Objective To evaluate the clinical value of computed tomography (CT)-enhanced and subtraction radiomic models in distinguishing clear cell renal cell carcinoma (ccRCC) from non-clear cell renal cell carcinoma (non-ccRCC).

Methods A retrospective analysis was conducted on clinical and imaging data of 458 patients with confirmed ccRCC and non-ccRCC. After excluding cases with poor image quality, 219 cases from Yijishan Hospital of Wannan Medical College (154 cases in the training set and 65 cases in the test set) and 41 cases from The Second People's Hospital of Wuhu (external validation set) were selected for analysis. Images were then exported, aligned, and subtracted and the regions of interest of the tumors were delineated by using ITK-SNAP software. FAE software was employed to extract the radiomic features of the tumor regions. Dimensionality reduction and feature selection were performed using Pearson's correlation coefficients and Relief methods, followed by Logistic regression to construct the radiomics model. Model performance was assessed using receiver operating characteristic (ROC) curve and area under the curve (AUC) analysis. Clinical-imaging features were incorporated in a combined model and visualized as a nomogram.

Results Regression analysis identified cystic or necrotic areas (odds ratio OR=3.282, 95% confidence interval CI:1.111–9.693, P=0.032) and tumor enhancement value of the corticomedullary phase (OR=1.058, 95%CI:1.024–1.093, P=0.001) as clinically independent predictors. In the training set, the radiomics model of the corticomedullary phase demonstrated the highest diagnostic efficacy (AUC=0.880) in the unenhanced and three-phase enhanced radiomics models, diagnostic efficacy was improved by subtraction (AUC=0.949). Additionally, the efficacy of both conventional and subtraction radiomics model improved with the inclusion of clinically independent predictors (AUC=0.903 and 0.973, respectively). These findings were validated in the internal and external validation sets (AUC of conventional radiomics=0.872 and 0.898, respectively; and AUC of subtraction radiomics model=0.908 and 0.920, respectively).

Conclusions The CT-enhanced subtraction radiomics model, integrating subtraction radiomics features of corticomedullary phase with clinical-imaging features, can effectively distinguish ccRCC and non-ccRCC.