Why this project exists
Surgery that sees more
A new member is joining the surgical team: an AI assistant. It recognises organs, vessels and instruments in the video, knows which phase the operation is in, and warns when an instrument comes close to a dangerous structure. Before the duct is clipped it can check whether the critical view of safety has been reached, and after the operation it can help review the recording and show a young surgeon where the key decisions were made.
The next step is autonomous surgical systems that perform individual steps of an operation under a surgeon’s supervision. The first experiments already exist. In 2022 the STAR robot performed a laparoscopic intestinal anastomosis in pigs on its own. In 2025 the SRT-H system, trained on videos of surgeons, performed the clipping and cutting phase of a cholecystectomy on pig organs outside a living body. Use in patients is still far away: it needs years of testing, clinical trials and regulatory decisions.
All of these systems share one foundation: recordings of operations annotated and checked by experts. The quality of that data decides what a system learns and whether it can be trusted. OpenSurgicalVision does not build a medical device: we build an open research foundation on which such systems can be trained and tested honestly.
Knowledge and technology for everyone
According to the Lancet Commission on Global Surgery (2015), about five billion people, two thirds of the world’s population, lack access to safe, affordable surgical and anaesthesia care. In low-income and lower-middle-income countries, nine people in ten go without it.
If AI for surgery is built only in closed labs and sold at prices that only large clinics in wealthy countries can pay, that gap will grow deeper. Technology that could help a surgeon in a district hospital will become a privilege of those who already have the best equipment and the best specialists.
We believe modern medical knowledge and technology should be a common good, not a privilege of wealthy countries or individual clinics. So we open everything we can: code, models, evaluation methods, experiment results including the failures, and teaching materials. A university, a hospital or a researcher in any country can check our work, repeat it and improve it. A student in any city learns from the same materials as students in leading centres.
What stands in the way today
Research on surgical video is hard to compare: teams use different data, annotate it differently and spend different amounts of compute. An impressive number in a paper often cannot be reproduced.
OpenSurgicalVision wants to change that. We are building an open, shared foundation:
- data with a clear origin and license;
- models anyone can reproduce;
- an experiment log that shows what worked and what did not.
The project is just starting. This page describes where we are heading; see the roadmap for what is already done.
Three moments of an operation
We think about the work through three stages of surgery:
- Before surgery – 3D models of organs built from CT and MRI.
- During surgery – recognising instruments, anatomy and phases in video.
- After surgery – a structured timeline and summary of the recording.
This is a map of research questions, not a finished system. Read more in How it works.
Why in the open
- A verifiable experiment is more useful than an impressive demo.
- Open work lets more than programmers take part: annotation, testing, writing, translation and teaching all matter.
- Negative results save other researchers time.
People make the medical decisions
Annotation rules, disputed cases and medical wording are approved by the clinical lead, a practising surgeon. That responsibility cannot be handed to a model or settled by a vote.
Learn about the roles in the team, ways to contribute and the project’s boundaries.
Contacts
- Questions, support and requests: support@opensurgicalvision.org
- Project mailbox: opensurgicalvision@proton.me
Both addresses are official. Never send patient data, medical records or recordings of operations by email.