Hi Daniel.
i'm currently applying mri_cvs_register and I'm very happy with the nice results. Congratulation for this very clever approach. However, some data sets which I work on have still aseg.mgz which are not segmented appropriately. As far as I understand aseg.mgz is not redefined after e.g. pial-edits or added control points. Thus, I think it might be better to use aparc+aseg or wmparc instead of aseg within mri_cvs_register. Taking this into account I have the following questions:
- Do you have experience with mri_cvs_register and --asegfname Option different from aseg? Does it impact the processing time? Are there options, which have to be adopted?
I only minimally experimented with this option, alowing users to use segmentation files that are edited or modified from the default version. processing time and registration options might change with a new file, but that will depend on thenature of the new file that you want to use. I belive that if you use either of the files that you mentioned above neither of those will (need to) change.
- Since I think it is better to use aseg instead of wmparc, since it contains fewer IDs and registration will not be overfitted, is there a convenient way to generate a new aseg.mgz, which includes manual corrections like pial edits?
If you make manual edits to the aseg.mgz file you can always save it under a different name and use the --asegfname option to indictae the new input to mri_cvs_register. For manual editing we recommend using Freeview and on the below page you may find useful advice http://surfer.nmr.mgh.harvard.edu/fswiki/FreeviewGuide/FreeviewWorkingWithDa...
- The last question regards the norm.mgz ... since it also often contains to much (outer brain) volume .. would it be better to mask norm.mgz with (e.g.) wmparc or a dilated version of wmparc before using it with mri_cvs_register ?
That never caused a problem for me, and the answer in your case will depend on the amount of error that exists in your datasets. If you do decide to use a mask though, you should do that consistently on all of your data sets.
Let me know if you have any further questions.
Lilla