Multiphase CT-based deep learning radiomics nomogram models for preoperative WHO/ISUP grading of clear cell renal cell carcinoma: a two-center validation study.
Authors
Affiliations (5)
Affiliations (5)
- Department of Urologic Oncology, Tianjin Medical University Cancer Institute & Hospital, National Clinical Research Center for Cancer, Key Laboratory of Cancer Prevention and Therapy, Tianjin's Clinical Research Center for Cancer, Huanhuxi Road, Hexi Distinct, Tianjin, 300060, China.
- Department of Urology, Sun Yat-Sen University Cancer Center, Guangzhou, 510060, P. R. China.
- State Key Laboratory of Oncology in South China, Guangdong Provincial Clinical Research Center for Cancer, Sun Yat-Sen University Cancer Center, Guangzhou, 510060, P. R. China.
- Department of Urologic Oncology, Tianjin Medical University Cancer Institute & Hospital, National Clinical Research Center for Cancer, Key Laboratory of Cancer Prevention and Therapy, Tianjin's Clinical Research Center for Cancer, Huanhuxi Road, Hexi Distinct, Tianjin, 300060, China. [email protected].
- Department of Urologic Oncology, Tianjin Medical University Cancer Institute & Hospital, National Clinical Research Center for Cancer, Key Laboratory of Cancer Prevention and Therapy, Tianjin's Clinical Research Center for Cancer, Huanhuxi Road, Hexi Distinct, Tianjin, 300060, China. [email protected].
Abstract
Preoperative determination of clear cell renal cell carcinoma (ccRCC) World Health Organization/International Society of Urological Pathology (WHO/ISUP) nuclear grade is crucial for surgical planning, yet current invasive biopsy approaches carry significant risks and diagnostic limitations. Existing radiomics and deep learning studies predominantly utilize single-phase imaging or isolated methodologies. We developed and validated a Deep Learning Radiomics Nomogram (DLRN) that integrates multiphase computed tomography (CT) imaging with machine learning for WHO/ISUP grading. This two-center study analyzed 1499 histologically confirmed ccRCC patients, allocated to training (n = 929), internal validation (n = 398), and external validation (n = 172) cohorts. Our DLRN model integrates three complementary data streams: radiomics features extracted from non-contrast, corticomedullary, and nephrographic CT phases; deep learning features from a multi-channel DenseNet201 architecture; and clinical variables. Model performance was evaluated using the area under the curve (AUC) and calibration analysis. Interpretability was enhanced through Shapley Additive Explanations (SHAP) and Gradient-weighted Class Activation Mapping (Grad-CAM) analyses. The DLRN model achieved favorable discriminative performance with AUC values of 0.935, 0.901, and 0.911 in training, internal validation, and external validation cohorts, respectively. This significantly outperformed individual component models in external validation: clinical (AUC = 0.730), radiomics (AUC = 0.845), and deep learning (AUC = 0.868) models (p < 0.05). Calibration analysis demonstrated excellent agreement between predicted probabilities and observed outcomes. SHAP analysis revealed deep learning features as dominant predictive contributors, while Grad-CAM visualization consistently focused on tumor heterogeneity patterns characteristic of different grades. The DLRN model based on multiphase CT imaging provides an accurate, non-invasive tool for preoperative prediction of ccRCC nuclear grade.