Radiomics and deep learning in upper tract urothelial carcinoma: advancing preoperative risk stratification and clinical decision-making.

Background: Upper tract urothelial carcinoma (UTUC) is a relatively rare but aggressive malignancy. Accurate preoperative assessment of tumor grade, invasiveness, and prognosis remains challenging using conventional imaging, cytology, and ureteroscopic biopsy alone. Radiomics and deep learning may provide noninvasive tools for improving risk stratification and clinical decision-making. Methods: This narrative review summarizes current evidence on radiomics, machine learning, and deep learning in UTUC. Relevant studies were identified from PubMed, Web of Science, and Scopus and synthesized according to clinical applications and methodological considerations. Results: Radiomics and deep learning models have shown promising performance in pathological grade prediction, differentiation between UTUC and renal cell carcinoma, muscle invasion assessment, and survival or recurrence risk stratification. However, most studies remain retrospective, single-center, and limited by small sample sizes, heterogeneous imaging protocols, inconsistent segmentation methods, insufficient external validation, and limited evidence of clinical utility. Conclusion: Radiomics and deep learning are promising approaches for noninvasive preoperative risk stratification in UTUC. Future studies should focus on methodological standardization, multicenter external validation, prospective evaluation, model interpretability, and demonstration of incremental clinical benefit before routine clinical implementation.