users > dilate into the void
Showing 1-3 of 3 posts
Display:
Results per page:
Jan 26, 2012  02:01 PM | Yaroslav Halchenko
dilate into the void
I wonder if any tool is already capable to dilate a volume carrying atlas labels, where unmarked voxels within X mm receive label matching the label of the nearest "non-void" voxel?

my motivation: some volumes were skull stripped tiny bit too aggressively by FSL's bet on highres anatomical -- as a result some border voxels after nearestneighbor reslicing of the sri24 atlas into the BOLD space lack labeling.  Obvious somewhat non-scientific approach I thought of is to dilate label map but so that labels information propagates  to the closest unoccupied voxels if multiple labels start to compete.

P.S. meanwhile I will just research on better parameters for bet to make it less aggressive ;-)
Jan 26, 2012  02:01 PM | Torsten Rohlfing - Google LLC
RE: dilate into the void
Sorry, no such tool. Also, how far should be dilated into the void? You will just end up labelling non-brain tissue, no?

Really, skull stripping needs to be fixed here. Suggestion - do you have other images in addition to T1w of the same subject? Say, an additional PDw and T2w? If so, strip each separately with BET, align to T1w, reformat T2w and PDw masks, and combine all three by voting.

TR
Jan 26, 2012  03:01 PM | Yaroslav Halchenko
RE: dilate into the void
well -- not much further -- anyways I would have then overlapped with the betted BOLD.  So I have only T1 and T2* sequences.

indeed I could try taking the mask of betted BOLD and use its union with T1 mask

P.S. someone eventually should start working on nipype interface for CMTK ;-)