Hi, I wonder the calculation of Dice in the volumetric images. If it means in the final prediction, we should use the volume of the 3D image of one patient? Because if we calculate it by slice, it would be curious for the whole zero label.
Dice scores will be evaluated on the patient level, and thus there will be no labels with all zeros.