Just to second Jonathan's comment, I would like to point out that in
the statistical literature concerned with observational studies (which
is almost invariably the case for VBM studies), TIV or TGM are not
legitimate covariates. Because the variable of interest may influence
TIV or TGM, a model including these variables as covariates is
considered as giving biased effects of the variable of interest
(example reference below). The interpretation of the results may be
quite different, as indicated by Jonathan.
A study including such variables as covariates cannot be interpreted
by the reader unless their effect on the outcome is also reported
(i.e. the regression on TIV or TGM), or, equivalently, the SPM without
the covariate.
Best wishes,
Roberto Viviani
Univ. of Ulm, Germany
Paul R. Rosenbaum, The consequences of adjustment for a concomitant
variable that has been affected by the treatment, J. R. Stat. Soc. A
(1984), 147:656-666.
Quoting Jonathan Peelle <[log in to unmask]>:
> Just to second Donald's comment, it's not that including TIV or total
> gray matter (TGM) is "right" or "wrong"; you're controlling for
> different effects, which may or may not be desirable depending on what
> you are looking at. Additionally, as Donald points out, CSF is not
> always estimated particularly well, which may be something to
> consider. In my experience TIV can significantly reduce variability
> (even with possibly inaccurate CSF), for example, particularly in
> reducing sex differences.
>
> Whatever covariates you include, just be sure to adjust your
> interpretation appropriately. I.e., if you include TGM, then your
> results show significant areas of difference "beyond" or unaccounted
> for by global differences in gray matter. In some comparisons this
> might be seen as an (appropriately) conservative approach; for
> example, if you are comparing patients with significant cortical
> atrophy to controls, they are likely to have less GM in a number of
> areas; including TGM as a covariate may help localize focal reductions
> that go above and beyond what could be explained by overall volume
> reduction. But there are likely to be regions that don't show up
> because, although having less GM than controls, this difference can be
> explained by global changes. So having included TGM as a covariate
> you wouldn't want to conclude that only the resulting significant
> regions have less GM than controls, but that only these regions show
> GM reductions *above and beyond that which can be explained by global
> differences* relative to controls. This may seem like a subtle point
> but I think it's important for the way we think about our results.
>
> Hope this helps a bit and didn't add to the confusion!
>
> Best regards,
> Jonathan
>
>
>
> On Mon, Mar 29, 2010 at 3:48 AM, MCLAREN, Donald
> <[log in to unmask]> wrote:
>> TICV is the preferred method as it gives a measure of potential size
>> irrespective of brain volume. For example, if you have severe atrophy, then
>> brain volume is confounded by the pathology whereas TICV should be less
>> influenced. However, older versions of SPM and other programs have been poor
>> at estimating the CSF volume as it is measuring the background of the T1w
>> image. This has led many groups to use TBV. If you have good T1w and T2w
>> images, then you can use multi-spectral segmentation to get an accurate
>> measure of the three tissue classes and use TICV.
>> TBV might not be a bad correction either, since you are then asking the
>> question, relative to the global shrinkage, were is the regional changes
>> more significant.
>> Another caveat when doing patients vs controls is to consider that maximal
>> brain size of patients relative to TICV might be lower, which could
>> complicate the analyses and interpretation.
>> Hope this helps.
>> Best Regards, Donald McLaren
>> =================
>> D.G. McLaren
>> University of Wisconsin - Madison
>> Neuroscience Training Program
>> Office: (608) 520-0586
>> =====================
>
|