Objectives: White light cystoscopy (WLC) remains the standard tool for the diagnosis and surveillance of bladder cancer (BC), but it may miss up to 20% of lesions, highlighting the potential role of artificial intelligence (AI) assisted systems such as CystoSmart™. The study aims to evaluate the sensitivity and specificity of CystoSmart™ in detecting papillary and flat lesions, using histopathology as the reference standard. Methods: This is a prospective observational, non-interventional study with retrospective AI application, led by “Sapienza” University of Rome. Patients undergoing primary or follow up WLC for suspected BC will be enrolled and will provide informed consent for anonymized recording. Frames will be retrospectively analyzed by the software without influencing clinical decision-making. Exclusion criteria include metastatic disease, inability to provide consent, inadequate video quality, and contraindications to cystoscopy. A total sample size of 158 patients was determined to ensure at least 138 positive cases. Statistical analyses will apply confidence intervals and exact binomial testing. Results: Based on the sample size calculation and the preliminary internal validation data, CystoSmart™ is expected to achieve sensitivity exceeding 90% for BC detection, with specificity also anticipated to exceed 90%. Diagnostic accuracy, sensitivity, specificity, positive and negative predictive values will be estimated at both the per-patient and per-lesion levels. Exploratory subgroup analyses will assess performance across primary diagnostic, intraoperative, and surveillance settings. Conclusion: AI-assisted cystoscopy has the potential to facilitate the integration of AI- based tools into routine BC clinical practice.
Giudice et al. (2026) studied this question.