Hi, Robin--

There is a lot going on here.... I for one have never run the FS recon-all longitudinal pipeline, so I am not sure what is happening there.

To be clear, is "@Make_Special_FS" actually "@SUMA_Make_Spec_FS"? I am also confused by what "your" T1 is here.

There are a lot of things that can be happening here. As your data changes (coordinates become less well-centered, obliquity is used or not the FOV size changes), some of these considerations can become more or less important.
+ If it were me, to start I would deoblique my T1w dataset; doing it as follows will preserve where the coordinate origin (x,y,z)=(0,0,0) is:

3dcopy DSET_INPUT tmp
3drefit -oblique_recenter tmp+orig
3drefit -deoblique tmp+orig 
3dcopy tmp+orig NEW_DSET.nii.gz

+ If there is a huge amount of non-brain material, this might negatively affect alignment, esp. in finding the initial starting point (and it makes the files bigger and more unwieldy, needlessly, too). For that, as you have shown, 3dZeropad can be useful.
+ If your coordinate origin is *not* in a reasonable spot, then using 3dCM can be very useful, too.

At this point your anatomical shoudl be in good shape for everything. If there is a mismatch between the initial anatomical and the FreeSurfer, regular recon-all output and the contents of the SUMA/ directory, I would assume that that is due to having obliquity in the original is not being propagated along. This can be verified with:

3dinfo -obliquity -prefix DSET_INITIAL DSET_IN_SUMA_DIR

Even though your acquired anatomical might have voxels that are not 1mm isotropic and matrix dimensions that are not 256x256x256, the initial dataset and SUMA/ directory contents should overlay if the initial did not have obliquity. This is because AFNI uses the coordinates to locate data in space, not just voxels stacked from a corner so voxel size and matrix dimensionality would matter.

Re. the longitudinal analysis, I don't know what the "template space" you refer to is. Is it something that overlaps neither T1w_1 nor T1w_2 (the anatomicals at each time point)? If that is the case, then I don't see why the volumetric ROIs *should* overlap eiither input dataset (by definition). Or is this just about obliquity again? The initial T1w_? dataset is oblique, but the output of recon-all and @SUMA_Make_Spec_FS is not?

--pt