Clinical And CT Features Of Renal Urothelial Carcinoma Mimicking Renal Cell Carcinoma And The Renal Clear Cell Carcinoma

Objects ⑴ To retrospectively assess the clinical and computed tomography(CT)features of renal urothelial carcinoma(RUC)mimicking renal cell carcinoma(RCC),and to correlate clinic-CT features with the pure infiltrative renal urothelial carcinoma(PIRUC)and infiltrative renal urothelial carcinoma histological variants(IRUCHV)in the RUCs.⑵To assess retrospectively clinical data and CT features of renal urothelial carcinoma(RUC)mimicking RCC and renal clear cell carcinoma(RCCC)with collecting system invasion(CSI).And to explore the association of clinical-CT features with RUC or RCCC.⑶To compare the ability of clinic-CT model and radiomics model for distinguishing RUC mimicking RCC from RCCC with CSI.Methods ⑴ Twenty-five patients with PIRUC and 18 patient with IRUCHV were retrospectively assessed.Logistic regression analysis was used to screen the independent predictor of PIRUC and PIUCV differential diagnosis.The diagnostic performance of independent predictor was evaluated by AUC,sensitivity and specificity.⑵Data from 46 patients with RUC and 289 patients with RCCC were retrospectively assessed.CT and clinical features were assessed by logistic regression analysis for screening independent predictors.Then independent predictors were used to construct the prediction model.ROC curves were drawn based on the independent predictors and prediction model,for evaluating the diagnosis performance.Statistical significance was set at P<0.05.⑶ Forty-four patients with RUC and 132 patients with RCCC were involved.Logistic regression analysis was used to screen out independent predictors of clinical-CT features.Clinc-CT model and radiomics models of multiphase CT were constructed respectively with the method of radiomics.The predictive performance was evaluated by AUC,sensitivity,and specificity of ROC curve.Results ⑴ The independent predictor was perinephric stranding.⑵Logistic regression analysis identified infiltrative growth pattern,diffused infiltrative growth pattern,hydronephrosis,heterogeneous enhancement,and hematuria as independent predictors that can be used to distinguish RUC from RCCC.The AUC of model-1 which differentiated RUC from RCCC and improved the diagnostic performance of predictors.⑶Performances in training-cohort and validation: Clinical-CT model> corticomedullary phase(CP)> nephrographic phase(NP)> excretory phase(EP)> unenhanced phase(UP).The boxplot of the average CT values of RUC and RCCC of multiphase CT showed the greatest difference in CP.Decision curve analysis showed that the clinical-CT model had a higher overall net benefit in distinguishing RUC mimicking RCC from RCCC with CSI.Conclusions ⑴The independent predictor may provide useful help in differentiating PIRUC from IRUCHV,which may provide useful information for clinical treatment decision-making.⑵ Independent predictors and predictive model may play an important role in preoperative differentiation between RUC mimics RCC and RCCC with CSI using clinic and CT features.⑶CP-radiomics model had the greatest value in the diagnosis of RUC or RCCC among radiomics models of multiphase CT.The performance of clinic-CT model was better than that of the radiomics model in distinguishing RUC from ERCCC.Part I Clinic and CT features of Pure Infiltrative Carcinoma vs Infiltrative Carcinoma Histological Variant in Renal Urothelial Carcinoma mimicking Renal Cell CarcinomaObjects To retrospectively assess the clinical and CT features of RUC mimicking RCC,and to correlate clinic-CT features with the PIRUC and IRUCHV in the RUCs.Methods Patients from Aug 2008 to Dec 2020 were included.Twentyfive patients(18 men and 7 women)with PIRUC and 18 patients(15 men and 3 women)with IRUCHV were retrospectively assessed.The average patient age was 67.00(63.00-69.00)and 67.50(62.25-72.75)years,respectively.The clinical data and CT characteristics of PIRUC and IRUCHV were analyzed by Mann-Whitney U test /T test(continuous variables)and χ~2/ Fisher’s exact test(categorical variables).Univariate logistic regression analysis was performed for statistically significant