| [Objectives]To foretell the grading and staging of clear cell renal cell carcinoma(ccRCC)preoperatively by using CT-based radiomic features.[Methods]This retrospective cohort study included 96 patients with pathologically confirmed clear cell renal cell carcinoma(ccRCC)from the Cancer Imaging Archive.1TK-SNAP software(v3.6.0)was used to delineate the region of interest(ROI)on the largest cross-section of tumor from the acquired DICOM-formatted nephrographic phase CT images,and then all data was uploaded to the Radcloud platform(Huiying Medical Technology Co.,Ltd)for further analysis.A total of 1409 radiomic features were extracted.96 patients were randomly divided into training set(n=76)and validation set(n=20)in a ratio of 4:1 labeled with Fuhrman grading(Grade 1～2/Grade3～4),T stage(T1～2/T3～4),N stage(N0/N1),M stage(M0/M1).respectively.Then variance threshold,select K best and least absolute shrinkage and selection operator(LASSO)algorithm methods were used to obtain the optimal features step by step.The selected features were constructed with logistic regression(LR)and random forest(RF)models to discriminate Fuhrman grading and the T,N and M staging of ccRCC.respectively.The effectiveness of the prediction model for the grading and staging of ccRCC was evaluated by the receiver operator characteristic(ROC)curve and parameters such as sensitivity and specificity.[Results]9 radiomic features were used to build the Fuhrman grading model.The area under the ROC curve(AUC value)of LR model predicting G3～4 ccRCC of training set was 0.889(95%confidence interval[CI]:0.79～0.99),and sensitivity and specificity were 0.80 and 0.78;the AUC value of validation set was 0.648(95%CI:0.40～0.90).and sensitivity and specificity were 0.69 and 0.43.The AUC value of RF model predicting G3～4 ccRCC of training set was 1.000(95%CI:0.98～1.00),and sensitivity and specificity were 0.98 and 1.00;the AUC value of validation set was 0.780(95%CI:0.57～0.99),and sensitivity and specificity were 0.62 and 0.71.For T staging model,16 optimal radiomic features were found.The AUC value of LR model predicting T3～4 ccRCC of training set was 0.941(95%CI:0.86～1.00),and sensitivity and specificity were 0.86 and 0.87;the AUC value of validation set was 0.823(95%CI:0.65～1.00),and sensitivity and specificity were 0.88 and 0.75.The AUC value of RF model predicting T3～4 ccRCC of training set was 0.998(95%CI:0.96～1.00),and sensitivity and specificity were 0.97 and 0.98;the AUC value of validation set was 0.865(95%CI:0.69～1.00),and sensitivity and specificity were 0.88 and 0.83.The number of cases of which regional lymph nodes could not be evaluated(stage NX)were too many to analysis.1 radiomic feature were selected to construct M staging model.The AUC value of LR model predicting M1 ccRCC of training set was 0.870(95%Cl:0.73～1.00),and sensitivity and specificity were 0.80 and 0.83;the AUC value of validation set was 0.941(95%CI:0.83～1.00),and sensitivity and specificity were 1.00 and 0.71.The AUC value of RF model predicting M1 ccRCC of training set was 0.997(95%C1:0.98-1.00),and sensitivity and specificity were 1.00 and 0.97;the AUC value of validation set was 0.951(95%CI:0.87～1.00),and sensitivity and specificity were 1.00 and 0.82.[Conclusions]Radiomic features could be used as biomarker for the preoperative evaluation of the ccRCC Fuhrman grading and T and M staging,which is helpful for clinical treatment decision making.The RF model had better prediction efficiency than LR model.The RF model is best in predicting M staging,followed by T staging,and the least effective in Fuhrman grading.Nevertheless,whether radiomic features can assess N staging of ccRCC still needs further study. |
PDF Full Text Request