Aligning FreeSurfer files to normalized Anatomy

Hi Guys:

So we’re in the process of making some anatomy-based ROIs from FreeSurfer segmentation files. We’ve run our EPI and anatomy data through most of a uber_subject preprocessing pipeline and will ultimately be wanting to do all of our analyses in a standardized space. I had figured that we could just take the aparc_aseg.mgz file that comes out of FreeSurfer and use the cat_matvec transformation matrix on this file to get it to be aligned with the subject anatomy in MNI space.
So we did this and the segmentation file is aligned fine to the anatomy, EPI, and MNI template (which are all aligned with each other well enough). The problem is, we used the following command: 3dallineate -base anat.al_strip+tlrc -input aparcaseg_shft+orig -1Dmatrix_apply mat.r01.warp.aff12.1D -prefix FS-SEG
The anat.al_strip+tlrc file has a grid space of 1x.97x.97 and the aparcaseg_shft+orig file has a grid space of 1x1x1.

So we didn’t use the -master or -mast_dxyz flags. The segmentation file should default to the grid space of the anat.al_strip file, is my understanding but instead the normalized segmentation file turns into a 2x2x2 file so it looks really blocky and bad. Is this because we’re using the 1Dmatrix we are? I’m not sure how it’s transforming into that voxel size.
Would it be kosher to just try align_epi_anat.py on the segmentation file to try to just the segmentation file into MNI space as an alternative? Ultimately we just want a high res segmentation file aligned with the EPI so that we can extract more precise ROI timecourses but as it stands these normalized segmentation files look too coarse to do that.

Thanks for your help.

It’s possible that specifying a -master or -mast_dxyz may help… However - if you’re planning to do things in group (standard) space, then I suspect you’d have the best luck downloading the SUMA surfaces for either TT_N27 or MNI and using those segmentations to do your ROI analysis.

There’s some problem with using a standard affine warp for moving your segmentations up to group space. The most striking is that affine warps tend to favor global alignment over local structures (Kochunov et al., 2002 in NeuroImage) and the Freesurfer segmentations have a lot of local structures defined! So if you can do things with the group template segmentation, that may help. Otherwise, if you want to use the individual subject segmentations, I’d recommend you use the AFNI nonlinear warp (3dQwarp, add -tlrc_NL_warp to your proc.py command). Then you can use the 3dNwarpApply to move your segmentations to standard space while preserving both local and global properties.

-Peter