The validation leaderboard reports Small_instance_tp/fn/fp/f1 for each region. Could the organizers clarify how an instance is classified as "small"? Specifically: Is there a fixed volumetric threshold (in voxels or mm³), and if so, what is it? Are there corresponding "medium" and "large" strata that are computed but not displayed, or is the stratification binary? Is the evaluation code for the 2026 metrics (including NSD and the instance-size stratification) publicly available?

Created by Moises Costa moisescunha1999
Hi @moisescunha1999 Please read the [detailed description](https://www.synapse.org/Synapse:syn74274097/wiki/639602) on the challenge page to find out the exact answers. The evaluation pipeline is available via [GitHub](https://github.com/BraTS/BraTS_evaluation). Best regards, /Mehdi

Definition of the "small instance" threshold in the Task 1 leaderboard metrics page is loading…