Dear list:

I recently exchanged some emails with Christian Gaser about VBM and
correcting for brain size. I found Christian's answers and literature
references very helpful and thought others might as well.

Regards,
Darren


---------- Forwarded message ----------
From: Christian Gaser <[log in to unmask]>
Date: Wed, May 19, 2010 at 7:47 AM
Subject: Re: TFCE
To: Darren Gitelman <[log in to unmask]>


Hi Darren,

this is a really good point, which I have not yet considered in my
thoughts. Thus, I have checked for some evidence of a relation between
TIV and atlas scaling factor with regard to dementia and found the
paper of Buckner
 (http://dx.doi.org/10.1016/j.neuroimage.2004.06.018). Figure 6 shows
a relation between atlas scaling factor (=inverse determinant of
affine transformation) and manually derived TIV for young, old, and
demented subjects. There is no clear deviation between the 3 lines.
This means that atlas scaling factor (ASF) is indeed a good
approximator for TIV regardless of a potential atrophy. To be honest I
have checked this relation only for 40 brains, where I had a manually
derived TIV (and found a strong correlation between TIV and ASF). This
was the motivation to not consider the affine term of the
transformation for modulation if you want to correct for different
head size or TIV. However, I am surprised (and happy) about the
independence of the ASF. This encourages me in my decision for the
modulation with non-linear terms only. Most of the variance explained
by age and gender will be removed from the data.

There is also a nice FS-wiki on that issue:
http://surfer.nmr.mgh.harvard.edu/fswiki/eTIV

Maybe this is also of interest for the spm maillist readers...

Regards,

Christian
Am 19.05.10 00:43, schrieb Darren Gitelman:
>
> I've had another thought that I wanted to run by you. I'm not sure if
> this is correct or not and perhaps I am just thinking of it the wrong
> way so I haven't posted it to the lists (SPM, FSL) yet.
>
> I am concerned that correcting for only the non-linear portion of the
> Jacobian and assuming that affine scaling takes care of the global
> (i.e., brain size) effects may be particularly wrong for disorders
> that are likely to affect large areas of the brain (i.e., nearly all
> neurodegenerative disorders).
>
> My thought is that the affine scaling attempts to match the subject
> brain to the template in overall size. If a brain is generally smaller
> than the template then the scaling will be increased. The software
> obviously doesn't know if the brain is smaller because it's normally
> smaller or it's smaller because it is atrophic. In the former case
> (i.e., small normal brain) the brain is properly scaled to the
> template. Then the size of the various structures are also
> proportionately scaled and I can see that there would not be a need to
> include TIV. In the case of atrophy, however, I think the brain is
> scaled in a sense inappropriately. Expanding the size of an atrophic
> brain would potentially create more brain. For example, if the brain
> had global cortical atrophy but the subcortical structures were not as
> affected, then in trying to match the overall brain size I think the
> affine scaling would expand the size of the subcortical structures
> possibly beyond normal.
>
> Essentially not correcting for the affine part of the Jacobian says
> that scaling a normal small brain and scaling an atrophic brain are
> equivalent, and I think it would lead to reducing differences between
> groups in the case of atrophy vs. when the full Jacobian was used for
> correction and then TIV used as a covariate.
>
> The other issue that bothers me, that I'm not sure how to fix, is that
> even using TIV assumes that the brains being compared were similar at
> some point in time. What I mean is that I think one has to assume that
> the proportions of gray matter, white matter and CSF were at some
> point similar between the groups. In this case, TIV would have a
> similar proportional relationship to members of each group. However,
> what happens if the proportions of gray/white/csf were never the same?
> Then the TIV correction has a different meaning for each group. Take
> the case of some developmental disorder that starts out with
> proportionately less gray matter. Now the measure of TIV isn't really
> related in the same way to the amount of actual brain in each group. I
> don't know if there is any way to correct for this.
>
> Darren