Here are my tips as someone who routinely registers longitudinal studies.
You do not need to play with the interpolation, use tri-linear (or spline if you want sharp edges)
1) Skull strip - this always helps (if your looking at the brain)
2) If it is a single subject, you may want to play with the DOF (generally 6 DOF for single subject multiple scans and single visit, 7 for longitudinal, i.e. multi-visit registrations).
3) Cost function, I always use normalized mutual information (normmi). My experience has taught me that this always provides the best registration. Feel free to try a variety of them, just be consistent.
4) Register to a template image (I use freesurfers mri_robust_template)
5) Worst case scenario, truncate one of the images z-axis to say 20 slices mid-brain. Then register this to the other image (care must be taken to then apply transform to original - non ROI image)
I have more tricks, but cannot think of them. I am not an expert, just someone who registers many brains.
Cheers,
Bryson