文献类型：期刊文献

中文题名：基于机器学习的腰椎间盘突出症中医证候分类模型构建及验证

英文题名：Development and Validation of Traditional Chinese Medicine Syndrome Classification Models for Lumbar Disc Herniation Based on Machine Learning

作者：王志鹏[1];张晓刚[1];张宏伟[1];赵希云[1];李元贞[1];秦大平[2];任真[3];郭成龙[1]

第一作者：王志鹏

机构：[1]甘肃中医药大学附属医院骨科,兰州730020;[2]甘肃中医药大学中医临床学院,兰州730020;[3]甘肃中医药大学医学信息工程学院,兰州730020

第一机构：甘肃中医药大学第二附属医院

年份：2025

卷号：20

期号：18

起止页码：3337

中文期刊名：世界中医药

外文期刊名：World Chinese Medicine

收录：;北大核心:【北大核心2023】；

基金：国家自然科学基金项目(82505367);张晓刚全国名老中医药专家传承工作室建设项目(国中医药人教发[2022]75号);甘肃省自然科学基金项目(24JRRA1037);甘肃省青年人才个人项目(2025QNGR72);兰州市科学技术局青年科技计划项目(2023-2-47,2023-2-48)。

语种：中文

中文关键词：腰椎间盘突出症;中医证候;预测模型;机器学习;人工智能算法

外文关键词：Lumbar disc herniation;TCM syndromes;Prediction model;Machine learning;Artificial intelligence algorithm

摘要：目的:基于机器学习算法建立并验证腰椎间盘突出症(LDH)常见中医证候分类预测模型。方法:通过流行病学调查方法,收集甘肃中医药大学附属医院LDH患者资料606例。根据中医证候采集量表收集患者中医四诊信息,将数据预处理和降维,随机按照7∶3划分为训练集(424例)和测试集(182例);采用支持向量机(SVM)、决策树(DT)、朴素贝叶斯(NB)、随机森林(RF)、极端梯度提升机(XGBoost)、人工神经网络(ANN)6种算法进行建模;运用10-折交叉验证进行调参优化模型,以准确率、灵敏度、特异度和受试者工作特征(ROC)曲线下面积(AUC)进行模型性能评价。结果:采用主成分分析对中医症状数据降维,提取8个公共因子。基于6种算法构建的LDH中医证候分类模型中,RF、SVM、DT、XGBoost、NB、BP的准确率分别为82.42%、91.21%、87.91%、90.11%、92.86%、88.46%;AUC分别为0.942 2、0.984 8、0.947 9、0.895 5、0.909 8、0.966 4。SVM和NB算法所构建模型的准确率和AUC均大于0.9。结论:该研究成功开发和验证LDH常见中医证候分类模型。SVM和NB算法构建的模型性能优于其余机器学习构建的模型,更适合实现对LDH中医证候的分类,可为中医证候规范化研究提供新的思路和方法。
Objective:To establish and validate machine learning-based prediction models for the classification of common traditional Chinese medicine(TCM)syndromes of lumbar disc herniation(LDH).Methods:Using an epidemiological survey,data from 606 patients with LDH treated at the Affiliated Hospital of Gansu University of Chinese Medicine were collected.TCM information from the four diagnostic methods was obtained according to a TCM syndrome data collection scale.After data preprocessing and dimensionality reduction,the dataset was randomly divided into a training set(424 cases)and a test set(182 cases)at a ratio of 7∶3.Six machine learning algorithms,including support vector machine(SVM),decision tree(DT),na ve Bayes(NB),random forest(RF),extreme gradient boosting(XGBoost),and artificial neural network(ANN),were used to construct classification models.Ten-fold cross-validation was applied for parameter optimization.Model performance was evaluated using accuracy,sensitivity,specificity,and the area under the receiver operating characteristic(ROC)curve(AUC).Results:Principal component analysis was applied to reduce the dimensionality of TCM symptom data,yielding eight common factors.Among the LDH TCM syndrome classification models constructed using the six algorithms,the accuracies of RF,SVM,DT,XGBoost,NB,and ANN were 82.42%,91.21%,87.91%,90.11%,92.86%,and 88.46%,respectively.The corresponding AUC values were 0.9422,0.9848,0.9479,0.8955,0.9098,and 0.9664,respectively.Models constructed using the SVM and NB algorithms achieved both accuracy and AUC values greater than 0.9.Conclusion:This study successfully developed and validated classification models for common TCM syndromes of LDH.Models constructed using the SVM and NB algorithms demonstrated superior performance compared with other machine learning models and are more suitable for LDH TCM syndrome classification,providing new ideas and methods for research on the standardization of TCM syndromes.