variables of the Mann-Whitney U test /T test and χ~2/ Fisher’s exact test to assess the association strength between clinical-CT features and PIRUC or IRUCHV.Next,the results of descriptive statistical variables with significant differences were made ROC curves respectively,and the variables with AUC <0.7 were excluded.Then,the variables of univariate logistic regression analysis with no significant differences were excluded from the multivariates regression analysis.The remaining variables were involved into multivariate logistic regression analysis,and the statistically significant variable was independent predictor of PIRUC and PIUCV differential diagnosis.The diagnostic performance of independent predictor was evaluated by AUC,sensitivity and specificity.Results Continuous variables in Mann-whitney U test /T test showed no significant difference.The χ~2/ Fisher’s exact test results of the categorical variables(renal stone history,renal sinus invasion,perinephric stranding,vessel in tumor,renal vein invasion and lymphatic metastasis)were significantly different.Vessel in tumor was excluded from multivariate logistic regression,due to P>0.05 in univariate logistic regression and ROC curve.Renal vein invasion and lymphatic metastasis were excluded,due to P>0.05 of ROC curve.History of kidney stone was excluded,due to AUC<0.7.Finally,perinephric stranding [P=0.001,(odds ratio,OR): 0.023] and renal sinus invasion(P=0.013,OR: 0.064)involved the multivariate regression analysis.The independent predictor was perinephric stranding(P=0.008,OR: 0.033).The performance of perinephric stranding(AUC=0.832,sensitivity: 94.44,specificity: 72)was shown in the ROC curve.Conclusions The independent predictor may provide useful help in differentiating PIRUC from IRUCHV,which may provide useful information for clinical treatment decision-making.Part Ⅱ Distinguishing Renal Urothelial Carcinoma Mimicking RCC from Renal Clear Cell Carcinoma with Clincal and CT FeaturesObjects To assess retrospectively clinical data and CT features of RUC mimicking RCC and RCCC with CSI.And to explore the association of clinical-CT features with RUC or RCCC.Methods: Data from 46 patients with RUC and 289 patients with RCCC were retrospectively assessed(from Aug 2008 to Dec 2020).Among them,there are 36 men and 10 women with RUC,208 men and 81 women with RCCC.The average age of RUC and RCCC was 67.00(62.25-70.25)and 53.00(45.00-61.00)years,respectively.CT and clinical features exhibiting significant differences in Mann-Whitney U test /T test and χ~2/ Fisher’s exact test were analysed using univariate logistic regression.Then,partial variables with P>0.05 or OR values being close to 1 in univariate logistic regression were excluded from multivariate logistic regression analysis,for screening independent predictors and avoiding the overfitting of the predictive model’s ROC curve.The results of descriptive statistical variables with significant differences were made ROC curves respectively,and the variables with P>0.05 or AUC<0.7 were excluded.Univariate logistic regressions analysis was used to analyze the associations of CT and clinical features with RUC or RCCC.Statistically significant variables(independent predictors)in the multivariate logistic regression analysis were used to construct the prediction model.Ultimately,ROC curves were drawn based on the independent predictors and prediction model,for evaluating the diagnosis performance of CT and clinical features with RUC or RCCC.Statistical significance was set at P<0.05.Results: Firstly,to avoid over-fitting ROC curve of the predictive model,CP,NP,EP,and age were excluded from multivariate logistic regression analysis,owing to their OR values being close to 1.Secondly,due to no significant difference in ROC curve analysis,distant metastasis was not included in the multivariate logistic regression analysis.Thirdly,perinephric stranding,flank pain,history of kidney,tumor shape,lymphatic metastasis,renal vein invasion,calculus,renal atrophy,and vessel in tumor were excluded from multivariate logistic regression analysis,with the AUC <0.7(low accuracy).Ultimately,infiltrative growth pattern,hydronephrosis,heterogeneous