肺纯磨玻璃结节瘤内及瘤周影像组学联合临床因素对病理分型预测价值探讨  

Value of intratumoral and peritumoral radiomic features combined with clinical factors predicting the pathological classification of pure ground-glass nodules

作　　者：张恒 刘吉兵 于永梅[2] 李正腾 王敏[2] ZHANG Heng;LIU Ji-bing;YU Yong-mei;LI Zheng-teng;WANG Min(Department of Intervention,Shandong Cancer Hospital and Institute,Shandong First Medical University and Shandong Academy of Medical Sciences,Jinan 250117,China;Department of Computed Tomography,Jining No.1 People’s Hospital,Jinan 272011,China)

机构地区：[1]山东省肿瘤防治研究院(山东省肿瘤医院)介入科山东省第一医科大学(山东省医学科学院),山东济南250117 [2]济宁市第一人民医院CT室,山东济宁272011

出　　处：《中华肿瘤防治杂志》2022年第7期508-515,522,共9页Chinese Journal of Cancer Prevention and Treatment

摘　　要：目的探讨瘤内及瘤周区域影像组学特征联合临床因素在预测肺纯磨玻璃结节(pGGN)病理分型中的价值。方法回顾性分析2016-09-05-2020-12-08在济宁市第一人民医院行术前胸部薄层CT平扫,并在2周内进手术切除的182例肺腺癌患者临床与影像资料,所有患者共200个pGGN。使用自动化软件勾画pGGN的感兴趣区域(ROI),使用膨胀算法沿结节边缘向外扩展5 mm。依次提取瘤内ROI和瘤周ROI影像组学特征,二者相结合生成融合特征。依据病理确诊结果,分为浸润性腺癌(IAC,100个)和非IAC(100个),将200个pGGN样本按照8∶2的比例随机划分为训练集(160个)和验证集(40个)。采用最小绝对收缩和选择算子(LASSO)进行特征降维,分别建立瘤内组学模型、瘤周组学模型、融合组学模型和融合组学联合临床(毛刺征、形状、血管改变、结节CT值、最大径)模型,通过绘制受试者工作特征曲线(ROC)分析评价预测模型的诊断效能,De-Long检验比较不同模型间的效能差异。结果融合组学模型的预测效能[训练集曲线下面积(AUC)为0.937,95%CI:0.888~0.970;验证集AUC为0.850,95%CI:0.702~0.943]高于瘤内组学模型[训练集AUC为0.921,95%CI:0.868~0.958;验证集AUC为0.845,95%CI:0.696~0.940]和瘤周组学模型(训练集AUC为0.891,95%CI:0.835~0.937;验证集AUC为0.843,95%CI:0.690~0.936)。融合组学特征联合临床因素的联合预测模型效能进一步提升(训练集AUC为0.958,95%CI:0.914~0.983;验证集AUC为0.895,95%CI:0.757~0.969)。联合预测模型较瘤内组学模型(Z=2.037,P=0.042)和瘤周组学模型(Z=3.084,P=0.002)在训练集中差异均有统计学意义。结论瘤周区域的组学特征在预测pGGN病理分型中具有重要价值,在融合瘤内组学特征并联合临床因素后诊断效能进一步提高。Objective To investigate the value of intratumoral and peritumoral radiomic features combined with clinical factors in predicting pathological classification of pure ground-glass nodules(pGGN).Methods Clinical data and radiological images of 200 pGGN in 182 patients with lung adenocarcinoma who received preoperative thin chest CT scan and underwent surgical resection within two weeks in Jining No.1 People’s Hospital from 2016-09-05 to 2020-12-08 were retrospectively analyzed.Region of interest(ROI)were delineated by automatic software,and the expansion algorithm was used to expand 5 mm along the edge of the nodule.Radiomic features were respectively extracted from the intratumoral ROI and peritumoral ROI.The radiomic features of intratumoral ROI and peritumoral ROI were combined to generate fusion radiomic features.According to the differences in final pathological results,they were divided into the invasive adenocarcinoma(IAC)group and the non-IAC group with 100 in each group.The data were randomly divided by a ratio of 8∶2 into a training set(160cases)and a validation set(40cases).the least absolute shrinkage and selection operator(LASSO)was used for dimensionality reduction of the features.Finally,the intratumoral radiomic model,peritumoral radiomic model,fusion radiomic model,and combination model combined fusion radiomic features and clinical features(including spiculation,nodular type,vessel changes,CT value of nodules,the maximal diameter of the nodule)were obtained.Receiver operating characteristic curve(ROC)analysis and the De-Long test were used to compare the diagnostic efficacy among models.Results The predictive efficiency of fusion radiomic model(training set AUC=0.937,95%CI:0.888-0.970;validation set AUC=0.850,95%CI:0.702-0.943)was higher than intratumoral radiomic model(training set AUC=0.921,95%CI:0.868-0.958;validation set AUC=0.845,95%CI:0.696-0.940)and peritumoral radiomic model(training set AUC=0.891,95%CI:0.835-0.937;validation set AUC=0.843,95%CI:0.690-0.936).The combined model of fusio

关 键 词：肺腺癌 磨玻璃结节 影像组学 体层摄影术 X射线计算机 瘤周微环境 

分 类 号：R734.2[医药卫生—肿瘤]