基于多标记深度森林的膝骨关节炎智能辅助诊断方法  

Intelligent assisted diagnosis method of knee osteoarthritis based on multi-label deep forest

作　　者：龙锦益 杨宇 张子龙 叶倩云 吴汉瑞 张荣华[2,5,6] 张佳 LONG Jinyi;YANG Yu;ZHANG Zilong;YE Qianyun;WU Hanrui;ZHANG Ronghua;ZHANG Jia(College of Information Science and Technology,Jinan University,Guangzhou 510632,China;Guangdong Key Laboratory of Traditional Chinese Medicine Information Technology,Jinan University,Guangzhou 510632,China;Pazhou Laboratory,Guangzhou 510335,China;College of Traditional Chinese Medicine,Jinan University,Guangzhou 510632,China;College of Pharmacy,Jinan University,Guangzhou 510632,China;Cancer Research Institution,Jinan University,Guangzhou 510632,China)

机构地区：[1]暨南大学信息科学技术学院,广州510632 [2]暨南大学广东省中医药信息技术重点实验室,广州510632 [3]广州琶洲实验室,广州510335 [4]暨南大学中医学院,广州510632 [5]暨南大学药学院,广州510632 [6]暨南大学癌症研究所,广州510632

出　　处：《中华中医药杂志》2024年第12期6811-6814,共4页China Journal of Traditional Chinese Medicine and Pharmacy

基　　金：国家自然科学基金项目(No.62106084,No.61773179);广东省自然科学基金项目(No.2022A1515010468,No.2019A1515012175);广州市科技计划项目(No.202201010498);广东省中医药信息化重点实验室(No.2021B1212040007)。

摘　　要：目的:利用多标记深度森林(MLDF)算法构建膝骨关节炎(KOA)智能辅助诊断模型,并探索多标记方法在中医数据集上的优势。方法:基于1421例临床样本,使用MLDF算法构建分类模型,在6个评价指标上与其他5种多标记算法进行对比;使用多标记算法Rank-SVM、ML-kNN和单标记算法SVM、kNN建模并对比。结果:使用MLDF构建的分类模型在6个评价指标上均优于其他5种对比算法,并且在KOA标记上的AUC为0.8122,远高于其他对比算法;Rank-SVM与ML-kNN分类准确率(0.7746、0.7787)高于其对应的单标记算法(0.7641、0.7570),且在大多数评价指标上均优于其对应的单标记算法。结论:在多证兼夹的中医数据集上,多标记分类算法性能优于其对应的单标记算法,MLDF算法在KOA的诊断结果上与真实诊断结果的一致性较好,具有较好的推广和应用前景。Objective:To construct an intelligent assisted diagnosis model for knee osteoarthritis(KOA)using a multilabel learning with deep forest(MLDF)algorithm and explore the advantages of the multi-label method in the traditional Chinese medicine(TCM)data set.Methods:Based on 1421 clinical samples,a classification model was constructed using a MLDF algorithm and compared with five other multi-marker algorithms on six evaluation indexes.Multi-label algorithms Rank-SVM and ML-kNN and single-label algorithms SVM and kNN were used for modeling and comparison.Results:The classification model built with the MLDF algorithm outperformed the comparison algorithms on all six evaluation metrics.The AUC for KOA was 0.8122,significantly higher than other comparison algorithms.Rank-SVM and ML-kNN showed higher classification accuracies(0.7446,0.7787)compared to their corresponding single-label algorithm(0.7641,0.7570),and they also outperformed these single-label methods in most evalution metrics.Conclusion:On the TCM data set with multiple syndromes and clips,the performance of the multi-label classification algorithm is better than that of its corresponding single-label algorithm.The diagnosis results of the MLDF algorithm in KOA are in good agreement with the confirmed diagnosis,and it has good promotion and application prospects.

关 键 词：膝骨关节炎 人工智能 机器学习 多标记学习 多标记深度森林 智能辅助诊断模型 

分 类 号：TP3[自动化与计算机技术—计算机科学与技术]