Artificial intelligence (AI), particularly deep learning–based computer vision, has the potential to enhance surgical safety. Several deep learning–based semantic segmentation models that recognize intraoperative anatomy have been developed for research1,2. However, these models remain at the proof-of-concept or feasibility study phase and clinical applicability has not yet been verified. This multicentre RCT aimed to verify the clinical applicability of a computer vision–based anatomical recognition model and provide evidence to support its utilization. An RCT was conducted at three hospitals in Japan from June 2023 to June 2024, including patients aged 18–80 years who underwent laparoscopic left-sided colorectal resection. The target structures for AI recognition were the left ureter and autonomic nerves (the right and left hypogastric nerves and aortic plexus). Included patients were randomized to the standard surgery or AI surgery group. In both groups, the intraoperative video was inputted into the system from the laparoscopic tower, and AI-inferred segmentation masks were superimposed on each frame in real time. In the AI surgery group, the superimposed video was relayed to a second adjacent monitor and used as an adjunct to intraoperative navigation (Fig. 1a).
No takes yet. Share an insight, caveat, or question.
Kitaguchi et al. (2025) studied this question.