out TCM intelligent syndrome differentiation of cancer pain.

关 键 词：人工智能 机器学习 癌痛 中医辨证模型 医案 辨证 

分 类 号：R222.19[医药卫生—中医基础理论]