The Value Of ¹⁸F-FDG PET/CT Radiomic Features In Predicting The Prognosis Of Cervical Cancer

Objective:To investigate the value of 18F-fluorodeoxyglucose(FDG)PET/CT radiomic features in predicting prognosis of cervical cancer patients before treatment.Method:From May 20,2015 to April 30,2021,a total of 113 patients with cervical cancer diagnosed in Sichuan Cancer Hospital were retrospectively collected,andall patients underwent 18F-FDG PET/CT within 2 weeks before treatment.All patients were randomly divided into training set and validation set in a ratio of 7:3,with 79 cases in training set and 34 cases in validation set.The LIFEx software wasused to to extract PET radiomic features from cercervical primary tumor,and the best radiomic features and clinical features were selected by R software packages "Hmisc" and"glmnet".Logistic regression,linear discriminant analysis(LDA),and random forest(RF)were used to establish radiomic feature models,and an optimal modeling method was selected.and logistic regression was used to establish radiomic feature model.Using logistic regression to build radiomic feature model,clinical feature model,and complex model combining with the two.The receiver operating characteristic curve(ROC)was used to determine the performance of the model to predict tumor prognosis,and the area under the curve(AUC)was calculated.Univariate and multivariate analysis of each prognostic factor was performed using COX proportional hazards regression model,and independent predictors of progression-free survival(PFS)were obtained.The survival curvewere drawn by the Kaplan-Meier method,and the difference in the survival curvebetween groupwas compared by the Log-rank method.Result:The five best radiomic features were selectedby R software packages" Hmisc" and "glmnet",namely SHAPE_Sphericity,GLCM_Correlation,GLRLM_RLNU,GLZLM_SZE and GLZLM GLNU.The two bestclinical features were selectedby R software packages "Hmisc",namelylymph node metastasis and staging.Logistic regression was used to construct a model of radiomics features,a model of clinical features and a complex model,and the AUROC were 0.810(p<0.05,95%CI0.663-0.956),0.875(p<0.05,95%CI0.602-0.898),and 0.845(p<0.05,95%CI0.711-0.979).Univariate and multivariate COX regression analysis showed that lymph node metastasis,tumor stage and GLRLM_RLNU were independent predictors of PFS(p=0.038,p<0.016,p=0.009).The survival curve was drawn by Kaplan-Meier,and the results showed that the PFS rate of the lymph node metastasis negative group,FIGO Ⅱ-Ⅲ stage,and GLRLM_RLNU<65.98 was significantly higher than that of the lymph node metastasis positive group,FIGO Ⅳ stage and GLRLM_RLNU≥65.98(p=0.001,p<65.98).0.001,p<0.001)..Conclusion:The prediction model constructed based on 18F-FDG PET/CT radiomic features has high performance in predicting the prognosis of cervical cancer,and the combination of clinical features can improve the prediction performance of the model.Therefore,18F-FDG PET/CT radiomics can assist in early clinical judgment of the prognosis of cervical cancer patients,optimize clinical decision-making,and improve patient prognosis.