CT影像组学标签预测肺腺癌表皮生长因子受体基因敏感突变  

CT radiomics signatures for prediction of epidermal growth factor receptor sensitive mutation in lung adenocarcinoma

作　　者：肖磊 顾潜彪 张堃[1,3] 李平 沈宏荣[1] 朱璐[4] XIAO Lei;GU Qianbiao;ZHANG Kun;LI Ping;SHEN Hongrong;ZHU Lu(Department of Radiology,First Affiliated Hospital of Hunan University of ChineseMedicine,Changsha 410007,China;Department of Radiology,Hunan Provinceical People's Hospital,Changsha 410005,China;College of Integrated Traditional Chinese and WesternMedicine,Hunan University of Chinese Medicine,Changsha 410208,China;Department ofUltrasonography,Hunan Provinceical People's Hospital,Changsha 410005,China)

机构地区：[1]湖南中医药大学第一附属医院放射科,湖南长沙410007 [2]湖南省人民医院放射科,湖南长沙410005 [3]湖南中医药大学中西医结合学院,湖南长沙410208 [4]湖南省人民医院超声科,湖南长沙410005

出　　处：《中国介入影像与治疗学》2019年第4期220-224,共5页Chinese Journal of Interventional Imaging and Therapy

基　　金：国家自然科学基金青年科学基金项目(81603482);中国博士后科学基金面上项目(2017M622586);湖南省自然科学基金(2016JJ6115);湖南中医药大学重点学科建设项目

摘　　要：目的建立基于CT平扫图像的影像组学标签,探讨其用于预测肺腺癌表皮生长因子受体(EGFR)基因敏感突变的可行性。方法根据EGFR基因突变情况将80例肺腺癌患者分为EGFR敏感组(n=37)及EGFR不敏感组(n=43)。收集2组肺部病灶CT平扫主观影像征象,并提取影像组学特征;采用LASSO法进行特征挑选,以多因素Logistic回归分别建立主观影像征象模型、影像组学标签及融合模型,采用ROC曲线评估各模型预测EGFR基因敏感突变的效能。结果 EGFR敏感组与不敏感组间CT主观影像征象差异均无统计学意义(P均>0.05)。通过特征选择,确定CT影像组学标签由4个影像组学特征构成。主观影像征象模型(AUC=0.66)、影像组学标签(AUC=0.77)及融合模型(AUC=0.83)预测EGFR基因敏感突变的效能差异均有统计学意义(P均<0.05),以融合模型预测的效能最优。结论基于CT平扫图像建立的影像组学标签能够预测肺腺癌EGFR基因敏感突变。Objective To establish radiomics signatures based on non-enhanced CT image features,and to evaluate their feasibility for prediction epidermal growth factor receptor(EGFR)sensitive mutation of lung adenocarcinoma.Methods Eighty lung adenocarcinoma patients were divided into EGFR sensitive group(n=37)and EGFR insensitive(n=43)group according to EGFR mutation status.Radiomics features and subjective image features were collected from non-enhanced CT images.LASSO regression model was used to select radiomics features.Subjective image features model,radiomics model and combined diagnostic model were developed with multiple factors Logistic models,respectively.The predictive performance of EGFR sensitive mutation of each model was evaluated with ROC curve.Results There was no significant difference of subjective CT image features between EGFR sensitive and insensitive group(all P>0.05).Through feature selection,4 radiomics features were enrolled.Subjective CT image features model(AUC=0.66),radiomics model(AUC=0.77)and combined diagnostic model(AUC=0.83)had statistically significant differences in the performance of predicting EGFR sensitive mutation(all P<0.05).The combined diagnostic model had the best predictive efficiency.Conclusion Radiomics signatures based on non-enhanced CT images can be used to predict EGFR sensitive mutation in lung adenocarcinoma.

关 键 词：肺肿瘤 腺癌 受体 表皮生长因子 体层摄影术 X线计算机 影像组学 

分 类 号：R734.2[医药卫生—肿瘤] R814.42[医药卫生—临床医学]