Dear everyone,
> which I hope does not contradict mine in too many places :)
Let's see ;- ) But I think we complement one another pretty well. Nonetheless let's make sure once more what the modules do:
1) Segment. The CAT12 algorithm relies on some templates, tissue priors, masks that would not be age appropriate and in any case, are not "study-specific". The CAT12 segmentation also includes a Dartel registration per default. This registration takes some time and is useless in case one wants to Dartel register onto a study-specific Dartel template at a later point anyway. I'm also not aware of any child Dartel template that one might want to enter instead. To avoid this part completely one can thus just modify cat_main.m by switching do_dartel from 1 to 0. Make sure that you don't tell SPM to write out normalised/modulated images, flow fields, jacobians, ... that rely on the Dartel registration, otherwise you would likely obtain some error messages. You can write out "Dartel exported" files though.
2) Based on these "Dartel exported" files create your template. This is already study-specific then and can be analyzed.
3) However, the study-specific template might not agree too well with MNI templates. To minimize discrepancies one could transform the images with "Normalize to MNI space" or "Dartel (Population to ICBM Registration)" = turning your study-specific template into an MNI / ICBM version of that.
If you go this way then you will have relied on some age inapproriate files for some initial steps during 1). This should be (much?) less of an issue as they are really only used for some initial steps compared to when working with SPM12 segmentation. This could be solved if you were able to create good tissue probability maps based on the output from 2). Then you could just repeat the steps 1), 2) (and possibly 3) ). However, it might require manual editing to obtain good Dartel images/TPMs for e.g. skull and such, so probably only an alternative in theory.
As a compromise, you could work with some pre-existing files for the initial steps in 1) that are closer to your population than the default files in CAT12. That is, you could specify different images in cat_defaults.m for cat.opts.tpm at least (some TPMs from the TOM toolbox, some other child-specific TPMs), ideally also different cat.extopts.brainmask and cat.extopts.T1. One might be able to obtain files like that rather easily based on some modified TOM ones, which leaves open cat.extopts.cat12atlas though. Possibly it's better to rely on the default ones then instead of changing only some of those. Christian Gaser should know about this.
Finally, to come back to Marko's statement "You need to do two rounds: one, to segment your data based on a standard template, then use the affine-registered images to create your own DARTEL template, then repeat the segmentation with your own template." You really only have to repeat the segmentation module if you think that the first affine registration is going to differ to a relevant part from an affine registration onto the study-specific template. This might be negligible if you start with some more appropriate files right from the beginning like stated above.
Best
Helmut
|