Three Phase CT-based Radiomics Model For Prediction Of Synchronous Distant Metastasis In Patients With Clear Cell Renal Cell Carcinoma

Objective:To develop a clinical-radiomics combined model based on three phase CT radiomics score combined with clinical information and radiologic features for individualized prediction of synchronous distant metastasis in patients with clear cell renal cell carcinoma(ccRCC),and to evaluate the performance of the model.Methods:395 ccRCC patients involved in the study were recruited in hospital of southern medical university from January 2015 to September 2022.According to inclusive and exclusive criteria,finally 311 patients were included in the study(SDM negative n=168,SDM positive n=40),randomly divided into training cohort(n=218)and validation cohort(n=93)in a 7:3 ratio.Meanwhile,the clinical information and radiologic features of the patients were collected,including age,gender,location,nuclear grade,Ki67,CK7,tumor length,intratumoral necrosis,intratumoral calcification,venous invasion,and intratumoral angiogenesis.The independent predictors of SDM were obtained by univariate and multivariate logistic regression analysis,and was constructed for clinical model.Then the three-phase CT data of ccRCC patients were collected,including unenhanced phase,corticomedullary phase,and nephrographic phase.Through manually segmentation,feature extraction,ICC analysis,feature selection by mRMR and LASSO,radiomics model was constructed.A clinical-radiomics combined model was develop by Logistic regression and SVM,and to evaluate the performance of the model.Result:Ki67 and tumor length were independent predictors of SDM,and constructed the clinic model based on Logistic regression.The clinical model had an AUC value of 0.83(95%CI,0.755-0.905)in training cohort and 0.763(95%CI,0.638-0.888)in validation cohort.The radiomics model based on the feature signature extracted from the combination of unenhanced scan,corticomedullary phase and nephrographic phase had the best performance,with an AUC of 0.949(95%CI,0,9180-0.9809)in training cohort and 0.826(95%CI,0.730-0.922)in validation cohort.The AUC values of the combined model based on Logistic regression in training cohort and validation cohort were 0.956(95%CI,0.927-0.986)and 0.894(95%CI,0.818-0.973),respectively.The AUC values of the combined model based on SVM in the training cohort and validation cohort were 0.807(95%CI,0.731-0.883)and 0.676(95%CI,0.559-0.794),respectively.Moreover,the prediction performance of the combined Logistic regression model was better than radiomics model and the clinical model.Conclusions:Our combined clinics-imaging model based on the three-phase CT imaging omics score combined with clinical and image semantic features can provide an ideal individualized prediction of SDM in patients with ccRCC.