RESUMO
This study aims to explore the possibility and bottleneck of clinical translation for an artificial intelligence (AI) diagnosis system for bladder cancer based on cystoscopy.We retrospectively collected videos of 101 bladder cancer patients from January to November 2023, at Sun Yat-sen Memorial Hospital, Sun Yat-sen University. Among these patients, with a median age of 63 years and 81.0% were male. The bladder cancer AI diagnosis system was utilized for diagnosis, and the accuracy of diagnoses from the videos was assessed. Additionally, a surgical evaluation scale was formulated to evaluate the quality of the videos, simulating clinical usage.The final test results showed a system sensitivity of 97.8%, a positive predictive value of 81.7%, specificity of 54.2%, and a negative predictive value of 92.3%. Furthermore, the surgical evaluation scale scores ranged from 3.96 to 4.69, indicating the feasibility of clinical application for this system.This study further quantitatively validated the accuracy of an artificial intelligence system using cystoscopy videos and assessed the potential for clinical application.