Are you registering all your subjects together, or doing something else with
the data? In principle, it should just take about 20 minutes per subject -
although there are a few extra overheads involved in a kind of smoothing of
the template data.
I still have to extend the DARTEL guide to explain how to use the option for
spatially normalising to MNI space. There is a little bit of online
documentation for it, which may be enough to get it working.
Note also that you may want to try the new segmentation toolbox for generating
"imported" GM and WM (and possibly also other tissue classes). If you use
this approach, then the importing would actually be done by setting the
appropriate outputs for the segmentation - as opposed to using the import
option that is part of the DARTEL toolbox.
Best regards,
-John
On Thursday 30 April 2009 18:24, Haakon Engen wrote:
> Dear John,
>
> Thanks for your input. It was that mail that prompted my questions, in
> a manner of speaking. I think I'll go for DARTEL based on this and the
> Klein et al article. One thing, though: You say that the registration
> should take 20 minutes. However, I spent some 2 hours watching the
> normalization step do it's magic yesterday. Is there perhaps some
> problem with my setup or hardware that's causing the slowing?
>
> On a tangential note, for the list: Does anyone know of a "recepie"
> for using DARTEL normalization on fMRI data? The Dartel guide is
> somewhat sparse when it comes to that subject (at least for those of us
> that are rather new to the game), and I haven't been able to find
> anything through Google, either.
>
> Thanks!
> Haakon
>
> John Ashburner skrev:
> > The actual registration takes about 20 minutes per subject, which could
> > be run overnight. I would suggest using the version in SPM8 (see email
> > from yesterday), which has the option to create DARTEL aligned images in
> > MNI space.
> >
> > One of the problems with methods that are able to achieve more extreme
> > deformations is that some regions can shrink by a great deal, and this
> > results in signal loss from those regions - especially after smoothing.
> > The option that warps to MNI space attempts to prevent the loss of this
> > fMRI signal by (hopefully) generating smoothed spatially normalised
> > images in a slightly more optimal way.
> >
> > Best regards,
> > -John
> >
> > On Thursday 30 April 2009 16:40, Haakon Engen wrote:
> >> Dear fellow SPM'ers,
> >>
> >> I am currently setting up a preprocessing pipeline for a group of 40
> >> subjects.Normalization is rather important for this project, seeing as
> >> how I am planning to do DCM modeling on rather small subcortical ROI's.
> >> I am therefore considering using DARTEL to spatially normalize my fMRI
> >> volumes. My initial test runs have been successful, but it has dawned on
> >> me that DARTEL normalization takes quite a bit longer than standard
> >> normalization.
> >>
> >> I was therefore wondering if any of you have experience with DARTEL
> >> normalization and could inform whether or not it is worth the extra
> >> processing time, with regards to accuracy
> >>
> >> Second I have some questions regarding the precise procedure of doing
> >> DARTEL with regard to fMRI normalization. Is it necessary to calculate
> >> the Templates and flow fields for each subject, or should I include all
> >> the GM and WM segmented images for all subjects in one calculation?
> >>
> >> Thanks in advance,
> >> Haakon Engen
> >> Graduate student/Research Assistant
> >> University of Oslo, Norway
|