CT影像组学分层鉴别三种常见孤立性肺结节的价值

Stratified differentiation of 3 common pulmonary nodules with CT radiomics

目的 探讨基于CT影像组学特征的逻辑回归(LR)模型分层鉴别肺孤立性腺癌、结核、非结核炎性结节的价值。方法 回顾性分析2018年1月至2022年1月于陆军军医大学第二附属医院经病理证实为肺腺癌、结核、非结核炎性结节患者的临床和肺部CT资料。通过勾画肺结节感兴趣区并提取、筛选CT影像组学特征,分别建立肺腺癌与炎性结节预测模型以及结核与非结核炎性结节预测模型。通过绘制ROC并计算AUC、灵敏度和特异度评估模型效能。结果 共收集526例孤立性肺结节,其中肺腺癌263例,结核结节99例,非结核炎性结节164例。基于CT影像组学特征和临床危险因素建立的肺腺癌与炎性结节LR预测模型在训练集中和测试集中的AUC分别为0.880、0.886。基于CT影像组学特征建立的结核与非结核炎性结节LR预测模型在训练集中和测试集中的AUC分别为0.921、0.853。结论 基于CT影像组学特征建立的LR预测模型在分层鉴别3种常见孤立性肺结节中具有较好的性能,具有较高的临床应用价值。

Objective To explore the value of a logistic regression(LR)model based on CT radiomics features for the stratified classification of isolated adenocarcinoma pulmonary nodules,tuberculous nodules and non-tuberculous inflammatory pulmonary nodules.Methods The clinical and CT data of the patients pathologically diagnosed with pulmonary adenocarcinoma,tuberculosis,and non-tuberculous inflammatory nodules in our hospital between January 2018 and January 2022 were collected and retrospectively analyzed.By contouring the region of interest for pulmonary nodules and extracting CT radiomics features,prediction models were established to distinguish pulmonary adenocarcinoma vs inflammatory nodules and tuberculosis vs non-tuberculous inflammatory nodules.The model’s performance was assessed by plotting receiver operating characteristic(ROC)curves and calculating area under curve(AUC),sensitivity,and specificity.Results A total of 526 solitary pulmonary nodules were collected,including 263 cases of pulmonary adenocarcinoma,99 cases of tuberculous nodules,and 164 cases of non-tuberculous inflammatory nodules.In the training and validation sets,the LR models based on CT radiomics features and clinical risk factors achieved an AUC value of 0.880 and 0.886,respectively,for distinguishing pulmonary adenocarcinoma from inflammatory nodules.For discriminating tuberculosis versus non-tuberculous inflammatory nodules,the LR models based on CT radiomics features yielded an AUC value of 0.921 in the training set and of 0.853 in the validation set.Conclusion The LR prediction models based on CT radiomics features demonstrate excellent performance in hierarchically identifying the 3 prevalent solitary pulmonary nodules,with substantial clinical significance.