I think I stumbled upon a bug in how FEAT runs FNIRT in FSL 4.1.4.
Here's the relevant output from running FEAT with the FNIRT option in 4.1.4:
/usr/local/fsl4.1.4/bin/fslmaths /Volumes/Storage/freylab_projects/exp37-02_MRI_robot_grasping/data/mri/1/anat/1_003_t1_mprage_tra_20100119/0001776_001_003_t1_mprage_tra_20100119_brain.nii.gz highres_head
and the corresponding output from running 4.1.5 (which has fixed the bug, though it wasn't mentioned in the "what's new" page (http://www.fmrib.ox.ac.uk/fsl/fsl/whatsnew.html#revisions):
/usr/local/fsl4.1.5/bin/fslmaths /Volumes/Storage/freylab_projects/exp37-02_MRI_robot_grasping/data/mri/1/anat/1_003_t1_mprage_tra_20100119/0001776_001_003_t1_mprage_tra_20100119 highres_head
The key is that fnirt wants the anatomical image that hasn't been skull-stripped. It assumes that the skull-stripped image ends with _brain (e.g. foo_brain) and that the un-skull-stripped image has the same name without the _brain ending (e.g. foo). It then uses the un-skull-stripped image to create a new image called highres_head, which is sent to fnirt. If you compare the calls to flsmaths, you'll see that in the 4.1.4 version, the skull-stripped anatomical is used (ends with _brain), while in the 4.1.5 version the correct version is used (doesn't end with _brain). To verify that this is what was actually happening, I compared the registration results for 4.1.4 versus 4.1.5. The "Registration of highres to standard" section for 4.1.4 shows a skull-stripped brain, while the 4.1.5 version does not.
Am I missing something?
cheers,
-marc
|