改进深度极限学习机在肺癌中医证型分类中的应用研究  

Application of Improved Deep Extreme Learning Machine in the Classification of Traditional Chinese Medicine Syndromes of Lung Cancer

作　　者：章新友[1] 徐华康 周小玲[1] 刘梦玲 李秀云 张亚明 张春强[1] 唐琍萍 Zhang Xinyou;Xu Huakang;Zhou Xiaoling;Liu Mengling;Li Xiuyun;Zhang Yaming;Zhang Chunqiang;Tang Liping(Computer College of Jiangxi University of Chinese Medicine,Nanchang 330004,China)

机构地区：[1]江西中医药大学计算机学院,南昌330004

出　　处：《世界科学技术-中医药现代化》2023年第6期2132-2139,共8页Modernization of Traditional Chinese Medicine and Materia Medica-World Science and Technology

基　　金：国家自然科学基金委员会地区科学基金项目(81660727):基于策略模式的中药性效数据挖掘方法研究,负责人:章新友;江西省教育厅科学技术研究重点项目(GJJ190635):基于网络药理学的抗癌中药“四气五味”物质基础研究,负责人:章新友。

摘　　要：目的利用特征选择及Likert分级法量化肺癌病历数据,构建基于麻雀搜索算法优化的深度极限学习机模型,对肺癌中医病历数据进行证型分类与预测,为中医证型分类研究提供科学有效手段。方法从江西中医药大学附属医院收集了2015年1月-2021年12月诊断为肺癌的497例病历,筛选412例病历作为研究对象。利用特征选择和特征重要性排序等方法归纳出不同证型的证型因子,并使用Likert分级法量化证型因子。构建基于麻雀搜索算法优化的深度极限学习机,对模型进行训练、测试。最后把本研究所建模型与其他机器学习模型按照3种评价标准进行比较。结果本研究建立的SSA-DELM模型的平均分类准确率为88.44%,而采用支持向量机和贝叶斯网络的平均准确率分别为83.39%和84.53%。SSADELM模型在5种证型上的召回率及F1值大部分在80%以上,也优于其他传统的机器学习模型。结论研究结果表明,利用特征选择结合Likert分级法量化后的肺癌病历数据,相比于0-1化处理的数据更能显现出数据的特征,提高了分类模型的准确率,SSA-DELM新模型相比其他传统的机器学习分类模型,有更好的表征学习能力及学习速度。该模型不仅为临床治疗肺癌的研究提供了科学的技术手段,也为中医辨证论治的信息化、智能化发展提供有益的借鉴。Objective To use feature selection and Likert grading method to quantify the data of lung cancer medical records,to construct a deep extreme learning machine model optimized by the sparrow search algorithm,to classify and predict the syndrome types of traditional Chinese medicine medical record data of lung cancer,and to provide scientific and effective research on syndrome type classification of traditional Chinese medicine.means.Methods The medical records of 497 cases diagnosed with lung cancer from January 2015 to December 2021 were collected from the Affiliated Hospital of Jiangxi University of Traditional Chinese Medicine,and 412 medical records were screened as the research objects.Syndromic factors of different syndromes were summarized by feature selection and feature importance ranking,and the syndrome factors were quantified by Likert grading method.Build a deep extreme learning machine optimized based on the sparrow search algorithm,and train and test the model.Finally,the model built in this paper is compared with other machine learning models according to three evaluation criteria.Results The average classification accuracy of the SSA-DELM model established in this paper was 88.44%,while the average accuracy of the support vector machine and Bayesian network was 83.39%and 84.53%,respectively.The recall rate and F1 value of the SSA-DELM model on the five syndrome types are mostly above 80%,which is also better than other traditional machine learning models.Conclusion The results of the study show that the use of feature selection combined with Likert grading method to quantify the lung cancer medical record data,compared with the 0-1 processing data,can show the characteristics of the data,improve the accuracy of the classification model,SSA-DELM new Compared with other traditional machine learning classification models,the model has better representation learning ability and learning speed.This model not only provides a scientific and technical means for the clinical treatment of lung cancer,but als

关 键 词：肺癌 证型分类 深度极限学习机 特征选择 

分 类 号：R-058[医药卫生]