Diagnostic performance of radiomics for detecting and characterising upper tract urothelial carcinoma (UTUC): a systematic review.
Authors
Affiliations (8)
Affiliations (8)
- Department of Urology, Royal Perth Hospital, East Metropolitan Health Service, Perth, WA, Australia. [email protected].
- Department of Urology, Joondalup Hospital, Joondalup, WA, Australia.
- Department of Radiology, St James' University Hospital, Dublin, Ireland.
- Trinity St James Cancer Institute, Trinity College Dublin, Dublin, Ireland.
- Northwell, New Hyde Park, NY, USA.
- Department of Urology, Lenox Hill Hospital, New York, NY, USA.
- Department of Urology, Royal Perth Hospital, East Metropolitan Health Service, Perth, WA, Australia.
- Department of Urology, Sir Charles Gairdner Hospital, North Metropolitan Health Service, Perth, WA, Australia.
Abstract
Upper tract urothelial carcinoma (UTUC) is a rare but aggressive malignancy where accurate preoperative assessment of tumour grade and stage is essential to guide treatment. Conventional tools such as ureteroscopic biopsy have limited accuracy and procedural risks. This systematic review evaluated the diagnostic and prognostic performance of radiomics models for predicting grade, stage, histotype, muscle invasion, and outcomes in UTUC. Following PRISMA guidelines (PROSPERO Registration ID: CRD420251141966), PubMed, EMBASE, CENTRAL, and grey literature were searched up to 6 September 2025. Eligible studies included radiomics-based analyses in adults with UTUC. Extracted performance metrics (AUC, sensitivity, specificity) were summarised as medians with ranges. Twelve studies (<i>n</i> = 1,685) were included, encompassing 45 distinct radiomics models: 20 for grade, 11 for stage, 10 prognostic, three histotype, and one for muscle invasion. Models predicting high-grade disease demonstrated excellent performance (median AUC 0.876, range 0.675–0.961; sensitivity 0.838; specificity 0.806). Prognostic models achieved a median AUC of 0.854 (range 0.750–0.933) and sensitivity of 0.909, effectively predicting recurrence-free and overall survival. Stage prediction models yielded a median AUC of 0.771 (range 0.711–0.860), while histotype models reported a median AUC of 0.84. The single study assessing muscle invasion achieved an AUC of 0.821. Most studies used CT-based radiomics with machine-learning classifiers, though validation methods and feature selection strategies were heterogeneous. Radiomics demonstrates high diagnostic and prognostic accuracy in UTUC, particularly for tumour grade and survival prediction. These findings support its potential as a non-invasive adjunct for risk stratification and surgical planning, though standardised multicentre validation is required before clinical adoption. The trial was prospectively registered on PROSPERO on 07/09/2025 under ID: CRD420251141966. The online version contains supplementary material available at 10.1007/s00345-026-06191-w.