> I am attempting to do VBM/TBM analysis using the method outlined below,
> which I found in the list archives, and I have a couple of questions.
>
> > “ 1) Coregister early and late images together (no reslicing).
> > 2) High-dimensional registration of early and late images.
> > Assume that the late image is the source, and the early
> > image is the reference (although it could be done the other
> > way around - or ideally could be done in a symmetric way).
> > - Deformations toolbox.
> > 3) Write out the Jacobian determinants from the resulting warp.
> > - Deformations toolbox.
> > 4) Apply the warp to the late image.
> > - Deformations toolbox.
> > 5) Do a "soft-mean" of the images (possibly including some global
> > scaling). This can be done with ImCalc: (i1+i2)./(eps + (i1~=0) +
> > (i2~=0)) 6) Spatially normalise this mean image (possibly via a
> > segmentation
> > step, as in "optimised VBM").
> > 7) Segment this spatially normalised mean image.
> > 8) Apply the spatial normalisation parameters to the Jacobian
> > determinants. 9) Use ImCalc, select the normalised GM, and the spatially
> > normalised Jacobian determinants, and evaluate the following expression:
> > i1.*(i2-1) 10) Modulate this result by the volume changes from the
> > spatial
> > normalisation. 11) Smooth the results by about 12mm.
> > 12) Do some stats on these.”
>
> I apologize if this has already been asked, but I could not find answers in
> the mail list archives. My questions are:
>
> 1) In step (9) above, what is the purpose for doing (i2-1), instead of,
> e.g. log(i2)? Wouldn't log(i2) be more appropriate here?
The idea is to compare regional tissue volumes. The procedure is analagous to
modulated VBM, where the grey matter is scaled to preserve the total volume.
In this case. For more info, I would suggest taking a look at
Ashburner J & Friston KJ (2001):
"Why voxel-based morphometry should be used."
NeuroImage 14:1238-1243
If you take logs of the Jacobians and then smooth, then you would be comparing
the logs of a kind of geometric mean (rather than simply comparing arithmetic
means) in the analysis.
>
> 2) Also in step (9) above, does multiplying by the GM probability at each
> voxel potentially confound the measure of interest, the Jacobian
> determinant?
The measure of interest is the difference in tissue volume between the scans.
Best regards,
-John
|