Combination Of ¹⁸F-FDG PET/CT Radiomics And Clinical Features To Predict EGFR Gene Mutation Status In Lung Adenocarcinoma

Objective:To identify epidermal growth factor receptor（EGFR）mutations in lung adenocarcinoma based on ¹⁸F-fluorodeoxyglucose（FDG）positron emission tomography/computed tomography（PET/CT）radiomics and clinical features,and to explore the feasibility of differentiating distinguish EGFR exon 19 deletion（19 del）and exon 21 L858R missense（21 L858R）mutations by using PET/CT radiomics.Methods:A total of 179 patients with lung adenocarcinoma were retrospectively included.74 patients were EGFR wild type and 105 patients were EGFR mutant.Age,gender,smoking history,maximum tumor length of primary lesion,maximum standard uptake value（SUVmax）,mean standard uptake value（SUVmean）,metabolic tumor volume（MTV）and total lesion glycolysis（TLG）were recorded for all patients.The tumor primary lesions were manually outlined layer by layer on PET and CT images,respectively,to generate the tumor three-dimensional region of interest（ROI）.Subsequently,2632 radiomic features（1316 for PET images and 1316 for CT images）were calculated based on the tumor three-dimensional ROI.They were randomly assigned to the training（n=125）and testing（n=54）groups in a 7:3 ratio.The radiomic features were screened using ANOVA,Pearson correlation analysis,and single-factor and multi-factor logistic regression analysis,and six radiomic features（two PET features and four CT features）were finally retained to calculate the radiomic score（rad-score）for each patient.Meanwhile,the PET/CT radiomic model was established by logistic regression for predicting EGFR mutations.Subsequently,a combined model was constructed by integrating the gender,smoking history,and maximum tumor length of lung adenocarcinoma patients based on the PET/CT rad-score,and the efficacy of the PET/CT radiomic model and the combined model in identifying EGFR mutation status in lung adenocarcinoma was compared by the receiver operating characteristics curve（ROC）and area under the curve（AUC）.In addition,among the 99 samples（46 cases 19 del,53cases 21 L858R）,the same ratio of 7:3 was divided into training and testing groups,and after screening the features of PET/CT radiomics,a differential model of EGFR mutation subtype was established by logistic regression to distinguish 19 del from 21 L858R.Results:There were no statistical differences in age,SUVmax,and SUVmean between EGFR mutant and EGFR wild type between the training and testing groups.There were more patients with no history of smoking in EGFR mutant than in EGFR wild type（P=0.001,P<0.001）.In the training group,women were more likely to have EGFR mutations（P=0.001）,while the difference was not significant in the testing group（P=0.074）.In the testing group,the maximum tumor length diameter,MTV and TLG were lower in the EGFR mutant than in the EGFR wild type（all P<0.05）,but the difference was not statistically significant in the training group.According to ROC curve,in the training group,the AUC of PET/CT radiomic model was 0.853,and the accuracy of diagnosing EGFR mutational status was 78.4%;the AUC and the accuracy of the combination model were 0.882 and 81.6%,respectively.In the testing group,the AUC of PET/CT radiomic model was 0.804,and the accuracy was 74.1%;the AUC and accuracy of the combination model were 0.837 and 74.1%,respectively.In the subsample of cases,there was no difference between the EGFR 19 del and 21 L858R in terms of gender,smoking history,maximum tumor length,SUVmax,SUVmean,MTV,and TLG（all P>0.05）.Moreover,the AUC and accuracy of the PET/CT radiomics model to distinguish the EGFR mutational subtypes was 0.708 and 66.7%,respectively,in the training group and 0.652 and 56.7%,respectively,in the testing group.Conclusion:The combination model constructed by integrating PET/CT radiomics and clinical features can predict the EGFR mutation status of lung adenocarcinoma.However,it was more challenging to distinguish EGFR 19 del and 21 L858R mutational subtypes using PET/CT radiomics.