¹⁸F-FDG PET/CT Radiomics Of EGFR Gene Phenotype In Lung Adenocarcinoma

Objective Targeted therapy for epidermal growth factor receptor（EGFR）is an effective treatment strategy to improve the prognosis of patients with lung adenocarcinoma,but due to factors such as technology,cost,tumor heterogeneity,it still faces great challenges in clinical practice.In view of this,we constructed a combined model based on ¹⁸F-FDG PET/CT radiomics features combined with clinical-imaging parameters to predict the EGFR genotype in patients with lung adenocarcinoma.Methods This study retrospectively analyzed 103 patients with pathologically confirmed lung adenocarcinoma from June 2017 to December 2021.All patients received PET/CT examinations before surgery.The results of genetic testing were 52 cases of EGFR positive and 51 cases of EGFR negative.Clinical data of patients,qualitative features of CT images and semi-quantitative parameters based on PET images were analyzed including SUVmax,SUVmean,SUVpeak,MTV（metabolic tumor volume）,and TLG（total lesion glycolysis）.The region of interest of the entire tumor was delineated on each axial plane of the CT and PET images,and the region of interest was identified using Py Radiomics software.Univariate analysis,correlation analysis and LASSO dimensionality reductionanalysis were used to screen relevant image feature parameters.Logistic regression（LR）classifier algorithm was used to classify EGFR positive/negative,and 5 models were established,namely clinical-imaging model,radiomics models（based on CT,PET,PET/CT images）and combined models.The predictive propertyof five models was compared by the area under the receiver operating characteristic curve（AUC）,and finally the calibration curve was to analyze and evaluate the performance of the prediction model.Results The analysis result of clinical-imaging characteristics showed that there were significant differences in gender and SUVmax between the mutant group and the wild group（P<0.05）.Finally,11 radiomics features and 2 clinical-imaging factors were selected to build a joint model,and the logistic regression classifier was used to model the joint model.The AUC of the training group was 0.802（0.747-0.856）,and the AUC of the validation group was0.794（0.752-0.835）.The PET/CT radiomics combined clinical-imaging parameter model can significantly distinguish EGFR wild-type from mutant EGFR,which is superior to the CT radiomics model alone,the PET radiomics model alone,and the PET/CT radiomics model.The combined PET/CT-based model showed better performance in predicting EGFR mutations.Conclusion（1）High-throughput radiomics features extracted from different image feature sets can be used to predict the EGFR mutation status of patients with lung adenocarcinoma,and show higher predictive power than pure clinical features and traditional image-derived indicators.（2）¹⁸F-FDG PET/CT radiomics features integrated with clinical-imaging parameters can effectively predict EGFR mutation status in patients with lung adenocarcinoma.