基于机器学习方法的原发性肝癌患者预后预测研究  

Prognosis of Primary Liver Cancer Patients Based on Machine Learning Method

作　　者：李琳[1] 杨日东 王哲 张学良[1] 周毅[2] LI Lin;YANG Ri-dong;WANG Zhe(Zhongshan School of Medicine,Sun Yat-sen University,Guangzhou 510030,Guangdong Province,P.R.C.)

机构地区：[1]新疆医科大学 [2]中山大学中山医学院

出　　处：《中国数字医学》2019年第3期34-37,共4页China Digital Medicine

基　　金：国家重点研发计划项目(编号:2018YFC0116902;2018YFC0116904;2016YFC0901602);国家自然科学基金项目(编号:61876194;11661007);NSFC-广东大数据科学中心联合基金项目(编号:U1611261);广州市科技计划项目(编号:201604020016)~~

摘　　要：目的:使用机器学习方法对原发性肝癌患者生存和复发时间进行预测,并比较不同机器学习方法的预测效果。方法:使用Logistic回归、随机森林、SVM、C5.0决策树、神经网络、Bagging算法和Adaboost算法构建患者肝切除手术后3年无瘤生存时间和3年总生存时间的预后预测模型。结果:通过Logistic回归模型找到了影响患者生存和复发的影响因素,并且支持向量机和随机森林有较好的效果,测试集上的AUC为0.7122和0.7103。结论:机器学习方法相较于Logistic回归模型有较好的预测效果,但是Logistic回归模型有较好的可解释性。在分析实际问题时,可以采用二者结合的方法。Objective:The survival and recurrence time of patients with primary hepatocellular carcinoma were predicted by machine learning method,and the predictions of different machine learning methods were compared.Methods:Logistic regression,random forest,SVM and C5.0 decision tree,neural network,Bagging algorithm and Adaboost algorithm were used to construct the prognostic prediction model of 3-year disease-free survival and 3-year total survival time after keratectomy.Results:Logistic regression model was used to find the factors affecting patient survival and recurrence,and support vector machines and random forest were effective,and the AUC on the test set was 0.7122 and 0.7103.Conclusion:Compared with the Logistic regression model,the machine learning method has better prediction effect,but the Logistic regression model has better explanatory ability.In the analysis of practical problems,the combination of the two methods can be adopted.

关 键 词：机器学习 原发性肝癌 LOGISTIC回归 支持向量机 预后预测 

分 类 号：R319[医药卫生—基础医学]