Dice Score Computation, Patients w/o Lesions

Dear Organizers,

Thanks for organizing this great challenge. I would be more than happy if you clarify the computation of the metrics for patients without lesions.

As stated in the evaluation tab, the dice score is solely considered for cases with FDG-avid lesions. Looking at the following two cases:

• Case 1: (ground truth: no lesion, prediction: non-empty, i.e. lesion predicted)
• Case 2: (ground truth: no lesion, prediction: empty, i.e. no lesion predicted)

How do both cases influence the computation of the dice score displayed on the leaderboard?

Many thanks and best regards, Lars

Thanks for your question! As you already mention, we will only consider the false positive metric for the cases of negative ground truth masks. As Grand-Challenge does not allow for NaN or missing values in the metric output; we will output a Dice score of 0 for all participants in these cases for the automated leaderboard - this will not affect the ranking. After the challenge we will update the metrics be removing these Dice scores.