AUTHOR=Yang Xiaohuang , Fang Chao , Li Congrui , Gong Min , Yi Xiaochun , Lin Huashan , Li Kunyan , Yu Xiaoping 

TITLE=Can CT Radiomics Detect Acquired T790M Mutation and Predict Prognosis in Advanced Lung Adenocarcinoma With Progression After First- or Second-Generation EGFR TKIs?

JOURNAL=Frontiers in Oncology

VOLUME=Volume 12 - 2022

YEAR=2022

URL=https://www.frontiersin.org/journals/oncology/articles/10.3389/fonc.2022.904983

DOI=10.3389/fonc.2022.904983

ISSN=2234-943X

ABSTRACT=Objective To explore the potential of CT radiomics in detecting acquired T790M mutation and predicting prognosis in patient with advanced lung adenocarcinoma with progression after first- or second- generation EGFR TKIs therapy.
Materials and methods Contrast-enhanced thoracic CT were collected from 250 lung adenocarcinoma patients (with acquired T790M mutation, n=146, without mutation, n=104) after progression on first- or second-generation TKIs. Radiomic features were extracted from each volume of interest. The maximum relevance minimum redundancy (mRMR) and the least absolute shrinkage and selection operator (LASSO) regression method were used to select the optimized features in detecting acquired T790M mutation. Univariate cox regression and LASSO cox regression were used to establish the Radiomics model to predict PFS of osimertinib treatment. Finally, nomograms combined clinical factors with Radscore to predict acquired T790M mutation and prognosis were built separately. And the two nomograms were validated by the concordance index (C-index), decision curve analysis (DCA) and calibration curve analysis where appropriate. 
Results Clinical factors including PFS of first-line EGFR TKIs, EGFR mutation and N stage and 12 radiomic features were useful in predicting acquired T790M mutation. The AUC values of ROC curve of Clinics, Radiomics and Nomogram models were 0.70, 0.74 and 0.78 in the training set, and was 0.71, 0.71 and 0.76 in the validation set, respectively. Decision curve analysis and calibration curve analysis demonstrated good performance of the nomogram model. Clinical factors including age and first-generation EGFR TKIs and 12 radiomic features were useful in patients’ outcome prediction. The C-index of the combined nomogram was 0.686 in the training set, 0.630 in the validation set, respectively. Calibration curves demonstrated relatively poor performance of the nomogram model.  
Conclusion Nomogram combined clinical factors with radiomic features might be helpful to detect whether patients developed acquired T790M mutation or not after progression on first- or second-generation EGFR TKIs. Nomogram prognostic model combined clinical factors with radiomic features might have the potential in predicting survival of patients harboring acquired T790M mutation treated with osimertinib.