Hi Elizabeth,
It's fine to create difference images. If anything, the difference is
probably more normally distributed than the original data. The idea is
also similar to the common fMRI approach of getting contrast (con)
images for subjects at the first level, and then analysing these at a
second level of statistics. In your case, instead of subject, you have
twin-pair, and your con image is the difference between the twins.
It's also quite common in simple longitudinal VBM studies to create
chronological difference images; one example of their use is that a
one-sampled t-test of a collection of differences images is equivalent
to a paired t-test on the originals.
The only thing I would suggest that you look out for, is that if you
don't have an explicit analysis mask to use, then you might want to
create one --- use of absolute or relative threshold masking is
inappropriate on the difference images. One solution to this is to set
sub-threshold voxels to "NaN" before taking the difference, and then
to use implicit masking in the analysis. This can be done
automatically for an absolute threshold using my make_diffs script:
http://www.cs.ucl.ac.uk/staff/gridgway/vbm/make_diffs.m
Hope that helps,
Ged
> This is probably very basic but we don't know the answer and it is
> important (to us!). We have scans for a large number of twin pairs and
> we would like to look at the differences between various cogntive
> measures in relation to differences between the twins in grey matter
> using VBM methods. The first thing to occur to us was to create a
> difference scan by subtracting one twin's scan from the other. But if
> we do this, we are concerned that we will end up with a scan in which
> most of the voxels will have a value of 0 or a very small value since
> the twins will probably not differ over much of the brain. Does this
> violate the mathematical assumptions of VBM?
>
> Thanks a lot.
>
> Elizabeth Isaacs
> Dr. Elizabeth Isaacs
> MRC Childhood Nutrition Research Centre
> Institute of Child Health, UCL
> 30 Guilford St.
> London WC1N 1EH
> England
>
> E-mail: [log in to unmask]
> Telephone: 020 7905 2246
> Fax: 020 7831 9903
|