Objective To observe the value of intratumoral and peritumoral CT radiomics for predicting Ki-67 expression level of lung adenocarcinoma with mixed ground glass nodules.Methods Totally 284 cases of pathologically confirmed lung adenocarcinoma with mixed ground glass nodules were retrospectively enrolled,among them 197 cases were taken as training set(54 cases with high and 143 cases with low Ki-67 expression)and 87 cases as validation set(27 cases with high and 60 cases with low Ki-67 expression).Intratumoral and peritumoral radiomic features were obtained from non-contrast chest CT,and radiomic models for predicting Ki-67 expression in lung adenocarcinoma with mixed ground glass nodules were established using adaptive boosting,light gradient boosting machine(LightGBM)and multilayer perceptron algorithms based on intratumoral features,peritumoral features,as well as intratumoral+peritumoral features,respectively,and the optimal radiomics signature was selected according to the area under the receiver operating characteristic curve(AUC).Univariate and multivariate logistic regression analysis were performed to identify independent impact factors of Ki-67 expression level,and a clinical model was constructed,and the efficacy of the models were evaluated.Results Among radiomics models,LightGBMintratumoral+peritumoral model had the highest AUC(0.934 in training set and 0.845 in validation set),which were superior to that of clinical model(0.616 in training set and 0.684 in validation set)(both P＜0.05)Conclusion Intratumoral and peritumoral CT radiomics had good efficacy for predicting Ki-67 expression level of lung adenocarcinoma with mixed ground glass nodules.

Objective To observe the value of intratumoral and peritumoral CT radiomics for predicting Ki-67 expression level of lung adenocarcinoma with mixed ground glass nodules.Methods Totally 284 cases of pathologically confirmed lung adenocarcinoma with mixed ground glass nodules were retrospectively enrolled,among them 197 cases were taken as training set(54 cases with high and 143 cases with low Ki-67 expression)and 87 cases as validation set(27 cases with high and 60 cases with low Ki-67 expression).Intratumoral and peritumoral radiomic features were obtained from non-contrast chest CT,and radiomic models for predicting Ki-67 expression in lung adenocarcinoma with mixed ground glass nodules were established using adaptive boosting,light gradient boosting machine(LightGBM)and multilayer perceptron algorithms based on intratumoral features,peritumoral features,as well as intratumoral+peritumoral features,respectively,and the optimal radiomics signature was selected according to the area under the receiver operating characteristic curve(AUC).Univariate and multivariate logistic regression analysis were performed to identify independent impact factors of Ki-67 expression level,and a clinical model was constructed,and the efficacy of the models were evaluated.Results Among radiomics models,LightGBMintratumoral+peritumoral model had the highest AUC(0.934 in training set and 0.845 in validation set),which were superior to that of clinical model(0.616 in training set and 0.684 in validation set)(both P＜0.05)Conclusion Intratumoral and peritumoral CT radiomics had good efficacy for predicting Ki-67 expression level of lung adenocarcinoma with mixed ground glass nodules.