> 1. can anyone point me to a good example of a paper where a standard VBM
> analysis, but using DARTEL, was used. I am unclear about the best way to
> write this up...
Look for papers by Stonnington, Kloppel or Draganski from the last year or so.
>
> 2. Secondly, some recent correspondence on DARTEL has drawn a dichotomy
> between 'VBM using DARTEL' and 'optimised VBM'. However, this seems odd to
> me - using DARTEL tools is still optimised VBM, is it not?
The idea is to use the most accurate pre-processing available at the time, in
order that any differences you detect will be easier to interpret in terms of
volumetric differences - rather than systematic misregistration or systematic
misclassification.
I'm not very keen on the terminology "optimised VBM". That was just one
approach, whereby we found that we could obtain more accurate spatial
normalisation by matching GM from the individual with a GM template image.
The unified segmentation of SPM5 kind of had this built in. The best
evidence I currently have available to me suggests that DARTEL should give
higher inter-subject registration accuracy than any of the other registration
approaches in SPM.
The other parts of the "optimised VBM" strategy were:
1) Scale the warped data by the Jacobian determinants ("modulation"). This
gives images that have a quantitative meaning in terms of tissue volume. Not
using the Jacobian scaling means that the warped GM images are not
quantitative, and the statisical tests are a kind of evaluation of
intersubject registration accuracy (differences indicate registration
errors).
2) Maybe the use of a "custom template". The main reason for this was that
the old segmentation routines did an initial affine registration by
minimising the least squares difference between a template and the image to
segment. Because each group seems to have MR scans with different
properties, then this registration does not necessarily work so well (as the
contrast etc of the images constituting the template differs from that of the
images being segmented). In the unified segmentation of SPM5 onwards, the
contrast in the images is less important, so there is less need for such
custom templates.
Best regards,
-John
|