Predictive Power Of PET/CT Uptake, Clinical Characteristics And Radiomics For Gene Mutations In Non-small Cell Lung Cancer Patients

Objective: 18-fluorodeoxyglucose positron emission tomography/computed tomography(18F-FDG PET/CT)and radiomics were used to investigate the predictive value of 18F-FDG maximum standard uptake value,clinical features and radiomics features in gene mutations of non-small cell lung cancer.Materials and Methods:The PET/CT data and clinical laboratory data of 127 patients with pathologically confirmed NSCLC were retrospectively analyzed,and the mutation status of EGFR was grouped into 68 patients with mutant type and 59 patients with wild type.All subjects were scanned with a GE Discovery STEPET/CT scanner in a quiet state,followed by a PET scan.PET and CT images were automatically fused at Xeleris or AW 4.4 workstations.The region of interest was delineated at the most obvious and concentrated part of the primary lesion,and the SUVmax value was obtained.Tumor lesions on CT and PET images were manually segmented and radiomics were extracted.Groups were grouped by EGFR mutation status,and t-test,Wilcoxon rank-sum test and chi-square analysis were used to analyze the differences in SUVmax,clinical laboratory characteristicsbetween groups,selectkbest and Variance Thresholdwere used to extract the radiomicsl features with Statistical significance.Logistic regression analysis was used for modeling,and ROC curve analysis was used to analyze the predictive value of clinical models,radiomics models and joint models based on CT and PET images for gene status.And make a comparison.RESULTS: 1.There was no significant difference between EGFR mutation and age,size,SUVmax,and carcinoembryonic antigen(CEA)(P>0.05),But was correlated with smoking history,gender,and carbohydrate antigen(CA125)(P<0.05).2.There was no significant difference between EGFR mutation and SUVmax(P>0.05).3.The clinical model established by combining smoking history,gender and carbohydrate antigen 125 has a good performance in predicting EGFR mutation.Area under curvewas 0.786(95%ci: 0.663-0.909,P=0.000),with sensitivity of 0.865 and specificity of 0.667.4.Three features that were significantly correlated with EGFR mutation were extracted from CT and PET images(P<0.05).They are Median firstorder wavelet-hhl,Correlation GLCM wavelet-lll and Elongation shape original respectively.Mean firstorder wavelet-HHL,Median firstorder wavelet-HLH,Short Run Emphasis GLRLM exponential.5.The model established based on the radiomics characteristics of CT images can distinguish the EGFR status(AUC= 0.775).Combining the model with the clinical characteristics(AUC=0.786)can improve the accuracy of prediction(AUC= 0.867),but it is not significant(P>0.05).The PET imaging radiomics model can distinguish the EGFR state well(AUC= 0.817),and combine the model with the clinical model(AUC= 0.786)can significantly improve the performance(AUC= 0.927).There was no significant difference in the predictive performance of joint model based on CT and PET(P>0.05).Conclusion: 1.There is no significant correlation between SUVmax and EGFR mutation state.2.The clinical model combining smoking history,gender and CA125 can provide a reference for clinical judgment of EGFR mutation status.3.The radiomics features(Median firstorder wavelet-hhl,Correlation GLCM wavelet-lll,Elongation shape original)based on CT and PET;Mean firstorder wavelet-HHL,Median firstorder wavelet-HLH,Short Run Emphasis GLRLM exponential can reflect the functional status and morphological changes within the region of interest from the overall level and microstructure,and have a good prediction performance for the EGFR mutation of NSCLC.4.Advanced biomarkers based on PET/CT imaging features and clinical features may be used to predict EGFR mutation status.