By: wildsquirrel on Jan. 29, 2025, 10:44 a.m.
Hi Challenge Organiser, Can you clarify if the quality of nuclei segmentation will be assessed (i.e. the shape of each segmentation mask/polygon)? To me it seems like we only care about the centroids of the predicted nuclei, since that's how the F1 score is calculated. Thank you.
By: mschuiveling on Jan. 29, 2025, 3:41 p.m.
Dear Wildsquirrel,
This is correct. Due to the problems with IoU based metrics in small object segmentation (see https://www.nature.com/articles/s41598-023-35605-7) we choose this centroid distance metric. Shape of segmentation has no importance for this challenge, it is purely about detection. For more information you can also look at this website and their accompanying paper (https://metrics-reloaded.dkfz.de/metric-selection)
Good luck with the challenge!
Kind regards, Mark
