Real-time Detection of Bladder Cancer Using Augmented Cystoscopy with Deep Learning: a Pilot Study

Area of Science:

• Urology
• Medical Imaging
• Artificial Intelligence

Background:

• Detecting bladder tumors during white light cystoscopy (WLC) is difficult and impacts treatment.
• Artificial intelligence (AI) shows promise for improving tumor detection, but real-time applications are unexplored.
• Previous AI use in bladder cancer detection focused on post-procedure analysis of recorded images.

Purpose of the Study:

• To evaluate the feasibility of integrating a real-time AI system into live cystoscopy and transurethral resection of bladder tumor (TURBT) procedures.
• To assess the performance of a novel real-time AI alert system (CystoNet) during live surgical video feeds.

Main Methods:

• Prospective enrollment of patients undergoing flexible cystoscopy and TURBT.
• Development and integration of a real-time AI alert device (CystoNet) with standard cystoscopy equipment.
• Real-time processing of streaming cystoscopy videos to provide synchronized alert boxes and measure per-frame diagnostic accuracy.

Main Results:

• Real-time CystoNet was successfully implemented in 55 procedures (21 clinic cystoscopies, 34 TURBTs) across 50 patients.
• For clinic cystoscopy, CystoNet achieved 98.8% per-frame tumor specificity with a median error rate of 3.6%.
• For TURBT, CystoNet demonstrated 52.9% per-frame sensitivity and 95.4% specificity for bladder cancers, with a 16.7% error rate.

Conclusions:

• The pilot study confirms the feasibility of real-time AI (CystoNet) for providing surgeons with active feedback during cystoscopy and TURBT.
• Further optimization of CystoNet may enhance its utility in AI-augmented cystoscopy for improved clinical outcomes.