基于^(18)F-FDG PET/CT的磨玻璃结节早期肺腺癌浸润性预测模型的构建及验证

Construction and validation of the predictive models for the pathological invasion of early lung adenocarcinoma presenting as ground glass nodules based on^(18)F-FDG PET/CT

目的基于^(18)F-FDG PET/CT构建并验证以磨玻璃结节(GGNs)为影像特征的早期肺腺癌病理浸润性的预测模型。方法回顾分析2011年10月至2019年10月间常州市第一人民医院的因GGNs行PET/CT显像并手术切除经病理证实为浸润前病变及微浸润腺癌(MIA)、浸润性腺癌(IAC)的患者149例,男44例、女105例,年龄(61.1±8.9)岁。将检出的GGNs按1∶1分为建模组和验证组。使用Mann-Whitney U检验或χ^(2)检验比较浸润前病变/MIA与IAC形态学定性特征(如形状、边缘特征等)、定量参数[如GGNs实性成分比例、磨玻璃成分CT值(CT_(GGO))等)]、功能学定量参数[如SUV_(max)、SUV指数(SUV_(index);即GGNs SUV_(max)/肝SUV_(mean))]的差异,基于有意义的参数构建logistic回归模型,应用ROC曲线验证模型的稳健性。采用Delong法比较AUC。结果170枚GGNs获手术切除并经病理证实。建模组(n=89)中,IAC患者混合性GGNs、不规则形状、边缘特征、支气管扩张/扭曲/截断征比例、GGNs最大径和实性成分最大径、实性成分比例、CT_(GGO)、SUV_(max)、SUV_(index)均高于浸润前病变/MIA患者(χ^(2)值:5.00~23.40,z值:-6.53~-2.70,均P<0.05)。基于PET/CT定性参数(GGNs类型、边缘特征)、定量参数(CT_(GGO)、SUV_(index))、联合定性和定量参数(GGNs类型、边缘特征、SUV_(index))分别成功构建模型1~3,三者在建模组中ROC AUC分别为0.896、0.880和0.931,其中模型2在验证组(n=81)中的AUC无明显下降(0.802;z=0.81,P=0.417)。结论联合^(18)F-FDG PET/CT形态学和功能学定量参数构建的早期肺腺癌病理浸润性预测模型具有较好的效能且模型稳健。

Objective To construct and verify of the predictive models for pathologic invasion of early lung adenocarcinoma with ground glass nodules(GGNs)based on^(18)F-FDG PET/CT.Methods A retrospective analysis was conducted on 149 patients(44 males,105 females;age(61.1±8.9)years)with pre-invasive lesions/minimally invasive adenocarcinoma(MIA)and invasive adenocarcinoma(IAC)confirmed by pathology after surgery in the First People′s Hospital of Changzhou from October 2011 to October 2019.All patients underwent PET/CT for GGNs.GGNs were randomly divided into a modeling group and validation group with the proportion of 1∶1.Mann-Whitney U test orχ^(2)test was used to compare the qualitative morphological characteristics(shape,edge characteristics,etc.),quantitative parameters(consolidation-to-tumor ratio,attenuation value of the ground glass opacity(GGO)component on CT(CT_(GGO)),etc.)and quantitative functional parameters(SUV_(max)and SUV_(index)(GGNs SUV_(max)/liver SUV_(mean))of pre-invasive lesions/MIA and IAC.Logistic regression analysis was used to construct the models,and the ROC curve was used to verify the models′robustness.Different AUCs were compared by Delong test.Results A total of 170 GGNs were removed by surgery and confirmed pathologically.In the modeling group(n=89),the proportion of mixed GGNs,irregular shape,edge characteristics,bronchiectasis/twist/truncation sign,GGNs_(max)imum diameter and solid component_(max)imum diameter,consolidation-to-tumor ratio,CT_(GGO),SUV_(max)and SUV_(index)in IAC group were significantly higher than those in pre-invasive/MIA group(χ^(2)values:5.00-23.40,z values:from-6.53 to-2.70,all P<0.05).Models 1-3 were constructed based on the qualitative parameters(GGNs type,edge characteristics),quantitative parameters(CT_(GGO),SUV_(index)),combined qualitative and quantitative parameters(GGNs type,edge characteristics,SUV_(index))of PET/CT,respectively,and the AUCs of ROC were 0.896,0.880 and 0.931 in the modeling group,respectively.And the AUC of model 2 was not decreased significantly in the validation group(n=81;AUC=0.802;z=0.81,P=0.417).Conclusion The model combined with morphological and functional quantitative parameters of^(18)F-FDG PET/CT can effectively predict the pathological invasion of early lung adenocarcinoma,and the constructed model is robust.