Hi Darren,
hope all is well.
> I have read through the list but still have a question about FSLVBM
> and setting up a proper model. My understanding is that the gray
> matter data are modulated by the Jacobian, but that the
> recommendation is that a covariate for total intracranial volume is
> not necessary in the analysis. I'm not sure I understand why. Does
> the Jacobian include both affine and non-linear aspects of the
> normalization? If so wouldn't modulating by the affine scaling mean
> that size differences would not be accounted for between subjects
> any more?
The contributions to the Jacobian from the affine and the non-linear
parts of the transform are kept separate within fnirt. That allows us
to specify the range of allowed Jacobians based on the full Jacobian
(which makes sense) while the modulation of the gray matter maps is
based on the non-linear part only.
It really isn't straightforward to say what is the correct thing to do
in terms of modulation.
Let us say e.g. that we have two brains that are identical, save for
one being twice as big (volume) compared to the other. So we perform
an affine registration without any modulation, and we conclude they
are identical. If on the other hand we modulate the gray matter
content based on that affine transform we would conclude that the
larger brain has twice as much gray matter, and hence that there are
massive differences all along the cortex.
What makes most sense?
I would say that from the perspective of what we are typically
interested in the former makes most sense, i.e. that in fact there is
no difference between the two brains (or in other words that any
difference can be attributed to an uninteresting size difference).
Let us now instead say that we have two other brains that are
identical in size save for a thinning of the somatosensory cortex. Let
us further say that we register the thin cortex guy to the thick
cortex guy using some very fancy registration algorithm that results
in two identical looking images. We segment these images, compare the
gray matter segments and conclude that the brains are identical.
This clearly doesn't make sense, and the suggested way to solve that
is by modulating by the local (non-linear) Jacobian which in this case
would mean that we multiply the gray matter segment of the thin cortex
guy by 0.5 along the somatosensory cortex. And when comparing the
modulated gray matter segment to thick cortex guy we conclude,
rightly, that thin cortex guy has less gray matter in the
somatosensory cortex.
So it is clear that it makes sense to modulate the very local volume
changes, and to me it also makes sense not to modulate global scaling
changes. So where is the cutoff in terms of spatial frequency of
warps? I don't really know, and clearly our choice to discard global
(affine) volume changes and modulate all non-affine volume changes is
an arbitrary one that is based more on procedural convenience of
division of volume changes than on anything else. On the other hand I
don't know if anyone has a more principled suggestion.
As for the inclusion/non-inclusion of global scaling parameters in the
subsequent GLM analysis I don't think there are any clear and hard
rules there either. If one e.g. think that "smaller brains tend to
have proportionally thicker cortices" (I have no idea if that is the
case or not) then as one scale those brains up to fit the template
their cortices will end up being slightly thicker (have "more" gray
matter) their bigger brained peers. Then the question is if this is an
interesting effect or not? If one considers this to be an
uninteresting general size effect then one would want to get rid of
this source of variability and the way to do that would be to include
it as a covariate.
> A second question is whether FSLVBM can be used on white matter? If
> so would it just be a matter of changing the Template specification
> at the top of the fslvbm_2_template script, and then making sure
> files were named appropriately through, e.g., WM for GM in file names?
Procedurally it would be very simple, much as you suggest. As for
interpretation I am uncertain as to what people think that white
matter VBM means.
Good luck Jesper
|