Should ROI masks be binarized individually before merging in FSL?

In FSL, when combining multiple probabilistic ROI masks (e.g., from an atlas), why is it recommended to threshold and binarize each mask separately before merging them, instead of just merging first and then thresholding once?

Hi @seymatakir, if you merge two overlapping probabilistic images, values in the overlapping region will be added together, so you will end up with invalid values there.

Note that this can also occur when merging binarised masks, if you use a threshold that is too low. E.g. if you add two binary ROIs which cover the same region, you will end up with values of 2 in that region.