White light cystoscopy (WLC) is the gold standard for bladder cancer (BC) detection and surveillance, but has limited sensitivity, particularly for small and flat lesions. We evaluated the diagnostic performance and safety of a real-time artificial intelligence (AI)-assisted support tool (CystoAID©) for bladder cancer detection.
This single-center, randomized, controlled non-inferiority trial included 64 patients undergoing transurethral resection of bladder tumor or laser fulguration for suspected primary or recurrent BC. Patients were randomized 1:1 to WLC alone or WLC followed by AI-assisted cystoscopy. The primary outcome was per-lesion sensitivity of CystoAID vs. WLC within the intervention group. The reference standard was histopathology or clinical evaluation. The predefined non-inferiority margin was -5%. Secondary outcomes included sensitivity for lesions ≤ 5 mm, false positives, procedural duration, and safety (30-day adverse events).
A total of 142 lesions were identified (84 intervention, 58 control). Sensitivity was 96.2% (95% CI 87.0-99.5) for CystoAID and 88.7% (95% CI 77.0-95.7) for WLC (difference 7.5%; 95% CI -2.7 to 17.8), demonstrating non-inferiority. In a sub-analysis for lesions ≤ 5 mm (n = 25), sensitivity was 100% (95% CI 86.3-100) vs. 80% (95% CI 59.3-93.2). AI assistance added a median of 2.1 min without further affecting workflow. Adverse event rates were low and similar between groups.
CystoAID demonstrated non-inferior sensitivity to WLC and could be safely integrated into the routine clinical workflow. Larger studies with improved representation of flat lesions are warranted, as well as to assess whether CystoAID can achieve superior sensitivity compared with standard WLC.
Cancers. 2026 May 26*** epublish ***
Peter Blak Hjort, Katharina Skovhus, Jørgen Bjerggaard Jensen, Andreas Ernst
Department of Clinical Medicine, Aarhus University, Palle Juul-Jensens Blvd. 82, 8200 Aarhus N, Denmark.