Dear ZhenliangNi, the metric computes the distances from the reference mask to the predicted masks. With these surface distances, TPs, FPs and FNs are computed: We define a TP if the distance is below a certain threshold which depends on the inter rater variabilty. If the distance is bigger than this threshold, it will be counted as FP or FN (depending on the direction). Based on those distance TPs, FPs and FNs, the formula of the DSC is used to compute the metric. Kind regards, the ROBUST-MIS organizers

Dear sir? It is difficult to read the code directly.Could you please provide mathematical description? Thank you.

Dear ZhenliangNi, please refer to the implementation of the metric, provided here: https://github.com/schnobi1990/robust_mis_2019/tree/master/evaluation Kind regards, the ROBUST-MIS organizers

Evaluation metricspage is loading…