Hello everyone,

We'd like to quickly address a question we’ve received a few times about the expected DSC scores/difficulty of the segmentation tasks. Tumor segmentation in this challenge is likely significantly “harder” than previous similar tumor segmentation challenges (e.g., such as HECKTOR). MRI-based segmentation in head and neck cancer is generally a noisier annotation process than, say, PET-based segmentation. While we would like to avoid giving any specific numerical ranges of anticipated performance, we do expect geometric evaluation scores to be lower than in past challenges, e.g., DSC likely below 0.75. Also, please remember we'll be using DSCagg for evaluation instead of traditional volumetric DSC. Don't be discouraged if your internal evaluations yield lower scores than initially thought - this is actually expected (especially for Task 2!).

Best,

Kareem