Prediction Of Pathological Grade Of Renal Clear Cell Carcinoma Based On Radiomics Features Of Energy Spectrum CT Multi-phase Enhanced Iodine Water Map

Objective:To investigate the value of preoperative prediction of pathological grade of clear cell renal cell carcinoma based on energy spectrum CT iodine(water)image,and to evaluate the predictive efficacy of the model.Data and methods:A total of 102 patients who underwent single-source dual-energy CT dynamic enhancement scans of kidneys in our hospital and were confirmed to be ccRCC by postoperative pathology were collected from January 2016 to September 2019.According to the pathological classification criteria,69 cases were divided into low-grade group(Fuhrman I-II)and 33 cases in high-grade group(Fuhrman III-IV).The 102 samples were randomly divided into training set and test set at a ratio of 7:3,of which 71 were in the training set(including 48 in the low-level group and 23 in the high-level group),and 31 in the test set(including 21 in the low-level group,10 cases in the high-level group).All patients underwent a dynamic enhanced CT scan of the kidney energy spectrum before operation,and post-processing was used to generate iodine(water)images in the cortical and medullary phases.Manually draw the outline of the entire lesion layer by layer on all continuous levels of the tumor and save the volumes of interest.The image was resampled and intensity normalized,and 402 radiological features were extracted from the cortical and medullary phases respectively.Spearman rank correlation test and Least absolute shrinkage and selection operator algorithm are used to reduce the dimensionality of features.Five-fold cross-validation was performed on the training set,and three multivariate Logistic imaging omics models were constructed: cortical phase,medullary phase,and the combined model of the cortical and medullary phase Logistic model.Choose the optimal imaging omics Logistic model and calculate the new variable radiomics score.Spearman rank correlation test was used to conduct single factor analysis of clinical related characteristics,and multivariate Logistic regression analysis was used to construct a clinical scoring model.Combine the clinical characteristics and the Radscore of the imaging omics model,apply multiple Logistic regression analysis to construct a clinical-imaging combined Logistic model,and draw a nomogram.The predictive performance of the radioomics model,the clinical scoring model and the clinical-image combined Logistic model are evaluated through the receiver operating characteristic curve,so as to obtain the area under the receiver operating characteristic curve,accuracy,sensitivity and specificity.Based on the Delong test,compare whether there is a difference between the AUC values of the three models.Based on the decision curve analysis to evaluate the clinical application value of the model,the calibration curve and Hosmer–Lemeshow test were used to evaluate the calibration performance of the model.Results:The Logistic model of radiomics based on the cortical and medullary phases showed the best predictive performance.The AUC and 95% confidence interval,accuracy,sensitivity,and specificity of the test set were 0.91(0.81,0.984)and 0.667,52.4%,100%.A clinical scoring model was established based on the 4 clinical independent risk factors obtained after screening.The AUC and 95% confidence interval,accuracy,sensitivity,and specificity of the test set were 0.676(0.488,0.843),0.516,42.9%,and 70.0%,respectively.Based on the 4 clinical risk factors and the optimal cortex-medullary imaging omics model Radscore,the clinical-imaging combined Logistic model and the corresponding nomogram were constructed,and the AUC of the test set and 95% confidence interval,accuracy,sensitivity,The specificity is 0.924(0.833,0.986),0.806,76.2%,and 90.0% respectively.ROC curve analysis shows that the predictive power of the clinical-image combined Logistic model is significantly higher than that of the clinical scoring model,and slightly higher than the imageomics Logistic model.The decision curve analysis shows that the clinical application value of the clinical-image combined Logistic model is higher than that of the radiomics model and clinical scoring model.The H-L test results showed that the P values of the clinical scoring model,radiomics model and clinical-imaging combined model were all greater than 0.05,and the goodness of fit was good.The calibration curve shows that there is good agreement between the predicted probability of the clinical-image combined Logistic model and the actual ccRCC pathological classification.Conclusion:The clinical-imaging combined with Logistic model nomogram based on preoperative energy spectrum CT enhanced scan cortical medullary phase iodine(water)map has certain clinical application value in predicting the pathological classification of ccRCC.