A CT−Based Radiomics Ensemble Model (CRIPEM) for Preoperative Prediction of Pathological Upstaging in Clear Cell Renal Cell Carcinoma

Background: Pathological upstaging (PU) of clear cell renal cell carcinoma (ccRCC) from clinical cT1 to pT3 stage often requires conversion from partial to radical nephrectomy. Preoperative PU prediction lacks objective, precise methods, hindering surgical decision−making. Methods: We developed and validated a computed tomography−based radiomics ensemble learning model (CRIPEM) integrating intratumoral and peritumoral features to predict PU in cT1 ccRCC. We enrolled a multicenter cohort of 309 cT1 ccRCC patients from three institutions, divided into training (n = 170), internal validation (n = 73), and external validation (n = 66) cohorts. Bioinformatics analysis confirmed that peritumoral regions harbor critical predictive value for PU. Results: A total of 7336 radiomic features were extracted from intratumoral and peritumoral (1, 2, 3 mm) regions on preoperative CT images, with 50 robust features retained via a rigorous five−step selection process. CRIPEM, fusing optimal base learners (IT−MLP for intratumoral features, PT1−RF for 1−mm peritumoral features), achieved area under the curve values of 0.872, 0.807, and 0.826 in the training, internal, and external validation cohorts, respectively. Subgroup, calibration, and decision curve analyses confirmed its stability, superiority over single base learners, and significant clinical net benefits, with individualized cases verifying clinical applicability. Conclusions: CRIPEM is an objective, accurate, and robust tool for preoperative PU prediction in cT1 ccRCC, which can optimize surgical strategy selection and improve patient clinical management.