enhancement,pseudo-capsule sign,reniform contour deformation,location,and hematuria were included in multivariate logistic regression analysis.Multivariate analysis identified infiltrative growth pattern [bean shape infiltrative growth pattern(P=0.014,OR: 6.386),diffused infiltrative growth pattern(P=0.858,OR: 1.142))],hydronephrosis(P<0.001,OR: 14.323),heterogeneous enhancement(P<0.001,OR: 0.058),and hematuria [gross hematuria(P<0.001,OR: 13.868);microscopic hematuria(P<0.001,OR: 10.253)] as independent predictors that can be used to distinguish RUC from RCCC.The AUC of model-1 was 0.954(sensitivity: 88.2,specificity: 91.3),which differentiated RUC from RCCC and improved the diagnostic performance of infiltrative growth pattern(AUC: 0.83,sensitivity: 72.3,specificity: 91.3),hydronephrosis(AUC: 0.746,sensitivity: 88.24,specificity: 60.87),heterogeneous enhancement(AUC: 0.745,sensitivity: 83.74,specificity: 65.22),and hematuria(AUC: 0.728,sensitivity: 80.97,specificity: 60.7).Conclusions: Independent predictors and predictive model may play an important role in preoperative differentiation between RUC mimics RCC and RCCC with CSI using clinic and CT features.Part III Value of Radiomics in distinguishing Renal Urothelial Carcinoma Mimicking Renal Cell Carcinoma from Renal Clear Cell CarcinomaObjects: To compare the ability of clinic-CT model and radiomics model for distinguishing RUC mimicking RCC from RCCC with CSI.Methods: In the retrospective study,44 patients with RUC(from Aug 2008 to Dec 2020)and 132 patients with RCCC(from Jan 2016 to Dec 2020)who underwent preoperative multiphase CT and subsequent surgery/biopsy were involved.Clinical data and CT semantic features were assessed,and multiphase CT radiomics features were extracted respectively.First,logistic regression analysis was used to screen out independent predictors of clinical-CT features.Second,clinc-CT model was constructed with radiomics five fold cross-validation.And radiomics models of multiphase CT were constructed respectively with least absolute shrinkage and selection operator algorithm(LASSO)and logistic regression classifier.The predictive performance was evaluated by AUC,sensitivity,and specificity of ROC curve.Results: Firstly,age,CP,NP,and EP were excluded from multivariate logistic regression analysis,owing to their OR values being close to 1(that reflect a weak strength association).Secondly,due to no significant difference in ROC curve analysis or AUC<0.7,variables were not included in the multivariate logistic regression analysis.Then,the residual variables(heterogeneous enhancement,hematuria,infiltrative growth pattern,and hydronephrosis)were included in multivariate regression analysis Infiltrative growth pattern,hydronephrosis,heterogeneous enhancement,and hematuria were the independent predictors.Ultimately,clinic-CT model was constructed with the predictors,and the radiomics models were constructed by the 22 best radiomic features.Performances in training-cohort: clinical-CT model(AUC: 0.924,sensitivity: 85.48,specificity: 82.14);UP-Radiomics model(AUC: 0.75,sensitivity: 74.05,specificity: 65.71);CP-Radiomics model(AUC: 0.89,sensitivity: 79.8,specificity: 91.4);NP-Radiomics model(AUC: 0.853,sensitivity: 77.14,specificity: 75.71);EP-Radiomics model(AUC: 0.823,sensitivity: 74.07,specificity: 77.78).Performances in validation: clinical-CT model(AUC: 0.914,sensitivity: 77.78,specificity: 88.89);UP-Radiomics model(AUC: 0.658,sensitivity: 62.96,specificity: 77.78);CP-Radiomics model(AUC: 0.815,sensitivity: 77.8,specificity: 77.8);NP-Radiomics model(AUC: 0.798,sensitivity: 88.89,specificity: 77.78);EP-Radiomics model(AUC: 0.772,sensitivity: 66.43,specificity: 75.71).The boxplot of the average CT values of RUC and RCCC of multiphase CT showed the greatest difference in CP.Decision curve analysis showed that the clinical-CT model had a higher overall net benefit in distinguishing RUC mimicking RCC from RCCC with CSI.Conclusion: CP-radiomics model had the greatest value in the diagnosis of RUC or RCCC among radiomics models of multiphase CT.The performance of clinic-CT model was better than that of the radiomics model in distinguishing RUC from ERCCC.