Evaluation, Reproducibility, Benchmarks Meeting 38 - Project-MONAI/MONAI GitHub Wiki
Minutes of Meeting 38
Date: 17th September, 2025
Present
- Rucha
- Annika
- Olivier
- Carole
- Nick
- Lena
- Nicola
Introductions
Welcome, Rucha!
Update from MONAI Advisory Board Meeting
- There was significant interest in the data licensing project
- Comment came up that the parent organization is what enforces custom language within existing licenses
- Any comments about community participation?
- No, nothing from the advisory board
Update From Licenses Project
- Is there a tool out there? There are some to explain licenses
- We could create a tool, but we need to find a legal expert to help, and finding one has been difficult
- Interesting note -- the parent organization may be playing a role in what license is being used
- Implications of this work
- Wake up call for people to be more careful
- Modeling good practice -- even if we don't have guidelines, we can provide case studies and focus on practical advice
- Is there a correlation between license use and citations? Industry vs academic use?
Lighthouse Challenge Re-Annotation Project
- Potential synergy with the MONAI Label team
- Nick to try to get in touch and have a conversation at MICCAI conference
Potential Collaboration with Other Working Groups
- Ultrasound group (new) expressed interest, but it is not a top priority on their side just yet. Annika will update us if this changes
- Potential synergy with federated learning working group -- Annika/Carole to reach out and just start a conversation
Updates from Olivier/Confidence Interval Project
- Background
- There is poor reporting of confidence intervals in our community
- This is at odds with regulatory needs, and also just a bad scientific practice in general
- Could help to explain limited translation
- Lack of CIs also leads to unjustified claims of superiority in new methods
- Goal of the project
- Point out the problem
- Present experimental results in medical imaging work specifically
- Not guidelines yet, but headed in that direction
- Update
- Finished additional experiments this summer
- Working on journal article that expands on workshop paper
- Each "benchmark instance" is one team's performance on one particular task (in decathlon)
- Plan
- Want to send this to Medical Image Analysis
Thoughts for Next Time
- How do we get more developers contributing?
- Unclear how much is done by core developers vs community contributions
- Next meeting October 22