18 Nov Stanford Study: Computer Vision Can Reliably Document When Patients Enter and Leave the Operating Room

Dr. Chloe Nobuhara
MedicalResearch.com Interview with:
Chloe Nobuhara, MD
General Surgery Resident
Stanford University
Postdoctoral Researcher in Surgical AI
German Cancer Research Center (DKFZ)
Dr. Nobuhara is researching how AI can be successfully implemented into surgical practice.
MedicalResearch.com: What is the background for this study? How does the technology work?
Response: Operating Room (OR) timestamps, such as when a patient enters and leaves the room, are still entered by hand, usually by the circulating nurse in the middle of other work. Ambient AI scribes are already automating documentation in clinics. The OR is harder, because the events worth documenting depend on visual cues, and real-world validation studies are scarce.
We used Aimbient (formerly the OR Black Box; Surgical Safety Technologies), which continuously captures audiovisual data in the OR. A convolutional neural network classifies each camera frame as “patient present” or “patient absent.” It then writes entry and exit timestamps directly into the electronic health record, with no action needed from staff.
MedicalResearch.com: What are the main findings?
Response: Across 290 multispecialty cases, we compared computer vision and manual nurse entry against blinded, human-annotated ground truth. Computer vision was noninferior to manual documentation for both patient entry and exit. Both methods were within one minute of ground truth in 90% to 95% of cases.
Computer vision showed no drift by time of day. Overnight, manual entries averaged about 106 seconds of error vs about 35 seconds for computer vision. To our knowledge, this is the first validation of computer vision for automated OR documentation across a broad multispecialty case mix.
MedicalResearch.com: What should readers take away from your report?
Response: Computer vision is now a reliable alternative to manual entry for routine OR documentation. Governance comes first: our deployment was possible because audiovisual feeds are fully deidentified and cannot be used for disciplinary action or case discovery at Stanford. The leadership in our Interventional Platform has seen the promise of ambient documentation frankly far ahead of its time, and we look forward to entering a world where monotonous, time-consuming tasks can be outsourced to AI.
AI can’t replace humans, but it can replace human error. Furthermore, I believe AI can give humans back the most precious resource, which is time, and allow us to focus on what brought us to medicine in the first place: being face-to-face with our patients.
MedicalResearch.com: What recommendations do you have for future research as a result of this study?
Response: We also plan to validate timestamps for other key intraoperative events, which would let the system automatically keep preoperative and postoperative care areas updated. Future studies should also measure the real-world impact of automation on staff time and OR efficiency.
MedicalResearch.com: Is there anything else you would like to add? Any disclosures?
Response: Disclosures: I am a consultant to Aimbient. I was first brought on to consult on Explorer, their platform where surgical residents use AI to review their own operative footage and get feedback on their performance. That work is separate from this study, but important to disclose nonetheless. Co-author disclosures are listed in the published article.
Citation:
Nobuhara CK, Sakref Y, Forrester JD, et al. Validation of Computer Vision in the Operating Room for Ambient Documentation. JAMA Surg. Published online November 18, 2026. doi:10.1001/jamasurg.2026.5500
For a broader overview of artificial intelligence in medicine, see this MedicalResearch.com overview of [anchor text].
The information on MedicalResearch.com is provided for educational purposes only, and is in no way intended to diagnose, cure, or treat any medical or other condition. Some links are sponsored. MedicalResearch.com and Eminent Domains Inc. do not warrant or endorse products or claims made by third party links. Always seek the advice of your physician or other qualified health provider and ask your doctor any questions you may have regarding a medical condition. In addition to all other limitations and disclaimers in this agreement, service provider and its third party providers disclaim any liability or loss in connection with the content provided on this website.
Last Updated on November 18, 2026 by Marie Benz MD FAAD