CT影像组学列线图评估肺腺癌脏层胸膜侵犯

CT radiomics nomogram for evaluating visceral pleural invasion of lung adenocarcinoma

目的建立CT影像组学列线图模型,观察其评估肺腺癌脏层胸膜侵犯(VPI)的价值。方法回顾性分析183例经术后病理证实的肺腺癌患者,以7∶3比例将其随机分为训练集(n=128)及验证集(n=55);根据有无VPI进一步分为浸润组和非浸润组。基于肺CT图像提取影像组学特征,构建影像组学评分(Rad-score)。以多因素logistic回归分析筛选训练集内组间差异具有统计学意义的影像学表现,构建常规模型,并结合Rad-score绘制列线图,对比观察其判断肺腺癌伴VPI的效能,以及列线图模型判断腺癌伴VPI结果与实际结果的一致性及其差异。结果最终以8个影像组学特征构建Rad-score。多因素logistic回归分析显示,病灶存在分叶征、瘤内坏死、胸膜牵拉及Rad-score是判断肺腺癌伴VPI的独立因素(P均<0.05)。以同时存在分叶征、瘤内坏死及胸膜牵拉为常规模型,结合Rad-score所绘制的列线图模型判断训练集与验证集肺腺癌伴VPI的AUC分别为0.875、0.865,优于常规模型在训练集的0.779、验证集的0.805,以及Rad-score模型在训练集的0.810和验证组的0.803,差异具有统计学意义(P均<0.05)。校准曲线及Hosmer-Lemeshow检验结果均显示,列线图模型判断训练集及验证集肺腺癌患者伴VPI与实际结果的一致性良好(P均>0.05)。结论CT影像组学列线图模型判断肺腺癌伴VPI应用价值良好。

Objective To establish CT radiomics nomogram model,and to evaluate its value for evaluating visceral pleural invasion(VPI)of lung adenocarcinoma.Methods Data of 183 patients with lung adenocarcinoma confirmed by surgical pathology were retrospective analyzed.The patients were randomly divided into training set(n=128)and validation set(n=55)at the ratio of 7∶3,and further divided into the invasive group and non-invasive subgroup according to the presence or absence of VPI.Radiomic scores(Rad-score)were constructed based on radiomic features the extracted from lung CT images.Multivariate logistic regression analysis was used to screen imaging findings being significant different between groups in training set,then a conventional model was constructed,and the nomogram was drawn with Rad-scores to comparatively evaluate the efficiency of models for judging lung adenocarcinoma with VPI,and the consistency and difference of relative results were compared with the actual findings of adenocarcinoma with VPI.Results Eight imaging features were selected to construct Rad-score.Multivariate logistic regression analysis showed that lobulation,intratumor necrosis,pleural traction and Rad-score were all independent factors for judging lung adenocarcinoma with VPI(all P<0.05).Taken the simultaneous presence of lobulation,intratumor necrosis and pleural traction as the conventional model,combining with the nomogram drawn by Rad-score,the AUC in training set and validation set was 0.875 and 0.865,respectively,better than conventional model for judging lung adenocarcinoma with VPI of training set and validation set(AUC=0.779,0.805)and Rad-score model of training set and validation set(AUC=0.810,0.803),respectively,and the differences were both significantly obvious(both P<0.05).The calibration curve and Hosmer-Lemeshow test showed that the judging results of the nomogram model for lung adenocarcinoma with VPI in training set and validation set were both in good agreements with the actual status(both P>0.05).Conclusion CT radiomics nomogram had good application value for judging lung adenocarcinoma with VPI.