Enhanced CT-Based Radiomics Analysis For Preoperative Prediction Of Fuhrman Grade And Survival In Clear Cell Renal Cell Carcinoma

Objictive To explore the value of enhanced CT-based radiomics in predicting Fuhrman grade in patients with clear cell renal cell carcinoma(ccRCC)before surgery;To develop and validate the radiomics nomogram combined clinical factors and radiomics to estimate the OS in patients with ccRCC,and assess the incremental value of radiomics for OS estimation.Materials and Methods In part one,127 ccRCC cases graded using Fuhrman grading system from May 2012 to December 2014 were included as a training cohort.62 ccRCC from January 2015 to December 2016 were set as an independent validation cohort.The 3-dimensional regions of interest(ROI)were contoured manually at cortical,nephrographic and excretory phases(CP,NP and EP).4227 radiomics features including intensity,shape and size,texture and wavelet features were extracted from each person.Inter-and intraclass correlation coefficients(ICCs)were used to exclude the inter-observer and intraobserver difference of ROI feature extraction.LASSO-Logistic regression was used to select radiomics features and build three phases radiomics signature.Radiomics score(Radscore)was calculated by the linear combination of selected features.Signal-factor analysis was used for the analysis of significant traditional CT features(tumor location,maximal diameter,irregular shape,necrosis,perinephric space invasion and intratumoral artery)between high grade and low grade ccRCC.The CT-feature model was constructed by multivariate Logistic regression analysis.And the comprehensive model combined CTfeature and Rad-score was constructed by multivariate Logistic regression analysis as well.The performance of the different models in predicting the Fuhrman grade of ccRCC was assessed using ROC curve in the training and validation cohorts,and the different prediction performances were compared with Delong test.The DCA was performed to evaluate the net benefits of the models.In part two,194 ccRCC cases diagnosed from May 2011 to December 2016 were included as a training cohort.188 ccRCC from June 2012 to May 2017 from another hospital were configured as an independent test cohort.After radiomics features extraction and ICC analysis,LASSO-Cox regression algorithm was applied to select optimal features correlated with OS.The clinical factors(age,sex,TNM stage,Fuhrman grade,presence of histologic necrosis,OS,ECOG-PS,hemoglobin,neutrophil count,lymphocyte count,platelet count,blood urea nitrogen,creatinine and neutrophil-lymphocyte ratio)were selected by the univariable and multivariable COX regression analysis.The radiomics nomogram was developed by combing the Rad-score and selected clinical factors.The clinical nomogram was built based on clinical factors to confirm the incremental value of Rad-score for assessment of OS.The prediction performance of the nomograms was evaluated by C-index.The difference was compared by t-test.Finally,risk group which was predicted by the nomogram with better discrimination capability was used as prediction factor to generate the Kaplan-Meier survival curves.The difference between survival curves was assessed by using the log-rank test.Results In part one,after selecting radiomics features with ICCs > 0.75,12 optimal features extracted from three phases after LASSO regression.The CT-feature model combined irregular shape and maximal diameter had a discrimination performance with an AUC of 0.723(95%CI: 0.621-0.826)in predicting the Furman grade of ccRCC in the training cohort,and showed improved diagnosis performance with an AUC of 0.823(95%CI: 0.749-0.908,P < 0.05)after incorporating with Rad-score.Comprehensive model showed better discrimination performance than CT-feature in the test cohorts(0.849,95%CI: 0.753-0.945 vs.0.728,95%CI: 0.587-0.868;P < 0.05).The DCA indicated that the three models had high overall net benefit in predicting the Fuhrman grade of ccRCC before operation.In part two,11 radiomics features were selected significantly associated with OS by LASSO-Cox regression analysis.TNM stage(HR: 2.431,1.709-3.459;P<0.05)and creatinine(HR: 1.019,1.003-1.036;P<0.05)were significantly related to OS by Cox regression analysis.The radiomics nomogram was developed by combing Rad-score and clinical factors.The clinical nomogram was merely built based on clinical factors.The radiomics nomogram presented higher discrimination capability than the clinical nomogram in the training(C-index: 0.884;95% CI: 0.808-0.940 vs.0.803;95% CI: 0.705-0.899,P < 0.05)and test cohorts(C-index: 0.859;95% CI: 0.800-0.921 vs.0.846;95% CI: 0.777-0.915,P < 0.05).The Kaplan-Meier survival curves were generated by using risk group based on survival probability at three or five years predicted by the radiomics nomogram as prediction factor.The significant difference was confirmed between the risk groups in both the training and test cohorts(P < 0.001).Conclusion Enhanced CT-based radiominc have good predictive performance in the prediction of Fuhrman grade and OS in ccRCC,which can improve the ability to predict and provide an new method for precise treatment.