If you consider a region of the data (after spatial normalisation), then the
sum of the Jacobians over that ROI will give an estimate of the volume of the
region for each subject. You can think of the results of spatial smoothing
in terms of a Gaussian weighted ROI centred at each voxel, so I guess whether
or not you smooth will depend on the spatial scale of the volumetric
differences that you want to detect.
Another aspect to consider may be whether or not to work with the logarithms
of these volumes, rather than the volumes themselves (allometry). Logarithms
of such measures can loosly be conceptualised as growth rates (under an
assumption that the rates remain constant), so may be a bit closer to the
actual causes of anatomical variability than the measures themselves. I
could go on, but maybe I'll save that for a paper.
Best regards,
-John
On Monday 02 March 2009 15:56, Michel Grothe wrote:
> Dear SPMers,
>
> I am using DARTEL for cross-sectional TBM and I wonder if I have to smooth
> the jacobian determinant maps before statistical analysis (two-sample
> t-test). I found this old post and the question is if the answer changed
> when using the DARTEL algorithm (or large diffeomorphic deformations with
> lots of degrees of freedom in general)?
>
> Any suggestions will be highly appreciated.
>
> Thanks in advance,
>
> Michel Grothe
|