@DiceMetric is a script that expects masks as input, and there can be multiple masks per dataset (1,2,3,...). Consider masking the data separately to a create a binary dataset that is your mask or using 3dABoverlap to compute overlap of two datasets. There's no Dice coefficient there though.
3dABoverlap -no_automask dset1 dset2
Use the -no_automask dataset if the dataset has already been thresholded, or take it out if it hasn't.