Dear FSL list,
I’m having some issues obtaining accurate segmentations using FIRST (in FSL v5.0.6), possibly similar to a recent discussion though without the twist of a special population https://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=FSL;cc0f2664.1404
I’ve attached images for an example subject (healthy adult). registration.png shows the alignment of my T1w image to standard template (i.e. slicesdir with MNI152_T1_1mm with T1w_to_std_sub.nii.gz), suggesting that registration works well (although the outline of the brain doesn’t match, I'd say the subcortical parts look good?). segmentation.png shows the first_roi_slicesdir with T1w image and the first_all_fast_firstseg results. These look much less satisfactory, e.g. the putamen and caudate are not properly covered (also see segmentation2.png for clearer image). The problem is the same in origsegs.png, so it doesn’t seem to be a boundary correction issue.
The only thing I can think of that is unconventional in my setup is that we use 0.8 mm images acquired using tilted slices. I had to change first_flirt to obtain a proper registration like so:
Original:
${FSLDIR}/bin/flirt -in $A -out ${B}_tmp_stage1 -ref ${FSLDIR}/data/standard/${refim} -omat ${B}_tmp_stage1.mat $inweightCom $costCom
Adjusted to search entire space:
${FSLDIR}/bin/flirt -in $A -out ${B}_tmp_stage1 -ref ${FSLDIR}/data/standard/${refim} -omat ${B}_tmp_stage1.mat $inweightCom $costCom -searchrx -180 180 -searchry -180 180 -searchrz -180 180
I’ve tried increasing the modes, which didn’t help, but I feel like I’m missing something more important here. Any suggestions on tracking down this problem would be greatly appreciated. I'd be happy to upload niftis if that helps.
Thanks and best wishes,
Peter
|