Prediction Study Of Surrounding Tissue Invasion In Clear Cell Renal Cell Carcinoma Based On Multi-phase Enhanced CT Radiomics

Objective To construct a combined model of clinical factor model and imaging feature model,and establish a fusion radiomics nomogragh with the three-phase enhanced CT radiomics model to explore the value of the fusion model in predicting the surrounding tissue invasion(STI)in clear cell renal cell carcinoma(ccRCC)before surgery.Methods Based on the inclusion and exclusion criteria,the clinical and imaging data of 65 patients with ccRCC of STI and 213 patients without surrounding tissue negative(STN)from medical center A and B were analyzed retrospectively.The data from medical center A were randomly segmented into the training cohort and internal validation cohort at a ratio of 7:3,and the data from medical Center B was the external validation cohort.All data were confirmed by postoperative pathology.Univariate and multivariate Logistic regression analysis were used to screen and evaluate the independent risk factors for STI in ccRCC,so as to construct the clinical imaging feature model.After the ROI was delineated along the tumor contour on CT scan,corticomedullary phase and parenchymal phase images,the tumor ROI was automatically expanded by 2mm,4mm,5mm,6mm and 2mm and shrinked by 2mm respectively.The image features of each ROI then were extracted and screened,and the radiomics model was constructed.Integrating clinical imaging feature model and the optimal radiomics model,a fused radiomics nomogram was developed.The area under curve was used to evaluate the predicted efficiency of each model for the presence of STI in ccRCC.Delong test was used to compare the difference of AUC values between the three models.Decision curve analysis and decision curve were used to evaluate the calibration degree and clinical value of the fusion model.Shapley Additive explanations(SHAP values)were used to explain the contribution of the image features to the models.Results In Medical Center A and Medical Center B,the number of STI and STN cases were 35 and 30,112 and 101,respectively(χ2=0.032,P=0.858).Arterial hypertension,Lumbalgia among the clinical risk factors and calcification,Intratumoral vasculature among the CT image features are independent risk factors(OR value is 2.994～4.613).The AUC of the clinical-image feature model,established on these independent risk factors in the training cohort,the internal validation cohort and the external validation cohort were 0.766,0.765 and 0.698,respectively.After gradient regression,LASSO dimensionality reduction and establishment of corresponding models for CT radiomics features,the CT radiomics model established based on the 4mm peritumoral ROI in the nephrographic phase has the highest diagnostic value,and the AUC of the optimal radiomics model in the training cohort,internal validation cohort,and external validation cohort were 0.837,0.831,and 0.762,respectively.Combining this model with the clinical image feature model to construct a fusion model,its AUC in the training cohort,internal validation cohort,and external validation cohort were 0.890,0.886,and 0.826,respectively.In the external validation group,Delong test results showed that the diagnostic efficiency of the fusion model was significantly higher than that of the clinical imaging feature model or the imaging group model(P<0.05);the calibration curve shows that the prediction results of the fusion model are consistent with the actual situation;DCA shows that the fusion model exhibits the highest net benefit.SHAP values analysis showed that the top three radiomics features were Small Dependence Low Gray Level Emphasis,Maximum 3D Diameter,Maximum Probability。Conclusion The combined model of clinical risk factors and CT imaging features,and a fusion model constructed with a CT radiomics model based on the peritumor ROI of 4 mm in the nephrographic phase,has the highest diagnostic performance in predicting the invasion of surrounding tissues of ccRCC,which provides an important basis for clinicians to formulate individualized treatment plan.Using the SHAP values,the role weight of individual feature in the CT radiomics model could be better interpreted objectively.