Application Of CT Radiomics In Differential Diagnosis And Fuhrman Classification Of Clear Cell Renal Cell Carcinoma

Part Ⅰ Construction of Predictive Model for Differential Diagnosis of ccRCC Based on CT RadiomicsObjective:The radiomics prediction models of renal tumors including clear cell renal cell carcinoma(ccRCC),angiomyolipoma without visible fat(AML.wovf),papillary renal cell carcinoma(pRCC),and chromophobe renal cell carcinoma(chRCC)based on CT radiomics were constructed.On this basis,the combined prediction model of renal tumor combined with clinical data,conventional imaging features,corrected CT enhancement parameters and CT radiomics was constructed.Objective to explore the application value of the two groups of models in the differential diagnosis of ccRCC and AML.wovf,ccRCC and non-ccRCC.Methods:A total of 301 patients with renal tumors confirmed by pathology and with complete clinical data and plain and enhanced CT images were collected.There were 32 cases of AML.wovf,212 cases of ccRCC and 57 cases of non-ccRCC,there were 44 cases of chRCC and 13 cases of pRCC in non-ccRCC.The basic clinical data and conventional imaging features of all patients were retrospectively analysed(including location,shape,longest diameter,plain scan density,calcification,bleeding,liquefaction and necrosis,enhancement mode and growth mode).The CT values of the lesions at different stages,the CT values of the normal renal cortex at the same layer,and the CT values of the abdominal aorta at the opening level of the renal artery of the affected side were measured,and the CT enhancement parameters were calculated and corrected(including progressive enhancement percentage,relative enhancement value,enhancement change value,tumor relative enhancement value,relative aortic enhancement rate,relative renal cortex change rate).At the same time,the radiomics program in the Siemens syngo.via post-processing workstation outlines the whole lesion in the CT images of cortical phase and parenchymal phase respectively,and extracts the radiomics features.Univariate logistic regression was used to compare the differences of basic clinical data,conventional imaging features,corrected CT enhancement parameters and radiomics features among ccRCC group and AML.wovf group,ccRCC group and non-ccRCC group.The radiomics features of P<0.001 were used to establish the radiomics model by LASSO regression.The clinical data,conventional imaging features and corrected CT enhancement parameters of P<0.001 were combined with the radiomics features,and the combined model was established by LASSO regression.Results:1、The gender,age,height,weight and BMI in the basic clinical data were statistically significant(P<0.05)between ccRCC and AML.wovf,and gender and weight were P<0.001.Among the conventional imaging features,the longest diameter,plain scan density,liquefaction necrosis and enhancement mode were statistically significant(P<0.05)and the liquefaction necrosis and enhancement mode were p<0.001.Most of the corrected CT enhancement parameters were statistically significant,and there were 6 parameters P<0.001.2、Finally,15 radiomics features were selected to construct the prediction model.The AUC values of training set and testing set of the established radiomics model were about 0.975(95%CI,0.951～0.998)and 0.906(95%CI,0.819～0.992)respectively.The AUC values of the training set and the testing set of combined model were about 0.995(95%CI,0.988～1)and 0.982(95%CI,0.953～1)respectively.3、There was no significant difference in all basic clinical data between ccRCC and non-ccRCC.The conventional imaging features of liquefaction necrosis and enhancement were statistically significant(P<0.001).Among the corrected CT enhancement parameters,most of them were statistically significant,and 10 of them were P<0.001.4、Finally,13 radiomics features were selected to construct the prediction model.The AUC values of training set and testing set of radiomics model were about 0.918(95%CI,0.872～0.963)and 0.884(95%CI,0.818～0.949)respectively.The AUC values of the training set and the testing set of combined model were about 0.964(95%CI 0.938～0.990)and 0.962(95%CI,0.927～0.997)respectively.Conclusion:1、The radiomics prediction model of renal tumor includeing ccRCC,AML.wovfc,chRCC and pRCC based on CT radiomics has high application value in the differential diagnosis of ccRCC and AML.wovf,ccRCC and non-ccRCC.2、The combined prediction model of renal tumor which combined with basic clinical data,conventional imaging features,correction of CT enhancement parameters and radiomics can effectively improve its diagnostic efficiency.Part Ⅱ Construction of Predictive Model for Fuhrman Grade of ccRCC Based on CT RadiomicsObjective:The prediction model of Fuhrman classification of clear cell renal cell carcinoma(ccRCC)based on CT radiomics was constructed.On this basis,the combined prediction model of ccRCC combined with clinical data,conventional imaging features,corrected CT enhancement parameters and CT radiomics was established.Objective to explore the application value of the two models in the Fuhrman classification of ccRCC.Methods:A total of 194 patients with ccRCC confirmed by pathology and with complete clinical data and plain and enhanced CT images were collected.There were 70 cases of high Fuhrman grade,124 cases of low Fuhrman grade.The basic clinical data and conventional imaging features of all patients were retrospectively analysed(including location,shape,longest diameter,plain scan density,calcification,bleeding,liquefaction and necrosis,enhancement mode and growth mode).The CT values of the lesions at different stages,the CT values of the normal renal cortex at the same layer,and the CT values of the abdominal aorta at the opening level of the renal artery of the affected side were measured,and the CT enhancement parameters were calculated and corrected(including progressive enhancement percentage,relative enhancement value,enhancement change value,tumor relative enhancement value,relative aortic enhancement rate,relative renal cortex change rate).At the same time,the radiomics program in the Siemens syngo.via post-processing workstation outlines the whole lesion in the CT images of cortical phase and parenchymal phase respectively,and extracts the radiomics features.Univariate logistic regression was used to compare the differences of basic clinical data,conventional imaging features,corrected CT enhancement parameters and radiomics features among ccRCC group and AML.wovf group,ccRCC group and non-ccRCC group.The radiomics features of P<0.05 were used to establish the radiomics model by random forest.The clinical data,conventional imaging features and corrected CT enhancement parameters of P<0.05 were combined with the radiomics features,and the combined model was established by random forest.Results:1、All the basic clinical data were not statistically significant between high Fuhrman grade and low Fuhrman grade.In conventional imaging features and corrected CT enhancement parameters,only the longest diameter,liquefaction necrosis,capsule condition,RE1 and TEV3 were statistically significant(P<0.05),while the rest were not statistically significant.2、Finally,8 radiomics features were selected to construct the prediction model.The AUC values of training set and testing set of RF radiomics model are about 0.828(95%CI,0.761～0.896)and 0.720(95%CI,0.530～0.909),respectively.The AUC values of training set and testing set of RF combined model are about 0.835(95%CI,0.769～0.899)and 0.740(95%CI,0.558～0.993),respectively.Conclusion:1、The radiomics prediction model of ccRCC based on CT radiomics has a certain value in the Fuhrman classification of ccRCC.2、The combined prediction model of ccRCC which was further combined with clinical data,conventional imaging features,corrected CT enhancement parameters and CT radiomics did not significantly improve the diagnostic efficiency.