@Article{cju.2026.080402,
AUTHOR = {Salvador Jaime-Casas, Roberta Corvino, Jan Łaszkiewicz, Amir Khan, Benjamin I. Chung, Vincenzo Asero, Valerio Santarelli, Dalila Carino, Sarah Carvalho Ribeiro, Stefano Impaloni, Francesco Del Giudice},
TITLE = {Artificial intelligence (AI)-based models for bladder cancer (BC) endoscopic detection: a systematic review},
JOURNAL = {Canadian Journal of Urology},
VOLUME = {},
YEAR = {},
NUMBER = {},
PAGES = {{pages}},
URL = {http://www.techscience.com/CJU/online/detail/27724},
ISSN = {1488-5581},
ABSTRACT = { <b>Background:</b> Artificial intelligence (AI)-based models are increasingly being explored for the diagnostic evaluation of patients undergoing assessment for bladder cancer (BC). The aim of this study was to systematically review the literature regarding the use of AI models for tumor detection during bladder endoscopy in patients with suspected BC, providing a comprehensive qualitative synthesis of their diagnostic performance. <b>Methods:</b> We systematically searched PubMed, Embase, and Scopus databases from inception to 30 September 2025 in accordance with Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines. Eligible studies reported the diagnostic performance of AI-based technologies applied to cystoscopy and transurethral resection of bladder tumor (TURBT), in patients with suspected BC. We excluded case reports, meeting abstracts, narrative reviews, case series, and non-original publications. Data extraction was performed using a standardized approach and included clinicopathological data and performance metrics. Study quality was independently assessed using the National Institutes of Health Quality Assessment Tool and the “Quality Assessment of Diagnostic Accuracy Studies (QUADAS-2)” tool. <b>Results:</b> A total of 12,840 patients across 13 studies published between 2018 and 2025 met the inclusion criteria. A total of 236,925 images/frames were analyzed, of which 36,890 contained lesions. AI models included convolutional neural networks (CNN), U-Net variants, GoogLeNet, ResNet, DenseNet, EfficientNet, CystoNet, and the Cystoscopy Artificial Intelligence Diagnostic System. 12 studies were retrospective, and 1 was prospective. 10 studies were conducted at a single institution, and 3 had a multicenter design. AI-based models demonstrated moderate diagnostic performance, reliably identifying BC with moderate to high accuracy. <b>Conclusions:</b> AI models are promising diagnostic tools, potentially adjunctive in clinical practice, but without yet-established standardized performance metrics. However, methodologies across studies are heterogeneous, and results remain context dependent.},
DOI = {10.32604/cju.2026.080402}
}