> I used the IXI templates for spatial normalization for my functional data - it seemed that for functional data, it did a better job of
> normalization across subjects for the subcortical regions.
This would require some testing. In general, one would assume high-dimensional non-linear registratations like Dartel to outperform lower-dimensional ones (except if there's "over-bending"/insufficient regularisation as seems to have been the case in earlier versions of Dartel). Thus CAT12 should indeed result in better registrations then those obtained from SPM12 Segment. For a proper comparison it would be necessary to contrast CAT12 outcomes with those from SPM12 Segment combined with a Dartel registration. As the IXI templates might be biased towards the CAT12 pipeline (I think it resulted from CAT12 preprocessing) one should probably rely on an independent template set for evaluation.
> having this affine operation already implemented in the CAT12 Segmentation would be the way to go if you use pre-existing templates
This is certainly an interesting feature of CAT12. Another interesting feature is that there are different options to control the resolution of the Dartel imported/exported files in the GUI, while SPM12 results in 1.5x1.5x1.5 mm^3. If your data has better resolution (which is likely the case for structural data) then the Dartel from CAT12 might well preserve more of the information, which might also well translate into the final templates (independent of the affine/rigid-body only registration and the remaining processing differences between CAT12 segmentation and SPM12 Segment)
> How many subject would you say is recommended before you'd use DARTEL for your own template brain?
This should depend on the homogeneity of the sample. E.g. the IXI template is based on 555 subjects, but files were acquired on three different scanners (two 1.5T and one 3T) and the age range is large. For smaller samples a similar heterogeneity could already become problematic when creating one's own templates despite relying on healthy people only.
The CAT12 manual suggests that you should have at least 50 - 100 subjects. I would just give a try based on the 34 and look at the outcome. In principal you could also try to pool across studies, e.g. some of your own work, some data from colleagues (e.g. something like a "scanner-specific template / student population").
To be continued...
Helmut
|