沈洁,张晔,金晓梅,孔杰俊.瘤内与瘤周CT影像组学模型结合临床及常规CT特征鉴别肺原位腺癌与微浸润性腺癌[J].中国医学影像技术,2024,40(6):869~873
瘤内与瘤周CT影像组学模型结合临床及常规CT特征鉴别肺原位腺癌与微浸润性腺癌
Differentiating adenocarcinoma in situ and microinvasive adenocarcinoma of lung based on intratumoral and peritumoral CT radiomics models combined with clinical and routine CT features
投稿时间:2024-01-27  修订日期:2024-04-12
DOI:10.13929/j.issn.1003-3289.2024.06.015
中文关键词:  肺腺癌  原位癌  体层摄影术,X线计算机  影像组学
英文关键词:adenocarcinoma of lung  carcinoma in situ  tomography, X-ray computed  radiomics
基金项目:南京医科大学科技发展基金(NMUB20230194)。
作者单位E-mail
沈洁 南京医科大学附属脑科医院胸科院区放射科, 江苏 南京 210009  
张晔 南京医科大学附属脑科医院胸科院区放射科, 江苏 南京 210009  
金晓梅 南京医科大学附属脑科医院胸科院区放射科, 江苏 南京 210009  
孔杰俊 南京医科大学附属脑科医院胸科院区放射科, 江苏 南京 210009 360364218@qq.com 
摘要点击次数: 130
全文下载次数: 114
中文摘要:
      目的 观察基于平扫CT瘤内及瘤周影像组学模型联合临床及常规CT特征鉴别肺原位腺癌(AIS)与微浸润性腺癌(MIA)的价值。方法 回顾性分析180例孤立性AIS及180例孤立性MIA肺结节患者,随机将其中各160例纳入训练集(n=320)、各20例纳入测试集(n=40)。以训练集AIS与MIA间差异有统计学意义的临床及常规CT特征构建临床模型;勾画瘤内(CTi)及包含瘤周2 mm(CTi+p2mm)、4 mm(CTi+p4mm)ROI,提取并筛选其影像组学特征,分别以之构建CTi模型、CTi+p2mm模型及CTi+p4mm模型;绘制受试者工作特征(ROC)曲线,计算曲线下面积(AUC),评价各模型效能,遴选预测测试集MIA效能最佳者,联合临床及常规CT特征构建联合模型,观察临床模型、最佳影像组学模型及联合模型的AUC、校准度及净收益。结果 训练集内,相比AIS,MIA结节直径较大、密度不均匀、伴血管穿行结节占比较高(P均<0.05)。CTi+p2mm模型鉴别测试集MIA与AIS效能最高(AUC=0.838,P<0.05);以之结合临床及常规CT特征构建的联合模型鉴别诊断效能更佳(AUC=0.867,P<0.05)。联合模型的校准度及0.60~0.90阈值概率区间的临床净收益较高。结论 基于平扫CT构建的瘤内和瘤周2 mm ROI影像组学模型能有效鉴别肺MIA与AIS,联合临床及常规CT特征可进一步提高模型鉴别效能。
英文摘要:
      Objective To observe the value of intratumoral and peritumoral radiomics models combined with clinical and routine CT features for differentiating adenocarcinoma in situ (AIS) and microinvasive adenocarcinoma (MIA) of lung. Methods Totally 180 patients with isolated AIS and 180 with isolated MIA were retrospectively included, among them 160 AIS cases and 160 MIA cases were randomly selected into training set (n=320), while the other 20 AIS cases and 20 MIA cases were selected into test set (n=40). In training set, clinical and conventional CT features being statistically different between AIS and MIA were obtained to construct clinical model. Besides, radiomics features were extracted from intratumoral (CTi) ROI, intra- and peritumoral 2 mm (CTi+p2mm) ROI and intra- and peritumoral 4 mm (CTi+p4mm) ROI, and then CTi model, CTi+p2mm model and CTi+p4mm model for differentiating MIA and AIS were constructed. The optimal radiomics model for predicting MIA was selected using the area under the curve (AUC) of receiver operating characteristic (ROC) curve, and a combined model was built based on the optimal radiomics model combining with clinical and conventional CT features. The AUC, calibration and net benefit of the clinical model, the optimal radiomics model and the combined model were assessed. Results In training set, the larger nodular diameter, higher percentage of inhomogeneous density and ratio of nodules with vascular signals were observed in MIA compared with those in AIS (all P<0.05). In test set, CTi+p2mm model had the highest efficacy (AUC=0.838) for differentiating MIA from AIS (P<0.05), and the combined model had better efficacy (AUC=0.867, P<0.05). The calibration of combined model was good, and the net benefit was high in 0.60-0.90 threshold probability range. Conclusion The radiomics model constructed with intratumoral and peritumoral 2 mm ROI based on plain CT was effective for differentiating MIA from AIS. Combining with clinical and routine CT features could furtherly improve differential diagnostic efficacy.
查看全文  查看/发表评论  下载PDF阅读器