杨昭,王小雷,李淑华,张群芳,张雪丽,李阳,李伟,王效静,刘浩,谢宗玉.CT影像组学列线图评估肺腺癌脏层胸膜侵犯[J].中国医学影像技术,2022,38(7):1017~1022
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
投稿时间:2021-12-07  修订日期:2022-04-04
DOI:10.13929/j.issn.1003-3289.2022.07.012
中文关键词:  肺腺癌  脏层胸膜侵犯  影像组学  列线图
英文关键词:adenocarcinoma of lung  visceral pleural invasion  radiomics  nomogram
基金项目:安徽省中央引导地方科技发展资金(2020b07030008)。
作者单位E-mail
杨昭 蚌埠医学院研究生院, 安徽 蚌埠 233004  
王小雷 蚌埠医学院研究生院, 安徽 蚌埠 233004  
李淑华 蚌埠医学院第一附属医院放射科, 安徽 蚌埠 233004  
张群芳 蚌埠医学院第一附属医院放射科, 安徽 蚌埠 233004  
张雪丽 蚌埠医学院第一附属医院放射科, 安徽 蚌埠 233004  
李阳 蚌埠医学院研究生院, 安徽 蚌埠 233004  
李伟 安徽省呼吸系统疾病(肿瘤)临床医学研究中心, 安徽 蚌埠 233004  
王效静 安徽省呼吸系统疾病(肿瘤)临床医学研究中心, 安徽 蚌埠 233004  
刘浩 北京医准智能科技有限公司, 北京 100089  
谢宗玉 蚌埠医学院第一附属医院放射科, 安徽 蚌埠 233004 zongyuxie@sina.com 
摘要点击次数: 203
全文下载次数: 90
中文摘要:
       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。
英文摘要:
      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.
查看全文  查看/发表评论  下载PDF阅读器