Predicting Preoperative Renal Clear Cell Carcinoma WHO/ISUP Classification Based On Enhanced CT Imaging Histological Features And Extracellular Volume Fraction

ObjectiveTo investigate the optimal classification of Clear cell renal cell carcinoma(cc RCC)based on multi-phase enhanced CT imaging histology and extracellular volume fraction(ECV)for preoperative prediction of WHO/ISUP pathological grading.Materials and Methods1.Clinical baseline informationThis study retrospectively analyzed 124 patients with renal clear cell carcinoma who underwent surgery for renal tumors between January 1,2018 and December 31,2021 at the Second Hospital of Dalian and were finally confirmed by pathology and immunohistochemistry,including 63 cases of high-grade and 61 cases of low-grade.Clinical data and subjective imaging features(including location,morphology,longest diameter,calcification,hemorrhage,cystic necrosis,enhancement pattern,growth pattern,envelope,etc.)were collected from all patients.The CT values of the lesions in the plain and excretory phases and the CT values of the abdominal aorta at the level of the affected renal artery opening during the same period were measured in all patients,and the extracellular volume fraction(ECV)of the tumor was calculated for each patient.3.Model constructionThe CT tumor images of each patient were manually segmented layer by layer using3 D Slicer software,including the scanned,cortical and parenchymal images of each patient.The region of interest(ROI)was outlined and the CT imaging histological features of each patient were extracted at the same time.The extracted image histology features were then uploaded to the imaging histology cloud platform,randomly divided into training set(n=86)and validation set(n=38)in the ratio of 7:3,and the most relevant image histology features were selected using the minimum redundancy maximum correlation(m RMR)feature filtering method,and a support vector machine classifier was used to construct single-phase(unenhanced,cortical and parenchymal phases)and threephase images respectively.The support vector machine classifier was used to construct single-phase(plain phase,cortical phase and parenchymal phase)and three-phase combined imaging histology models.Finally,one-way logistic regression analysis was applied to compare the differences between the clinical data,subjective imaging features,ECV and combined imaging histological features between the two groups,and the clinical data,subjective imaging features and ECV combined with imaging histological features with P<0.001 were used to construct the clinical radiological model,ECV model and combined prediction model,respectively,using the support vector machine classifier.The models were evaluated by subject operating characteristic curve(ROC)and parameters such as sensitivity and specificity.Analysis was performed to evaluate the efficacy of each model in predicting WHO/ISUP classification in the validation set and validation set.Result1.Clinical baseline informationFor the analysis of all basic clinical data of patients,there was no statistical significance between the cc RCC WHO/ISUP high-grade and low-grade groups(P > 0.05);there was a statistically significant relationship with red blood cell pressure(HCT)and surgical modality(all P less than 0.001).Among the subjective imaging features,the maximum tumor diameter,necrosis,intratumoral artery,choroidal carcinoma thrombus,perinephric invasion,and envelope were statistically significant(all P less than 0.001),and the rest were not statistically significant(P > 0.05).2.ECV model prediction model performance evaluationThe ECV of renal clear cell carcinoma in the low-grade group was lower than that in the high-grade group,and the difference was statistically significant(0.43(0.40,0.45)in the low-grade group and 0.60(0.58,0.61)in the high-grade group,P < 0.001).The AUC of the training set was about 0.907 and the sensitivity,specificity and F1 scores were 0.952,0.727 and 0.851,respectively;the AUC of the validation set was about 0.868,and the sensitivity,specificity and F1 scores were 0.889,0.750 and 0.821,respectively.Prediction performance and comparison among modelsThe AUC of the training set of the three-stage combined model was about 0.950,sensitivity,specificity and F1 score of 0.929,0.818 and 0.876,respectively;the AUC of the validation set was about 0.910,sensitivity,specificity and F1 score of 0.889,0.750 and 0.821,respectively.the AUC of the training set of clinical radiology model was 0.917,sensitivity,specificity and F1 score of 0.810,0.957 and 0.883,respectively;the AUC of the validation set was 0.909,sensitivity,specificity and F1 score of 0.889,0.700 and 0.782,respectively.the AUC of the training set of the integrated model is about 0.971,sensitivity,specificity and F1 score 0.952,0.887 and 0.920 respectively;the AUC of the validation set was about 0.950,sensitivity,specificity and F1 score of 1.000,0.850 and 0.878,respectively.Conclusion:The SVM imaging histological model constructed based on the CT single-phase or multi-phase enhanced imaging histological features can predict cc RCC WHO/ISUP pathological grading more accurately,with the highest accuracy of the combined threephase model.In addition,the combined model constructed by adding some statistically significant basic clinical data,conventional imaging features and ECV could improve the diagnostic efficacy of the prediction model to some extent.