CT Based Radiomic Features For Prediction Of Epidermal Growth Factor Receptor Mutation In Advanced Lung Adenocarcinoma

Objective: To predict epidermal growth factor receptor(EGFR)mutation status of advanced lung adenocarcinoma using quantitative radiomic features.Methods:(1)The study included 339 patients with advanced lung adenocarcinoma admitted in Affiliated Hospital of Qingdao University from April 2013 to September 2019.Patients were divided into EGFR mutant group and EGFR wild group according to their EGFR mutation status.Retrospective study was carried out to compare the age,gender,smoking history,and TNM staging between the two groups.SPSS 22.0 software was used for data analysis.Chi-square test was used for comparison between two groups,and there was significantly difference when P<0.05.(2)There were 237 patients in the training group and 102 patients in the validation group.ITK-SNAP software was used for two-dimensional(2D)and manual segmentation of the largest layer of the tumor.A.K software was used for the extraction of quantitative radiomic features from the region of interest(ROI).The minimum redundancy maximum relevance(mRMR)method as well as the least absolute shrinkage and selection operator(LASSO)method was used to select the features in the training group.A multivariable logistic regression model was built using the radiomic features which were highly correlated with EGFR gene mutation.The receiver operating curve(ROC)was used for the evaluation of model performance in training group and validation group.The larger the area under the curve(AUC),the better the performance of the model.Results:(1)There were 182 patients of lung adenocarcinoma in the EGFR mutant group,and the mean age was 63.5±8.8 years(range from 39 to 84 years).157 patients of lung adenocarcinoma were in the EGFR wild group,and the mean age was 64.3 ± 8.1years(range from 38 to 82 years).The proportion of females in the EGFR mutant group was higher than that in the EGFR wild group(P<0.001).The proportion of nonsmokers in the EGFR mutant group was higher than that in the EGFR wild group(P=0.001).No statistical difference was found in the TNM staging between the two groups(P>0.05).(2)In total,396 radiomic features were extracted from the arterial phase ROI.At first,mRMR was performed to eliminate the redundant and irrelevant features,20 featureswere retained.LASSO was conducted to choose the optimized subset of features,then the most predictive subset including 15 features were chosen to construct the arterial phase radiomic model.AUC for arterial phase radiomic model is 0.75 in the training group and0.70 in the validation group.(3)In all,396 radiomic features were extracted from the venous phase ROI.At first,mRMR was performed to select 20 features.Then LASSO was conducted to choose 5 features to construct the venous phase radiomic model.AUC for venous phase radiomic model is 0.69 in the training group and 0.68 in the validation group.(4)We combined the features extracted from the arterial phase ROI with the features obtained from the venous phase ROI.Firstly,mRMR was performed to select 30 features in total.Then LASSO was conducted to choose 10 features to construct the combined radiomic model.AUC for combined radiomic model is 0.76 in the training group and 0.74 in the validation group.Conclusions:(1)EGFR mutations were correlated with females or nonsmokers with advanced lung adenocarcinoma.(2)Radiomic features were predictive for EGFR mutation status in advanced lung adenocarcinoma.(3)The arterial phase radiomic model showed better performance than venous phase radiomic model.(4)The combined radiomic model showed best performance in predicting the EGFR mutation status.