Noninvasive Imaging Biomarker Of PET/CT Features For Predictions Of VEGF And Prognosis In Patients With Gastric Cancer

Objective:Vascular Endothelial Growth Factor(VEGF)is associated with poor prognosis in gastric cancer.There is a lack of non-invasive dynamic predictive indicators for VEGF in clinical practice.This study aims to establish a non-invasive model for predicting VEGF status and survival prognosis in gastric cancer patients through radiomics of[18F]FDG PET/CT.Methods:In this retrospective clinical study,according to the results of pathological diagnosis,the patients were randomly divided into training cohort,validation cohort and test cohort.All patients received[18F]FDG PET/CT examination and met the criteria established in this study.Experienced imaging physicians were then invited to use software to outline features of the tumor tissue region of interest in PET/CT images.The outlined regions of interest were extracted using the R analysis package.The methods of radiomics were used to calculate the radiomics features that can predict VEGF status and to establish the Radiomics Score(RS)formula and calculate the score for each patient.The RS threshold was set through the Receiver Operating Characteristic curve(ROC).A noninvasive prediction model of VEGF status was constructed by combining other clinical predictive features that could predict VEGF status counted by univariate and multivariate logistic regression analysis.For survival,RS and other clinical predictive features that could predict the survival of patients were measured by univariate and multivariate Cox regression analysis,and a noninvasive prediction model of survival was constructed.In both models,we used ROC and calibration curves to assess the diagnostic accuracy and diagnostic efficiency of the models.The Decision Curve Analyses(DCA)approach was used to determine the clinical impact and clinical utility of the noninvasive prediction model of VEGF status.To assess the predictive power of the noninvasive prediction model of survival in predicting the overall survival of gastric cancer patients,we used Harrell’s concordance index(C-index)to assess the predictive discrimination of the model.Results:We identified radiomics features that could predict VEGF status and established a formula for calculating the RS,and determined the threshold of RS as0.0137.After logistic regression analysis,the AUCs of ROC in the noninvasive prediction model of VEGF status combining RS and other clinical features were 0.861(95%CI,0.791-0.915)in the training cohort and 0.857(95%CI,0.758-0.927)in the validation cohort,respectively.While the calibration curves exhibited the combination of RS and other clinical features of the noninvasive prediction model of VEGF status had good predictive and discriminatory ability.The decision curve analysis demonstrated its good clinical utility.In terms of patients’ prognosis,multivariate Cox regression demonstrated that RS was an independent risk prognostic factor for overall survival(OS)(HR:5.063,95%CI,1.018-25.176).Harrell’s concordance index demonstrated that the noninvasive prediction model of survival with RS in combination with other clinical characteristics had better clinical predictive power than clinical characteristics alone(C-index:0.783,95%CI,0.730-0.836).Conclusion:The VEGF status and survival prediction model of gastric cancer based on[18F]FDG PET/CT could predict the VEGF status with gastric cancer patients,and patients predicted to be VEGF(+)had poor prognosis...