Answers are grounded in this paper and cite their sources.
Study integrates clinical data and ML for predicting recurrence-free survival in ccRCC, indicating potential for personalized care.
Background: Integrated multimodal systems improve recurrence-free survival (RFS) prediction in surgically resected clear cell renal cell carcinoma (ccRCC) to overcome traditional model limitations. Methods: A total of 1284 patients were enrolled and divided into training (n = 738), internal validation (n = 317), and external test (n = 229). A machine learning (ML)-based model combined clinical data, radiomics signature (Rad-sign), and deep learning signature (DL-sign) to predict recurrence at 3-, 5-, and 7-year post-surgery. Radiomics features were extracted from CT images using ML, with the Rad-sign showing the best performance. A three-dimensional vision transformer (3D-ViT) for volumetric feature extraction outperformed ResNet models. Results: Cox regression identified clinical features associated with RFS ( p < 0.05). An optimal clinical-DL nomogram was chosen from 8 ML algorithms via 5-fold cross-validation. Clinical features alone had limited predictive power (external test AUC: 0.576, 95% CI: 0.472–0.672), while radiomics modeled by Naive Bayes (NB) performed better (AUC: 0.728). Rad-sign's C-index was 0.729, and the 3D-ViT model achieved an external test AUC of 0.846. The nomogram's AUCs for predicting 3, 5, and 7-year RFS post-surgery were 0.920, 0.935, and 0.938, with C-indices of 0.890 and 0.910, respectively. It effectively stratified risk groups, outperforming UISS and SSIGN via calibration and decision curve analysis. Conclusion: The ML-based clinical-DL model shows promise for accurate ccRCC RFS prediction, which may aid in personalized risk assessment and decision-making.
No takes yet. Share an insight, caveat, or question.
Wang et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: