基于多序列MRI影像组学预测肺腺癌EGFR基因表型  

Predictive value for EGFR gene phenotype of lung adenocarcinoma based on multi-sequences MRI radiomics

作　　者：唐兴 白国艳 王虹 印弘 张艰 徐肖攀 康晓伟 TANG Xing;BAI Guo-yan;WANG Hong(Department of Radiology,Xijing Hospital,the Air Force Medical University,Xi′an 710032,China)

机构地区：[1]空军军医大学第一附属医院放射科,西安710032 [2]陕西省人民医院检验科,西安710068 [3]空军军医大学第一附属医院呼吸与危重症医学科,西安710032 [4]空军军医大学军事生物医学工程学系,西安710038 [5]西安市人民医院(西安市第四医院)医学影像中心,西安710199

出　　处：《放射学实践》2021年第8期1010-1015,共6页Radiologic Practice

基　　金：国家自然科学基金青年项目(81901698);陕西省重点研发计划项目(2017ZDXM-SF-044);西京医院助推计划(XJZT5ZL04);西安市人民医院(西安市第四医院)科研孵化基金(CX-17)。

摘　　要：目的:探讨基于多序列MRI影像组学在预测肺腺癌EGFR基因表型中的应用价值。方法:回顾性分析2015年1月-2018年12月行肺部MRI检查及EGFR基因检测的74例肺腺癌患者的临床、病理和影像资料。对肿瘤标本进行基因检测,证实EGFR突变型32例,野生型42例。MRI序列包括T_(2)WI、DWI及ADC图。临床资料包括性别、年龄、吸烟史、CEA、Ki-67、位置、最大直径和病理分级。分别在T_(2)WI、DWI和ADC图上于肿瘤最大截面手动勾画感兴趣区,共提取1404个影像组学特征。然后,利用Student-t检验和基于非线性支持向量机的递归特征消除(SVM-RFE)策略进行特征优选后建立预测模型,并应用受试者工作特征曲线(ROC)评估模型的预测效能。结果:最终选取16个最优纹理特征构建EGFR表型预测模型,其预测EGFR突变型的敏感度为53.1%,特异度为92.9%,符合率为75.7%,曲线下面积(AUC)为0.826。在此基础上进一步联合性别因素构建模型,预测符合率提高到78.9%。结论:基于多序列MRI影像组学方法可在一定程度上预测肺腺癌的EGFR基因表型,为术前肺腺癌患者的个体化风险分层提供参考。Objective:To explore the feasibility of prediction of epithelial growth factor receptor(EGFR)gene mutation for lung adenocarcinoma based on multi-sequence MRI radiomics.Methods:A retrospective study was conducted on 74 cases with pulmonary adenocarcinoma(EGFR-mutant 32 cases and wild-type 42 cases diagnosed by a genetic test for the tumor)confirmed by postoperative pathology.All patients underwent 1.5T chest MRI examination before surgery.The clinical index including gender,age,tobacco smoking history,lesion location,maximum diameter,CEA and Ki-67 level,and histopathological grade were recorded.On the T_(2)WI,DWI and ADC images of each patient,ROIs was drawn in the largest tumor region respectively,and totally 1404 radiographic features were extracted.Then,the Student t-test and support vector machine recursive feature elimination(SVM-RFE)method were used to select out the optimal characteristics and establish the prediction model.ROC was used to analyze the performance of the predictive models.Results:In the performance evaluation of EGFR mutation in lung adenocarcinoma,16 optimal characteristics were selected out for establishing the prediction model,and its sensitivity,specificity,accuracy and AUC for predicting EGFR-mutant were 53.1%,92.9%,75.7% and 0.826,respectively.When the prediction model was further added with gender factors,and the accuracy reached 78.9%.Conclusion:Radiomics based on multi-sequence MRI can predict EGFR gene phenotypes of lung adenocarcinoma with relative high accuracy,thus can be helpful for preoperative individualized risk stratification of lung adenocarcinoma patients.

关 键 词：肺肿瘤 腺癌 影像组学 磁共振成像 扩散加权成像 表观扩散系数 血管内皮细胞生长因子 

分 类 号：R734.2[医药卫生—肿瘤] R445.2[医药卫生—临床医学]