AI-based smoke removal in robotic urooncologic surgery: a clinical pilot study

Abstract

Abstract Surgical smoke from electrocautery limits visibility during robot-assisted oncological surgery by obscuring tumor margins, blood vessels and anatomical landmarks. AI-based smoke removal demonstrated promising technical performance, structured clinical evaluation remains limited. This pilot study examined surgeons’ perceptions of a temporally consistent 3D U-Net applied to intraoperative smoke. Twenty urologists and residents retrospectively assessed ten pairs of endoscopic videos offline, showing original and AI-processed clips. A questionnaire evaluated general impressions (seven items, five-point Likert scale), direct comparisons across four image dimensions, open-ended feedback and overall utility. Visibility of anatomical details and depth perception were rated equal to the smoke-affected original by 75% of participants and slightly better by 25%. Color fidelity was rated equal by 85% and slightly better by 15%. None rated these dimensions worse (0/20, exact 95% CI 0 to 16.8%), and no participant selected the highest rating. Image stability was rated worse by 20%. Half disagreed that the processed footage appeared clearer. Overall utility was rated positively by 40%, neutrally by 35% and negatively by 25%. In this small, unblinded, single-institution retrospective pilot study, AI processing was not associated with a perceived worsening of anatomical detail, depth perception or color fidelity. However, confidence intervals remain wide and the study was not designed to establish equivalence. Most participants perceived no improvement in image clarity or orientation support. Artifact reduction, real-time performance and blinded multicenter evaluation are required before clinical use.