What we teach
We want to teach people to check claims, not just to run models. We plan an introductory five-session course on medical computer vision, with notebooks and cards from real experiments. The course is in development.
Five questions for any experiment
- Where does the data come from? Origin, permissions and the split – before looking at a metric.
- What is it compared with? Which baseline, and which conditions are fixed.
- What changed? One change, one testable hypothesis and an ablation.
- How certain is the result? Variation between runs and confidence intervals.
- What did not work? Negative and unfinished results are material for discussion too.
For students
A term paper or thesis can start with documentation, a test on synthetic data or a reproducibility check – before any GPU access. Assignments never require patient files or access to clinical systems. A sound method and honest limitations count, not a positive result at any cost.
Medical students have their own page about annotating surgical video, and there are 56 research topics built on project data. Computer science students: engineering and machine learning in the project.
For teachers
Propose an exercise through a repository issue (without any student details), or see the education track. Medical educators: how to run a student research group or theses on the project.