决策树及支持向量机与深度学习模型在肝癌鉴别诊断中的比较研究  

Comparative Study of Decision Tree,Support Vector Machine and Deep Learning Model in Differential Diagnosis of Liver Cancer

作　　者：黄辛迪 黄慧 刘佳俊 丁长松 HUANG Xin-di;HUANG Hui;LIU Jia-jun;DING Chang-song(School of Information Science and Engineering,Hunan University of Chinese Medicine,Changsha 410208,Hunan,China;Big Data Analysis Laboratory of Traditional Chinese Medicine in Hunan Province,Changsha 410208,Hunan,China)

机构地区：[1]湖南中医药大学信息科学与工程学院,湖南长沙410208 [2]湖南省中医药大数据分析实验室,湖南长沙410208

出　　处：《医学信息》2023年第15期70-74,共5页Journal of Medical Information

基　　金：湖南省中医药科研计划重点课题(编号:2020002);长沙市自然科学基金项目(编号:kq2202265);湖南中医药大学校级科研项目(编号:2019XJJJ029)。

摘　　要：目的使用数据挖掘技术研究肝功能检查数据,分析肝功能检查指标与肝癌诊断的关联,探究肝癌早诊断、早治疗的辅助数据分析方法。方法构建决策树C4.5模型并提取决策方法,并以Bagging方法优化;采用网格划分法和粒子群优化算法优化支持向量机模型;构建多层感知机(MLP)和卷积神经网络(CNN)进行性能比较。基于决策树和SVM模型进行特征属性分析和最优特征子集选择。结果Bagging决策树模型、SVM、MLP模型的10交叉检验准确率分别为95.18%、95.60%、90.17%,测试准确率分别为94.34%、93.40%、89.78%。在肝功能检查指标中,碱性磷酸酶、谷丙转氨酶、天门冬氨酸转氨酶、年龄、直接胆红素是主要贡献指标,三指标联合诊断对肝癌预测率达86.08%。结论决策树、支持向量机、多层感知机建立的肝癌分类器模型都可用于肝癌辅助诊断,SVM模型略优,预测模型对肝癌早期鉴别有较好的辅助作用。Objective To study liver function test data with data mining technology,analyze the correlation between liver function test indicators and liver cancer diagnosis,and explore methods for early diagnosis and early treatment of liver cancer.Methods The decision tree C4.5 model was constructed and the decision method was extracted and optimized by Bagging method.The grid division method and particle swarm optimization algorithm were used to optimize the support vector machine model.Multi-layer perceptron(MLP)and convolutional neural network(CNN)were constructed for performance comparison.Based on decision tree and SVM model,feature attribute analysis and optimal feature subset selection were carried out.Results The 10-cross-check accuracy rates of Bagging decision tree model,SVM,and MLP model were 95.18%,95.60%,and 90.17%,respectively,and the test accuracy were 94.34%,93.40%,and 89.78%,respectively.Among the liver function test indicators,alkaline phosphatase,alanine aminotransferase,aspartate aminotransferase,age and direct bilirubin were important indexes in the diagnosis of liver cancer with three-index combined diagnosis up to 86.08%in accuracy.Conclusion The liver cancer classifier models established by decision tree,support vector machine and multilayer perceptron can all be used in diagnosis of liver cancer,SVM model slightly better.The prediction models are supplementary measures for early identification of liver cancer.

关 键 词：肝癌 决策树 支持向量机 深度学习 多层感知机 卷积神经网络 

分 类 号：R735.7[医药卫生—肿瘤]