Correlation Between PET And CT Imaging Texture Features And EGFR Mutation Status In Advanced Non-Small Cell Lung Cancer

Purpose To investigate the correlation of texture features of the primary tumor on F-18 Fluorodeoxyglucose positron emission tomography(¹⁸F-FDG-PET)and computed tomography（CT）with epidermal growth factor receptor（EGFR）mutation status in patients with advanced non-small cell lung cancer（NSCLC）.MethodsIn this retrospective study,seventy-one patients with advanced NSCLC（AJCC eighth edition）were included between April 2014 and August 2018 in our institution,who underwent pretreatment ¹⁸F-FDG-PET/CT scanning and EGFR mutation tests.Age,gender and smoking status were collected from the institutional information system.EGFR mutation status was detected using amplifed refractory mutation system（ARMS）PCR technology,pyrosequencing or next-generation sequencing（NGS）.The PET and CT DICOM data were introduced into an in-house MATLAB code（Mathworks Inc,USA）.Seventy PET and Fifty-eight CT texture features were extracted from volumes of the primary tumor using an CGITA image heterogeneity analysis plug-in（Chang-Gung Image Texture Analysis toolbox,Taiwan province,China）.The chi-square test was used for categorical variables between 2groups.The correlation between texture features was investigated to address collinearity issues.The highly correlated features（correlation>0.8）were regarded as dependent features,which were not considered in this analysis.Thus,38 of 128features were considered as mutually independent features for analysis of factors associated with EGFR mutation.The method of logistic regression used was forward-conditional.The stepwise probability was set to 0.05 for entry and 0.10 for removal.The classification cut-off was 0.5 and the maximum number of iterations was 20.Omnibus tests of model coefficients were also conducted.Receiver operating characteristic（ROC）curves for each model were constructed and the AUC was calculated with EGFR mutation status.P<0.05 was considered as significantly statistical difference.SPSS 20.0 software and R 3.5.0 software were used for the statistical analysis.Results71 patients with Stage IV NSCLC who underwent pretreatment¹⁸F-FDG PET scans were analyzed,and all of them were lung adenocarcinoma.Thirty-nine patients were EGFR mutation,whereas 32 patients were wild-type.The median age was 58（range,27–84）and 32（45.1%）were female.The EGFR Mutation was significantly associated with female sex（P=0.026）.21 PET and 17 CT features were considered as mutually independent features for analysis of factors associated with EGFR mutation.Four PET（cooccurance second angular moment,cooccurance contrast,intensity-size-zone low-intensity large-zone emphasis and texture spectrum max spectrum）and two CT quantitative features（intensity-size-zone high-intensity zone emphasis and normalized cooccurance second angular moment）were independent predictors of EGFR mutation in multiple logistic regression analysis.The models with clinical variables and texture features were built,and the predictive ability of the model based on texture features and clinical variables was more precise than texture features alone（AUC,0.897 vs 0.864）.ConclusionTexture features of primary tumors can be useful to predict EGFR mutation in patients with Stage IV NSCLC,and to some extent the models of texture features and clinical variables can show useful information regarding EGFR mutations in Stage IV NSCLC when adequate specimens for the detection of EGFR mutation status can not be obtained.