利用Logistic回归和神经网络分析乳腺癌的预后因素  

Study on Prognostic Factors of Breast Cancer with Logistic Regression and Artificial Neural Network

作　　者：章鸣嬛[1] 陈瑛 郭欣[1] 张璇 季萌 ZHANG Minghuan;CHEN Ying;GUO Xin;ZHANG Xuan;JI Meng(Lab of Big Data Analyses and Process,Information Science and Technology,Shanghai Sanda University,Shanghai 201209)

机构地区：[1]上海杉达学院大数据分析与处理研究中心,上海201209

出　　处：《计算机与数字工程》2020年第3期617-622,共6页Computer & Digital Engineering

基　　金：2016年上海市民办高校重点科研项目(编号:2016-SHNGE-01ZD);2015年IBM大学合作部联合研究项目(编号:D-2111-15-001)资助。

摘　　要：研究以SEER数据库中1990~2014年间的乳腺癌数据为研究对象,分别利用Logistic回归和神经网络两种机器学习算法进行建模,以寻找影响乳腺癌5年预后的因素。研究表明:1)肿瘤分期、肿瘤分级、肿瘤尺寸、雌激素水平、年龄分组和孕激素水平等因素对于乳腺肿瘤预后具有较大影响,与临床诊断经验相吻合。2)在此两种模型下,模型测试集上的灵敏度和特异度均介于75.4%~78.2%之间,模型的ROC曲线面积(AUC)均处于0.847~0.850之间。因此,Logistic回归和神经网络算法可有效探寻模型输入变量间的关系,构建乳腺癌患者的优化预后模型,辅助医生判断患者预后情况及治疗效果。On the basis of the breast cancer data from 1990 to 2014 in the SEER database,this study developes,models with the Logistic regression and the artificial neural network,two kinds of machine learning classification algorithms,with the aim to exploring the factors affecting the 5-year prognosis of breast cancer. The results show that:1)such factors as tumor stage,tumor grade,tumor size,estrogen level,progesterone level,age grouping have a greater impact on the prognosis of breast tumors,which is in consistence with the clinic practice. 2)from the results of two models,both the sensitivity and specificity of the model test set are between 75.4% and 78.2%,and the ROC curve areas(AUC)of the two models are between 0.847 and 0.850. Therefore,it could be certain to conclude that Logistic regression and decision tree algorithms could be used to effectively explore the relationship between model input variables,develop a prognostic model for breast cancer patients,and assist doctors to some extent to assess the prognosis and the treatment effect.

关 键 词：SEER 乳腺癌 LOGISTIC回归 神经网络 预后因素 

分 类 号：Q334[生物学—遗传学]