CT影像组学联合CT特征预测肺亚实性结节侵袭性  

CT radiomics combined with CT features for predicting invasiveness of subsolid pulmonary nodules

作　　者：吴雯丽 宋庆平 董连军 杨帅 于秋凤 朱正 赵燕风 WU Wenli;SONG Qingping;DONG Lianjun;YANG Shuai;YU Qiufeng;ZHU Zheng;ZHAO Yanfeng(Medical Imaging Center,Liaocheng Tumor Hospital,Liaocheng 252000,China;Department of Chest Surgery,Liaocheng Tumor Hospital,Liaocheng 252000,China;Department of Diagnostic Radiology,National Cancer Center/National Clinical Research Center for Cancer/Cancer Hospital,Chinese Academy of Medical Sciences and Peking Union Medical College,Beijing 100021,China)

机构地区：[1]聊城市肿瘤医院影像中心,山东聊城252000 [2]聊城市肿瘤医院胸外科,山东聊城252000 [3]国家癌症中心/国家肿瘤临床医学研究中心/中国医学科学院北京协和医学院肿瘤医院影像诊断科,北京100021

出　　处：《中国介入影像与治疗学》2023年第10期615-620,共6页Chinese Journal of Interventional Imaging and Therapy

摘　　要：目的 观察CT影像组学联合CT特征预测肺亚实性结节侵袭性的价值。方法 回顾性分析170例肺亚实性结节患者资料,包括6例非典型腺瘤样增生(AAH)、12例原位腺癌(AIS)、58例微浸润性腺癌(MIA)及94例浸润性腺癌(IAC),将AAH、AIS和MIA归为非侵袭组、IAC归为侵袭组。按7∶3比例将患者分为训练集(n=119,含5例AAH、9例AIS、36例MIA及69例IAC)和验证集(n=51,含1例AAH、3例AIS、22例MIA及25例IAC)。采用单因素及logistic回归分析训练集患者一般资料及病灶CT表现,筛选预测肺亚实性结节侵袭性的独立危险因素并建立CT模型;基于训练集提取及筛选病灶最佳影像组学特征,以之构建影像组学模型。基于CT模型及影像组学模型构建联合模型,并以列线图将其可视化。绘制受试者工作特征(ROC)曲线,计算曲线下面积(AUC),基于验证集评估各模型诊断效能;以校准曲线评价联合模型的校准程度。结果 CT所示结节长径和最大CT值为预测肺亚实性结节为IAC的CT相关独立危险因素,以之构建CT模型。基于训练集筛选出6个最佳影像组学特征并构建影像组学模型。CT模型、影像组学模型及联合模型预测验证集肺亚实性结节侵袭性的AUC分别为0.772、0.785及0.869;联合模型的AUC高于CT模型(Z=2.336,P=0.019)而与影像组学模型差异无统计学意义(Z=1.925,P=0.054),其预测结果与实际结果的一致性较高。结论 CT影像组学联合CT特征可有效预测肺亚实性结节侵袭性。Objective To observe the value of CT radiomics combined with CT features for predicting invasiveness of subsolid pulmonary nodules.Methods Data of 170 patients with subsolid pulmonary nodules,including 6 atypical adenomatous hyperplasia(AAH),12 adenocarcinoma in situ(AIS),58 minimally invasive adenocarcinoma(MIA)and 94 invasive adenocarcinoma cancer(IAC)confirmed by surgical pathology were retrospectively analyzed.AAH,AIS and MIA were enrolled as non-invasive group and IAC as invasive group.Then patients were divided into training set(n=119,including 5 AAH,9 AIS,36 MIA and 69 IAC)and test set(n=51,including 1 AAH,3 AIS,22 MIA and 25 IAC)at the ratio of 7∶3.The general information of patients and CT manifestations of lesions in training set were analyzed with univariate and multivariate logistic regression to screen the independent risk factors for predicting invasiveness of subsolid pulmonary nodules and establishing a CT model.Based on data of training set,the optimal radiomics features were extracted and screened to construct a radiomics model,and finally a combined model was constructed based on CT model and radiomics model,which was visualized by drawing a nomogram.Receiver operating characteristic(ROC)curves were drawn,and the area under the curves(AUC)were calculated to evaluate the diagnostic efficacy of each model in test set,and the calibration degree of combined model was evaluated using calibration curve.Results The length and the maximum CT value of nodules shown on CT were independent risk factors for predicting pulmonary subsolid nodule as IAC,which were used to construct a CT model.Six optimal radiomics features were screened based on the training set,and a radiomics model was constructed.In test set,the AUC of CT model,radiomics model and combined model for predicting invasiveness of subsolid pulmonary nodule was 0.772,0.785 and 0.869,respectively,of combined model was higher than that of CT model(Z=2.336,P=0.019)but not statistical different with that of radiomics model(Z=1.925,P=0.054).There w

关 键 词：肺肿瘤 肿瘤侵袭性 体层摄影术 X线计算机 影像组学 

分 类 号：R734[医药卫生—肿瘤]