AI enhanced surgeons' faster responses to intraoperative bleeding in laparoscopic surgery: a preclinical randomized controlled trial
In brief
AI alerts helped surgeons spot laparoscopic bleeding 0.7 seconds faster
In a preclinical randomized trial of 77 surgeons, AI assistance shortened the time to identify a bleeding site from 3.23 to 2.53 seconds and lowered reported negative emotion scores. The detection model was trained on videos from 20 medical centers, but the trial tested surgeons' responses, not whether AI improves outcomes for patients in actual surgery.
- Journal
- Surgical endoscopy (Q1)
- Published
- 13 October 2026
- Study design
- Randomized controlled trial
- Evidence level
- Level 1, High (CEBM 1b)
- Authors
- Jing Zhang, He Cai, Junjie Zhang, Zixin Chen, Yang Zhou, Ran Hu, et al.
- PMID
- 42869971
- DOI
- 10.1007/s00464-026-13490-1
Why clinicians should know about it
- Picked for Surgery (paper of the day, 16 October 2026): Preclinical RCT, AI shortens bleeding reaction time
Abstract
OBJECTIVE: To develop and validate an artificial intelligence (AI) model for real-time detection and tracking of intraoperative bleeding sites during laparoscopic surgery, and to evaluate whether AI assistance improves surgeons' bleeding‑identification performance. METHODS: From January to September 2023, laparoscopic surgical videos were retrospectively collected from 20 medical centers in China. Senior surgeons identified bleeding events, extracted relevant video clips, and manually annotated bleeding frames to train the AI model. Model performance was evaluated using the F1-score, mean absolute error (MAE), and mean distance error (MDE). A preclinical randomized controlled trial was subsequently conducted. Eligible surgeons were randomly assigned to the AI-assisted (experimental) group or the control group. The primary outcome was reaction time, defined as the time elapsed from the onset of bleeding to the participant's first successful click on the bleeding site. RESULTS: A total of 257 laparoscopic surgical videos were collected from 20 medical centers. The training, validation, and test sets comprised 145 (56.42%), 55 (21.40%), and 57 (22.18%) videos, respectively. The model achieved an F1-score of 0.860 (95% CI, 0.838-0.878). The MAE was 3.315 (95% CI, 3.051-3.614), and the MDE was 0.087 (95% CI, 0.086-0.088). A total of 77 surgeons were randomly assigned to the experimental group (n = 38) and the control group (n = 39). Compared with the control group, the experimental group demonstrated a significantly shorter reaction time (2.53 s vs. 3.23 s, P < 0.001) and a lower negative emotion score (1.68 vs. 2.95, P < 0.001). CONCLUSION: The AI model achieves high accuracy in real‑time detection and tracking of bleeding‑site targets within laparoscopic surgical videos. Furthermore, AI‑assisted support significantly shortens surgeons' reaction time for bleeding‑site identification and reduces their negative emotional burden when confronted with bleeding events. TRIAL REGISTRATION: Chinese Clinical Trial Registry (ChiCTR2600118185).
Abstract as published, via PubMed.
For healthcare professionals. The summary is generated by AI from the published abstract, and the evidence level is assigned automatically from the study design on the Oxford CEBM hierarchy. Neither is medical advice. Read the full paper before changing practice.