影像组学特征对磨玻璃结节型肺腺癌病理亚型的预测作用

The role of radiomics features in predicting pathological subtypes of ground glass nodular lung adenocarcinoma

目的验证影像组学模型鉴别磨玻璃结节(GGN)型肺腺癌病理亚型(浸润与非浸润)的准确性。方法回顾性收集184例结节影像学表现为GGN、术后病理证实为肺腺癌的180例手术病例,按照7∶3的比例将入组患者的结节随机分为训练组及验证组,训练组有129例结节,验证组有55例结节。选取磨玻璃结节CT图像中的1 mm平扫序列勾画感兴趣区域(ROI),从中提取960个影像组学特征,通过t检验及LASSO回归方法筛选出7个特征。利用支持向量机(SVM)的方法在训练组中构建影像组学模型,并在验证组中验证其准确性。应用受试者工作特征(ROC)曲线评估模型对GGN型肺腺癌病理亚型的预测效能。结果单因素分析结果显示,性别和病变位置与GGN的病理亚型差异无统计学意义(P>0.05),而年龄及CT图像中结节最大直径与GGN的病理亚型之间的差异有统计学意义(P<0.05)。影像组学模型在训练组的AUC(95%CI)值为0.94(0.89~0.99),准确度为91%、灵敏度95%、特异度为84%;在验证组的AUC(95%CI)值为0.88(0.83~0.93),准确度为83%、灵敏度84%、特异度为81%。结论基于影像组学特征建立的模型对GGN型肺腺癌病理亚型具有预测作用,可以为制定GGN型肺腺癌个体化诊疗策略提供依据。

Objective To verify the accuracy of the radiomics features model in identifying the pathological subtypes(infiltrating and non-infiltrating)of ground glass nodule(GGN)lung adenocarcinoma.Methods A retrospective collection of 184 cases of nodules showed GGN on imaging and 180 cases of lung adenocarcinoma confirmed by postoperative pathology was performed.The nodules of the enrolled patients were randomly divided into a training group and a verification group according to the ratio of 7∶3.129 nodules in the control group and 55 nodules in the verification group.The region of interest(ROI)was delineated by the 1 mm plain scan sequence in the ground-glass nodule CT image,and 960 radiomics features were extracted from it,and 7 features were screened out by t test and LASSO regression method.The radiomics model was constructed in the training set by using the method of support vector machine(SVM),and its accuracy was verified in the validation set.Receiver operating characteristic(ROC)curves were used to evaluate the predictive performance of the model for pathological subtypes of GGN lung adenocarcinoma.Results The results of univariate analysis showed that there was no significant difference between gender and lesion location and the pathological subtype of GGN(P>0.05),and there were statistically significant differences between age and the maximum diameter of nodules in CT images and the pathological subtype of GGN(P<0.05).The AUC(95%CI)value of the radiomics model in the training group was 0.94(0.89~0.99),the accuracy was 91%,the sensitivity was 95%,and the specificity was 84%.The AUC(95%CI)value in the verification group was 0.88(0.83~0.93),the accuracy was 83%,the sensitivity was 84%and the specificity was 81%.Conclusion The model established based on radiomics features can predict the pathological subtypes of GGN lung adenocarcinoma,and can provide the basis for formulating individualized diagnosis and treatment strategies for GGN lung adenocarcinoma.