Segmentation performance will be evaluated using the Dice Similarity
Coefficient (DSC). A python script for evaluating model performance
can be downloaded here.
Ranking will be based on the overall score: weighted dice, with
weights inversely proportional to the size (volume) of each anatomical
structure (cartilage, bone).
A score of zero will be assigned to those with missing results.
Reconstruction will not be evaluated. Participants may choose to
develop a segmentation model directly from the undersampled k-space or
image reconstruction as an intermediate step.
