CT-based Radiomics And Traditional Imaging Features For Classification Of Renal Tumors:A Pilot Study

Part 1 Exploring the diagnostic value of CT multi-phase improvedenhanced parameters in the diagnosis of benign and malignant renaltumorsObjective:To explore the diagnostic value of differentiating benign and malignant renal tumors based on the CT multi-phase improved parameters of unenhanced and three-phase enhanced CT images.Materials and Methods:A total of 226 pathologically proven renal malignant tumors(143 clear cell RCCs,25 papillaxy RCCs,and 29 chromophobe RCCs)and renal benign tu,mors(10 oncocytomas,19 lipid-poor AMLs)from January 2013 to October 2018 treated in Nanfang hospital was included in our study.Univariate analysis was used to compare the differences in traditional CT imaging features between renal malignant tumors and benign renal tumors group,oncocytoma group and lipid-poor angiomyolipoma group.The quantitative features with P<0.05 in the above three groups of univariate analysis were included in multiple logistic regression analysis,and the independent influencing factors of imaging diagnosis of renal malignant tumor were obtained.For the variables obtained in the multivariate analysis in each comparison group,the regression coefficients of the varnables,the OR values,the P values,and the AUC values,sensitivity,specificity,and accuracy of the diagnosis of renal malignant tumors were calculated.Results:The relative enhancement in unenhanced phase of renal malignancy was lower than that of benign renal tumor and LP-AML,which was an independent factor for the diagnosis of patients(P<0.001).The diagnostic efficacy AUC values were 0.727 and 0.742,respectively.The enhancement change in corticomedullary phase(P =0.006),percentage enhancement(p=0.015),relative aortic corrected CT value(P =0.028),relative kidney cortical correction CT value(P =0.025)in nephrographic phase and relative aortic CT value in excretory phase(P =0.012)were lower than that of renal oncocytoma,which were independent factors for the diagnosis of patients.Among them,relative aortic CT value n excretory phase had the highest diagnostic value,The AUC value is 0.844.With a critical value of 83.0142,the sensitivity and specificity are 84.26%and 80%,respectively.The multi-factor combined diagnosis to distinguish renal malignant tumor from renal oncocytoma has a diagnostic AUC value of 0.990.Conclusion:The improved enhanced parameters based on CT multi-phase have a good diagnostic value for identifying benign and malignant tumors of the kidney.Part 2 Exploring the diagnostic value of CT multi-phaseRadiomics models and traditional imaging models in the diagnosis ofclear cell renal cell carcinomaObjective:To explore the diagnostic value of differentiating clear cell renal cell carcinomas based on the CT multi-phase Radiomics features and traditional imaging features.Materials and Methods:A total of 197 pathologically proven renal masses(143 clear cell RCCs,25 papillary RCCs and 29 chromophobe RCCs)from January 2013 to October 2018 treated in Nanfang hospital was included in our study.The above patients with renal tumors were divided into clear cell and unclear cell renal cell carcinomas groups.First,a traditional imaging feature model was constructed.The CT values of the lesion and the corresponding attenuation of the aorta and adjacent normal renal cortex were measured by two radiologists,and the improved enhanced parameters were obtained.According to the univariate analysis,the three most discriminative features identified with each phase were entered as predictors in logistic regression models to construct a traditional diagnosis model.Secondly,a diagnostic model of Radiomics features was constructed.Two radiologists manually segmented the 3D tumor images and extracted the image radiomics features.LASSO regression screening was used to construct the Rad-scores based on the unenhanced and three enhanced phases.Finally,all the above parameters with statistical difference in single factor comparison and rad-scores were included in the multivariate logistic regression analysis to build the combined diagnosis prediction model.To determine all of the above model diagnostic performance,the AUC,sensitivity,specific,positive predictive value,negative predictive value and accuracy were concluded.Delong test was used to compare the diagnostic performance of each diagnostic model.Results:Differences in Rad-scores and enhancement changes of aortic version in the corticomedullary phase,nephrographic phase,and excretory phase were independent influencing factors in the differential diagnosis of clear cell renal cell carcinoma and unclear cell renal cell carcinoma(P<0.001).AUC values of the traditional diagnostic model,Rad-score model and the combined diagnostic model in differentiating the two groups were 0.748-0.823,0.776-0.887 and 0.864-0.9,respectively.The AUC values of the three models were statistically significant(P<0.001).Conclusions:By combining traditional CT improved parameters and Radiomics features to construct a combined diagnostic model,the predictive efficacy of clear cell renal cell carcinomas was significantly improved,which provided a reliable basis for clinical treatment options.