基于SFS-SVM的乳腺癌预测模型的构建  

Construction of breast cancer prediction model based on SFS-SVM

作　　者：赖胜圣[1] 刘虔铖[1] 余丽玲[1] 刘文平 杨蕊梦 金浩宇[1] LAI Shengsheng;LIU Qiancheng;YU Liling;LIUWenping;YANG Ruimeng;JIN Haoyu(School of Medical Devices,Guangdong Food and Drug Vocational College,Guangzhou 510520,China;Department of Radiology,the Second Affiliated Hospital of South China University of Technology,Guangzhou First People's Hospital,Guangzhou 510180,China)

机构地区：[1]广东食品药品职业学院医疗器械学院,广东广州510520 [2]广州市第一人民医院/华南理工大学附属第二医院放射科,广东广州510180

出　　处：《中国医学物理学杂志》2019年第7期826-829,共4页Chinese Journal of Medical Physics

基　　金：广东省高等学校珠江学者岗位计划自主项目(2016);广东省自然科学基金(2018A030313282);广州市科技计划项目(201607010038);广东省医学科学技术研究基金(A2018338,A2019465);广东食品药品职业学院校级科研项目(2015YZ0012,2015YZ0020)

摘　　要：目的:构建基于序列前向选择算法(SFS)与支持向量机算法(SVM)分类器融合的乳腺癌预测模型,提高计算机辅助诊断技术对乳腺癌细针穿刺细胞病理的准确率。方法:对456组乳腺肿瘤病理数据作为训练集,利用SFS-SVM算法对30个特征进行筛选,得到最优的特征组合,再用112组乳腺肿瘤病理数据作为测试集验证,构建乳腺癌预测模型。该模型的预测精度通过5折交叉验证进行评价。评价指标包括:受试者工作特性曲线(ROC)下面积(AUC)、准确率(ACC)、敏感度和特异度。结果:构建了基于SFS-SVM的乳腺癌预测模型,该模型(AUC为98.39%,ACC为97.35%)相对于单独SVM算法(AUC为97.00%,ACC为92.42%)有一定的提高。结论:基于SFS特征选择的SVM分类器乳腺癌预测模型能较好地对乳腺癌进行辅助诊断。Objective To improve the accuracy of computer-aided diagnosis for fine needle aspiration pathology in breast cancer by employing the breast cancer prediction model based on sequential forward feature selection (SFS) algorithm and support vector machine (SVM) classifier.Methods The pathological data of 456 breast tumors were used as training set.A total of 30 features were screened by SFS-SVM algorithm to obtain the optimal feature combination,and then the pathological data of 112 breast tumors were used as test set to construct breast cancer prediction model.The prediction accuracy of the constructed model was evaluated with 5-fold cross-validation method.The evaluation indicators included area under the receiver operating characteristic curve (AUC),accuracy,sensitivity,and specificity.Results Compared with SVM-based model which had an AUC of 97.00% and an accuracy of 92.42%,the breast cancer prediction model based on SFS-SVM had better performances,achieving an AUC of 98.39% and an accuracy of 97.35%.Conclusion The breast cancer prediction model based on SFS-SVM exhibits a good predictive efficacy on the auxiliary diagnosis of breast cancers.

关 键 词：乳腺癌 预测模型 序列前向选择算法 支持向量机算法 

分 类 号：R318[医药卫生—生物医学工程]