Hi, groups,
I noticed the updates of longitudinal analysis in the new version of Freesurfer. However, I still have a couple of questions about the goal of longitudinal analysis.
1. The goal of longitudinal analysis is to detect the atrophy (thinning), but the transformation in the registration step will minimize any atrophy. An example of outcomes of FLIRT and FNIRT on FSL website exactly demonstrates that the atrophy disappears in the 'best' transformation (FNIRT) http://www.fmrib.ox.ac.uk/fsl/fnirt/index.html. In our experience, the number of vertex of post-transformed tp2 is very close to tp1. It is not clear to me how the difference in the original thickness between tp2 and tp1 is preserved in this transformation.
One may argue that processing tp1 and tp2 independently does not solve this problem because the registration is also used in the routine recon. Is it true?
2. The new longitudinal process is still using volume-based registration if I read it correctly. Why not switch to surface-based since people have argued the advantage of surface-based over volume-based?
Thank you for any clarification,
Wang
Hi Wang,
1. An affine transform from tp1 to tp2 would only remove global atrophy, but not the region effects we are interested in.
2. For intra-subject registration the surface-based is not needed as we expect the changes to be small.
cheers, Bruce
On Thu, 23 Oct 2008, Wang, Xin wrote:
Hi, groups,
I noticed the updates of longitudinal analysis in the new version of Freesurfer. However, I still have a couple of questions about the goal of longitudinal analysis.
- The goal of longitudinal analysis is to detect the atrophy (thinning), but the transformation in the registration step will minimize any atrophy. An example of outcomes of FLIRT and FNIRT on FSL website exactly demonstrates that the atrophy disappears in the 'best' transformation (FNIRT) http://www.fmrib.ox.ac.uk/fsl/fnirt/index.html. In our experience, the number of vertex of post-transformed tp2 is very close to tp1. It is not clear to me how the difference in the original thickness between tp2 and tp1 is preserved in this transformation.
One may argue that processing tp1 and tp2 independently does not solve this problem because the registration is also used in the routine recon. Is it true?
- The new longitudinal process is still using volume-based registration if I read it correctly. Why not switch to surface-based since people have argued the advantage of surface-based over volume-based?
Thank you for any clarification,
Wang
freesurfer@nmr.mgh.harvard.edu