基于CT影像组学列线图预测肺腺癌EGFR突变的研究  

CT-Based Radiomics Nomogram to Predict EGFR Mutation in Lung Adenocarcinoma

作　　者：李淑华[1,2] 杨昭[3] 王小雷 张群芳 张雪丽 李阳 李伟 王效静 刘浩 谢宗玉 LI Shuhua;YANG Zhao;WANG Xiaolei(Department of Radiology,The First Affiliated Hospital of Bengbu Medical College,Bengbu,Anhui Province 233004,P.R.China)

机构地区：[1]蚌埠医学院第一附属医院放射科,233004 [2]蚌埠医学院医学影像诊断学教研室,233000 [3]蚌埠医学院研究生院,233000 [4]安徽省呼吸系统疾病(肿瘤)临床医学研究中心,233004 [5]北京医准智能科技有限公司,100089

出　　处：《临床放射学杂志》2022年第9期1676-1682,共7页Journal of Clinical Radiology

基　　金：安徽省中央引导地方科技发展基金项目(编号:2020b07030008);安徽省重点研究与开发计划项目(编号:2022e07020033);蚌埠医学院自然科学重点项目(编号:2021byzd091)。

摘　　要：目的 探讨基于胸部平扫CT的影像组学列线图术前预测肺腺癌表皮生长因子受体(EGFR)基因突变状态的价值。方法 回顾性分析经病理证实的183例肺腺癌的临床及胸部平扫CT影像资料,包括EFGR突变型(110例)及EGFR野生型(73例)。评估每位患者的病灶CT特征。勾画整个病灶的三维容积感兴趣区(VOI)。将所有患者的重建肺窗图像及病灶的VOI上传至“医准-达尔文”智能科研平台,按照7∶3比例分成训练组(128例)及验证组(55例)。采用最大绝对值归一化、最小收缩、最优特征筛选、迭代筛选及选择算子(LASSO)算法对数据进行降维处理,计算影像组学分数(Rad-score),建立影像组学模型。运用单因素及多因素Logistic回归分析,筛选具有显著预测价值的临床-影像独立影响因素,建立临床模型。将具有显著预测价值的临床-影像独立影响因素及Rad-score通过Logistic回归,得到影像组学列线图模型。通过ROC曲线和校准曲线评估3个模型的诊断效能。结果 最终筛选出11个影像组学特征构建影像组学分数。多因素Logistic回归分析结果显示吸烟史、胸膜牵拉及影像组学评分是肺腺癌患者EGFR基因突变的独立危险因素。ROC曲线分析训练组及验证组中影像组学列线图模型曲线下面积最大,分别为0.816(95%CI:0.743~0.890)、0.830(95%CI:0.720~0.941),相较于临床模型与影像组学模型对肺腺癌EGFR基因突变的预测效能更好。Hosmer-Lemeshow检验显示训练组及验证组的影像组学列线图预测模型与真实结果没有显著差异(P>0.05)。结论 基于CT的影像组学评分联合临床-影像独立危险因素构成的影像组学列线图能够在术前有效预测肺腺癌EGFR基因突变状态。Objective To evaluate the value of non enhanced chest CT based radiomics nomogram in preoperative prediction of EGFR mutation in lung adenocarcinoma. Methods This retrospective study included 183 lung adenocarcinoma patients, including 110 patients with EGFR mutations and 73 patients with EGFR wild types and the corresponding clinical data and non enhanced chest CT images were analyzed.CT features of the lesions in each patient were evaluated.Three dimensional volume area of Interest(VOI) delineating the entire lesion.All patients’ reconstructed lung window images and VOI of lesions were uploaded to the “Med Darwin” intelligent scientific research platform, and divided into the training group(128 cases) and the validation group(55 cases) in a 7∶3 ratio.The maximum absolute value normalization, minimum shrinkage, optimal feature screening, iterative screening and selection operator(LASSO) algorithm were used to reduce the dimension of the data, and rad score was calculated to establish the radiomics model.Univariate and multivariate logistic regression analysis was used to screen the clinically and image independent factors with significant predictive value and establish the clinical model.The clinical imaging independent influencing factors and Rad score with significant predictive value were used by Logistic regression to obtain the radiomics nomogram model.The diagnostic performance of the three models was evaluated by ROC curve and calibration curve. Results 11 radiomics features were finally selected to construct the Rad score.Multivariate Logistic regression analysis showed that smoking history, pleural traction and rad score were independent risk factors for EGFR gene mutation in lung adenocarcinoma.The AUC of the radiomics nomogram model in the ROC analysis training group and the validation group were 0.816(95% CI:0.743-0.890) and 0.830(95% CI:0.720-0.941),respectively.Compared with the clinical model and radiomics model, the radiomics nomogram model was more effective in predicting EGFR gene mutat

关 键 词：肺腺癌 表皮生长因子受体 体层摄影术 X射线计算机 影像组学 

分 类 号：R730.44[医药卫生—肿瘤] R734.2[医药卫生—临床医学]