In short
You annotate recordings of real operations in your browser: outline organs and instruments and mark the phases of the procedure. Computer vision models for surgery are trained and tested on this annotation.
For you, it is a way to see hundreds of operations before your first night on call, to test your anatomy on real tissue, and to get data for a term paper, a thesis or a conference talk.
What you get
A real medical AI project
This is not a classroom exercise. It is an open research project. You work with the same datasets that research groups use around the world – Cholec80, Endoscapes2023, AutoLaparo – in a professional annotation tool, CVAT, under the quality rules that scientific publications expect. Your annotation becomes part of the data that models learn from and are tested on.
Case exposure
Laparoscopic cholecystectomy is one of the most common operations in general surgery. Over one semester in the project you will watch dozens of them from start to finish, from trocar placement to gallbladder extraction.
You will see how the same operation unfolds in different conditions, what bleeding, smoke and adhesions look like, and how a surgeon reaches the critical view of safety before clipping the duct. No textbook gives you that volume, and neither does a single visit to an operating room where you mostly see the team’s backs.
Anatomy with feedback
Your first tasks are calibration tasks: frames that already have a reference annotation by experts. You outline the gallbladder, the cystic duct and the liver, and you get a report on how closely your outline matched the reference for each structure.
It is like an anatomy practical where every specimen has an exact answer, except the tissue is living and seen through a laparoscope. Disputed frames are reviewed by a surgeon, the project’s clinical lead, and the decisions go into a shared log that every participant can read.
How medical AI really works
You will follow the whole path from a frame to a model: how data is collected, why annotators must agree, how a model learns and why it fails.
When a “smart” assistant arrives in your hospital, you will know what data it was trained on and where it must not be trusted. Every doctor will soon need that critical eye.
A skill people value
Annotating medical images and video is a competence of its own, between medicine and data. Experience of working to a strict guideline, with quality control and agreement metrics, is useful in research groups, in clinical trials and in companies that build medical software.
Research on real data
Student research usually lacks data: it has to be collected, permitted and annotated. In this project that work is done or under way. We have collected more than 50 research topics, from the anatomy of Calot’s triangle to how far annotators trust automatic suggestions.
Conferences
You can present results from the project at student research conferences and in the young-researcher sessions of surgical and anatomical congresses. The project helps with data, methods and a review of your abstract before you submit it.
Recognition
- A profile on the website in the list of contributors, with your consent. It shows your role and the hours of accepted annotation.
- Co-authorship of project publications under the same rule as for every annotator: at least 20 hours of annotated and accepted material, under a contribution agreement.
- Proof of participation: on request, the project writes a letter describing your role and the amount of work, for a portfolio, a scholarship or an application.
Publications and talks help your portfolio. Many universities count them as individual achievements in residency and doctoral admissions. Rules depend on the university and the year, so check yours.
People
You work alongside surgeons, machine learning engineers and students from other universities and cities. These connections help when you choose a specialty, a residency and a supervisor.
How it works
- Apply. Write to us: your year, your university, what interests you.
- Onboarding. A short online meeting, the annotation guideline and training in CVAT. An administrator creates your account; there is no self-registration.
- Calibration. Tasks on frames with a reference annotation, and a report on how closely you matched it.
- Real tasks. Batches of 50–100 frames or one video. A second participant reviews every batch; a surgeon reviews disputed frames.
- Your own topic. Optionally, a term paper, a thesis or a talk from the list of topics or your own idea.
Time: about 2–4 hours a week, whenever suits you and from any city. You need a computer with a browser and a mouse; a graphics tablet is more comfortable but not required.
Who we are looking for
- Medical and paediatrics students who have passed human anatomy.
- Senior students and residents interested in surgery, anatomy, radiology or medical informatics.
- Biomedical and computer science students, for work at the boundary of medicine and data.
You do not need to know how to code.
What to know first
- Only open and permitted data. The project works with published de-identified datasets and with videos whose authors gave written permission. Never upload recordings from your hospital or placement, photos of patients or any information about them.
- Annotation is not a diagnosis. Your outlines are research data. The project’s clinical lead has the final word on medical definitions.
- Research use only. The project’s models are not intended for treating patients.
- The project is just starting. We are preparing data and getting the annotation guideline approved; pilot tasks begin after that. Now is a good time to join and be part of the first group of annotators.
How to join
Write to support@opensurgicalvision.org with the subject “Student: annotation”, or leave a message in the project discussions.
Studying computer science? You have your own page.
Are you a teacher who wants to bring a group? See the page for educators and departments.