Dear Robin,

Regarding your last question, 'Normalise to MNI space' adds an affine component, while 'Population to ICBM Registration' reflects a non-linear registration. The second difference is that 'Normalise to MNI space' refers to MNI space as represented in the TPM.nii, which is very close to the asymmetric version of the MNI ICBM 152 non-linear 6th Generation atlas (its symmetric version is available at http://nist.mni.mcgill.ca/?p=858 ), whereas the non-linear registration is conducted onto (a smoothed version of) the ICBM 152 Nonlinear atlas version 2009 http://nist.mni.mcgill.ca/?p=904 . Funnily, without modifying the scripts the target space is coupled with the type of registration, although these two aspects are unrelated in terms of content.

Whether to prefer the 6th gen or the 2009 ICBM version is one issue. The 2009 version is certainly the newer (and latest) official version, created with a more complex registration technique. However, it has not become a common target space so far; SPM and FSL continue relying on an asymmetric version of the 6th gen version as represented e.g. by the TPM.nii in SPM. I would stick with the 6th gen space as long as the 2009 space has not become a common space, as it eases comparisons with other studies.

The other issue is complexity of the registration. Whether an affine registration onto a template is sufficient to bring your data in good alignment depends on your data and your personal demands. It might work well on a global level, but on a local level, there might still be rather strong spatial deviations from the template.

For your ROI analyses you should ensure that the space of the atlas corresponds well to the space of your final data, be it 6th gen, 2009 or another MNI version. Therefore I would go with an atlas in which the raw structural volumes are available, so that you can re-segment them and bring them into the corresponding space. Some of those have been created several years ago, relying on older MNI versions, so they might be unnecessarily imprecise and in any case, differ from "your" space. One option would be the Hammersmith n30r83 atlas http://brain-development.org/brain-atlases/ , another would be the LPBA40 http://loni.usc.edu/atlases/Atlas_Detail.php?atlas_id=12 .
1) Working with the n30r83 should be very easy, just segment the files, determine the flow fields necessary to warp onto your template (or template plus additional transformation into "MNI" or "ICBM"), apply these onto the individual label images, either create probabilistic maps for the relevant regions or go with an MPM (can be constructed with Imcalc, select the 30 normalised label images and go with mod(X) ).
2) It would be more complicated with the LPBA40, as the volumes with the individual labels have been reoriented into what they call "delination space" (the file is called "LPBA40 Subjects Delineation Space: MRI and label files in delineation space"). Within this space they only provide structural volumes that have already been skull-stripped, with cerebellum and brainstem removed. The raw structural volumes are stored in "native space" still ("LPBA40 Subjects Native Space: MRI data and brain masks in native space"), with necessary reorientation parameters from native to delination space stored in .air format, for which you will probably need some of the LONI software. Maybe it is sufficient to coregister the skull-stripped structure in native space onto the skull-stripped cerebrum-only structure in delination space and apply this onto the raw structural volume. Then segment those structures, apply the deformations and so on like in 1)
For atlases with unavailable raw volumes but known preprocessing space one could at least try to warp from the corresponding template volume into your space. E.g. if a certain atlas represents 6th gen MNI space and you want it in 2009 MNI space you could segment the 6th gen T1 template (or take GM, WM, ... files if available, this is going to depend on the templates) and then go with a 'Population to ICBM Registration'. With spm_norm_population.m you could warp into other spaces as well, you just have to replace the icbm152.nii in line 13 by some other volume.

Hope this helps

Helmut