I am not the slightest bit surprised that the results differ, as an
analysis of different pre-processed data is likely to produce
different findings. The accuracy of the results, both in terms of
sensitivity and interpretability, will depend on the accuracy of the
pre-processing models used, as well as the quality of the original
scans. According to various validations of image registration
algorithms (eg by Klein et al), Dartel appears to be more accurate
than the older registration approaches in SPM, so I would generally
trust it rather more.
Best regards,
-John
On 4 May 2011 10:28, Diederick Stoffers <[log in to unmask]> wrote:
> Hi all,
> I ran a DARTEL VBM analysis on an old dataset (two-group GM comparison)
> that had previously been analysed in spm5 without DARTEL, I am unable to
> reproduce any of the group differences. I checked al the steps in the old
> analysis (it used the best practices available at that time (e.g. segment,
> linear and non-linear registration to a study-specific half-way template
> including modulation), all seems just fine. The DARTEL stream
> (vanilla-flavored, analysis in MNI space) also doesn't show any obvious
> problems. Both analyses used 8mm smoothing, 2mm voxel size and the same
> segmentation algorithm.
> I am unsure how to best explain the differences between the analyses and how
> to check whether previously found differences were veridical. Is the most
> likely culprit a difference in folding between groups? Should I use lower
> smoothing in the DARTEL analysis? Other explanations?
> Any pointers on how to find the cause of these differences would be greatly
> appreciated!
> Cheers,
> Diederick
>
>
|