> I am wondering, in the HDW toolbox, is it possible to specify a mask image
> over which the likelihood potentials are calculated?
No. Not without quite a lot of additional coding.
> Or does it even make
> sense to do that? I am wondering if it is possible to constrain the
> warping algorithm such that its cost function is calculated only in the
> masked regions. I have noticed that in some images where there is
> extremely bright dura near the brain, the presence of the dura in one image
> but not the other seems to be driving the warping, i.e. the warping field
> calculated over the brain voxels near this bit of dura seems to be heavily
> influenced by the presence or absence of the dura. In other words, we are
> finding large regions of "expansion" or "contraction" in the brain voxels
> near the bright dura, but it seems to be a result of the algorithm trying
> to match the dura between the two scans. I have tried simply extracting
> the brain from both images, which seems to help in some ways, but I believe
> it introduces artificial edges which have their own unintended effects on
> the result. Sorry for the long winded explanation, but is there a way to
> mask out certain voxels from the likelihood potential calculation without
> introducing an artificial edge? Thanks for any insight you can offer,
I don't have any bright ideas that would help here. Are you doing
inter-subject registration? If so, then you could try the DARTEL toolbox,
which warps GM to GM and WM to WM.