| Research BackgroundLung cancer is the cancer with the highest morbidity and mortality in China.Mutations in the epidermal growth factor receptor(EGFR)are most common in lung adenocarcinoma.In recent years,targeted therapy has become an important treatment method for patients with advanced lung cancer,especially tyrosine kinase inhibitors(TKIs)targeting EGFR kinase domains.EGFR mutations,in addition to being valuable in the selection of treatment options,can also be used as a prognostic indicator of survival benefits.Therefore,it is important to determine whether patients with lung cancer have EGFR mutations.A number of scholars have tried to identify EGFR mutations by imaging,and many previous studies have explored the value of single CT image signs and CT radiomics in the identification of lung cancer gene mutations.With more and more application of PET/CT in the clinical practice,18F-FDG PET radiomics researches increase gradually,which are mainly concentrated on curative effect evaluation and prognosis prediction,etc..In terms of genetic mutations,there are always many researches on the relationship between EGFR mutations and PET parameters,while few on PET radiomics.Therefore,the purpose of this study is to investigate the predictive value of PET/CT radiomics for EGFR mutation in lung adenocarcinomaResearch purpose1.To compare the advantages and disadvantages of traditional PET metabolic parameters and feature parameters in predicting EGFR mutations in lung adenocarcinoma.2.To compare the advantages and disadvantages of PET radiomics,CT radiomics and PET combined CT radiomics in predicting EGFR mutations in lung adenocarcinoma3.To establish a combined model for predicting EGFR gene mutation in lung adenocarcinoma based on PET combined CT radiomics and CT image signs,and evaluate its diagnostic efficacy.Materials and methodsPatients with lung adenocarcinoma who underwent PET/CT examination in Nanfang hospital PET center from January to December 2017 and met the inclusion criteria were retrospectively collected.Clinical and imaging data of the patients were collected.The CT images of the primary tumor were collected,and the contour of the primary tumor was manually delineated layer by layer on PET and CT images respectively,and the corresponding radiomics parameters,including traditional parameters and feature parameters,were extracted.According to the results of homogeneity and normality of variance tests,the two independent sample t test or approximate t test or mann-whitney U test were used for the continuous variables,and the categorical variables were used 2 or Fisher’s exact probability test.For the parameters of radiomics,the least absolute shrinkage and selection operator(LASSO)and ten-fold cross-validation were used to verify the preliminary parameters,and logistics regression was used to further select parameters and build the model.The receiver operating characteristic(ROC)curve was used to analyze the diagnostic efficacy of each model,and the corresponding area under the curve(AUC),sensitivity,specificity and compliance rate were calculated and compared.Internal validation was used to verify the model with Bootstrap resampling method and calibration curves were drawn.ResultsA total of 114 patients(53 with EGFR mutation and 61 with EGFR wild type)were included.Univariate analysis showed no significant difference in EGFR expression in age,gender,smoking,and TNM stage(all P>0.05).Among the traditional PET parameters,only the difference between SUVmax in the EGFR mutant group and the wild type group was statistically significant(P=0.029),and its AUC was 0.600(P=0.066).Among CT image signs,only emphysema was statistically different between the two groups(P=0.027).PET feature parameters,PET feature parameters combined with traditional parameters(namely PET radiomics parameters),CT radiomics parameters,PET+CT radiomics parameters,PET+CT radiomics parameters combined with emphysema(namely combined parameters)were performed separately by LASSO screening and Logistic regression modeling to build 5 models:PET feature model(Mpetfeature),PET radiomics model(Mpet),CT radiomics model(Mct),PET+ CT radiomics model(Mpetct),combined Model(Mcombine).The AUC of the above 5 models were:0.661,0.730,0.752,0.866,0.677 respectively,sensitivity of 69.8%,83.0%,58.5%,92.5%,79.2%respectively,specificity of 60.7%,59.0%,82.0%,70.5%,83.6%respectively,the compliance rate were 64.9%,70.2%,71.1%,80.7%,81.5%respectively.Delong’s test showed that there was no significant statistical difference in AUC between Mpet and Mpettexture(P=0.060).Among Mpet,Mct,and Mpetct,there was no statistical difference in AUC between Mpet and Mct(P=0.749),but there were statistical differences in AUC between Mpetct and Mpet(P<0.05),Mpetct and Mct(P<0.05).In Mpetct and Mcombine,there was no statistical difference in AUC(P=0.408).The differences in sensitivity(P<0.05)and specificity(P<0.05)between the two models were statistically significant.In summary,of all the models,Mcombine has the highest AUC,specificity,and coincidence rate,and the Mpetct has the highest sensitivity.ConclusionsIn this study,it was found that both PET radiomics and CT radiomics could predict the EGFR mutation status of lung adenocarcinoma,and there was no statistical difference in the prediction efficacy between the two,both of which were superior to the traditional PET parameter SUVmax.The prediction efficiency of SUVmax was poor.In the prediction of EGFR mutation in lung adenocarcinoma by PET/CT radiomics,the combined application of PET and CT radiomics with the same machine can significantly improve the efficacy,both of which were better than the single PET radiomics,CT radiomics and PET traditional parameters.After adding emphysema to the PET combined CT radiomics model,the AUC and coincidence rate were slightly increased,and the specificity was significantly improved.Therefore,in the study of PET radiomics,adding the same machine CT radiomics and CT image sign could significantly improve the prediction efficiency. |
PDF Full Text Request