Artificial intelligence in the diagnosis and prognosis of upper tract urothelial carcinoma: A comprehensive review of current evidence, methodological standards, and future directions.

Abstract

870 Background: Upper tract urothelial carcinoma (UTUC) is a rare malignancy (5–10% of urothelial cancers) with challenging diagnosis and heterogeneous outcomes. Artificial intelligence (AI)—including radiomics, machine learning (ML), deep learning (DL), and digital pathology—has emerged as a tool to improve detection, risk stratification, and treatment planning. Methods: A systematic literature review (PubMed, Scopus, Embase) to August 2025 identified original human studies using AI for UTUC diagnosis or prognosis. Extracted data included sample size, AI methodology, validation type, and key performance metrics. Results: Twenty-one studies met inclusion criteria (2018–2025): CT urography radiomics (n = 12), prognostic perirenal fat/peritumoral texture analysis (n = 2), AI-assisted urine cytology (n = 2), DL-based digital pathology (n = 2), and multimodal models integrating imaging and clinical/molecular data (n = 3). Radiomics achieved AUCs of 0.80–0.94 for grading/staging; multimodal approaches improved predictive accuracy (ΔAUC ≈ +0.06). AI-assisted cytology reached sensitivity > 85% for recurrence detection, and digital pathology predicted lymph-node status with AUC up to 0.85. Most studies were retrospective and single-center; only 14% reported external validation. Conclusions: AI shows strong promise for UTUC diagnosis and prognostication, particularly via CT radiomics and multimodal integration. However, clinical adoption requires multicenter prospective validation, standardized imaging/pathology workflows, regulatory approval, and cost-effectiveness evaluation.