Application Of Radiomics In Preoperative Prediction Of WHO/ISUP Grading Of Renal Clear Cell Carcinoma

Objective: To investigate the value of nomograms constructed based on the radiomics features of enhanced CT combined with clinical features in preoperatively predicting WHO/ISUP grading of renal clear cell carcinoma(ccRCC).Methods: The imaging and clinicopathological data of 186 cases with preoperative enhanced CT examination and pathologically confirmed ccRCC were retrospectively analyzed.The cases were divided into high-grade(Ⅲ+Ⅳ)47 cases and low-grade(Ⅰ+Ⅱ)139 cases according to WHO/ISUP classification,and divided into training and validation groups according to 7:3,including 130 cases in the training group and 56 cases in the validation group.The region of interest(ROI)along the edge of the lesion was manually delineated on the image,and the features of the plain scan,cortical medullary stage and medullary stage were extracted to screen out the best features and construct a radiomics model;relevant CT imaging features and clinical features were incorporated,and the most valuable features were selected after logistic regression to construct the clinical model;The joint model was constructed based on the clinical model and the radiomics model,and the nomogram was established.The diagnostic efficacy of the three models was evaluated by subject operating characteristic curves(ROC),and the corresponding AUC values,accuracy,sensitivity,and specificity were obtained.The clinical application value of the models was evaluated using decision curve analysis(DCA),and the calibration performance of the models was evaluated by calibration curves.Results: 2817 radiomics features were extracted from CT images,27 radiomics features were finally screened out after dimensionality reduction processing,and 1 clinical radiological feature was screened out based on CT image characteristics and clinical information,and the radiomics models,clinical radiology models and clinical-radiomics models were constructed,The joint model had the best prediction performance among the three groups of models,and its AUC values in the training and validation groups were The AUC values were 0.823(95% CI 0.744-0.891)and 0.726(95% CI 0.564-0.875)in the training and validation groups,respectively.The decision curve analysis showed that the clinical application value of the clinical-radiomics model was higher than the other two groups of models,and there was some agreement between the predicted and actual probabilities of the clinical-radiomics model as seen by the calibration curve.Conclusion: Contrast-enhanced CT-based nomogram can non-invasively predict the WHO/ISUP pathological grade of ccRCC before surgery and provide high predictive performance.