Hi Jason, Iroise,
I was about to write along the same lines when you beat me to the finish
line :) and I fully agree: GM volume, for example, is highly non-linear,
and peaks at different times in different regions. When using age as a
covariate, I would at least suggest to include non-linear extensions
(quadratic, cubed) in the equations. The combination of these variables
explains a substantial portion of the variance.
The point about "unfair" spatial normalization I have, I believe,
mentioned before :) and it is quite easily illustrated in your example:
the 20-year-old will be much closer to the MNI template (where the
average age in the classical version was 27) than the 6-year-old, and
whill therefore experience less deformation when warped to it. This
deformation will correlate with age, and will therefore bias your
analysis. One way to address that would be to either create your own
template based on all subjects (if you have enough) or to take a
template which is "somewhere near the middle" so that, on average, the
deformation is not correlated with age anymore.
Not sure this helps you, though...
Marko
Jason Craggs wrote:
> Iroise,
>
> Thanks for the information. However, for the proposed analyses you will
> probably need to consider some additional factors, if you haven't already.
> The use of age as a covariate makes the assumption that changes in brain
> development (e.g., growth and functional complexity) are linear, which will
> likely need to be discussed amongst your group. For both sets of analyses,
> you are likely to run into the 'spatial normalization' problem. That is,
> children's brain should not be simply warped into adult brain space, they
> are not directly spatially comparable. That is, the size, shape and ratio of
> white/grey matter of structure X for a 6-year old should be very different
> from than that of a 20-year old. Consequently, careful consideration is
> needed to determine if a the covariate of age, in months, is sufficient to
> explain differences in brains if any other factor aside from 'getting older'
> is being investigated.
>
> These are just my thoughts/concerns, and I don't have a ready answer to the
> problem, but it's an issue that we are dealing with also (but on a 'back
> burner). I'd be happy to continue to discuss this and get additional
> feedback from the other users about how to deal with these types of issues.
>
> Cheers
> Jason
>
>
>
>
> On 9/17/10 10:08 AM, "Iroise Dumontheil"<[log in to unmask]> wrote:
>
>> Hi Jason,
>>
>> we are using this model both for VBM (Dartel) and fMRI data.
>>
>> With fMRI we run a Condition B - Condition A contrast from the first level
>> individual subject analyses and they are the images we use in the 2nd level
>> analysis.
>>
>> With VBM we use individual subjects' grey matter images, so I guess it means
>> there is only one level of analysis.
>>
>> But we'd like to run the same analyses on both types of data if possible.
>>
>> 'Developmental level' is in the model as an age (in months) covariate. So we
>> have both the testing time/round and age at testing in the model. Does this
>> make sense? (and is it correct?!).
>>
>> Iroise
>
--
=====================================================================
Marko Wilke (Dr.med./M.D.)
[log in to unmask]
Universitäts-Kinderklinik University Children's Hospital
Abt. III (Neuropädiatrie) Dept. III (Pediatric neurology)
Hoppe-Seyler-Str. 1, D - 72076 Tübingen
Tel.: (+49) 07071 29-83416 Fax: (+49) 07071 29-5473
=====================================================================
|