Hi Daniel,
> We are attempting to create a average brain template in SPM2 for the
> subjects in our study (~30 children), following a method similar to that
> of Marko Wilke et al (2002).
Well then perhaps I should venture an answer :) Note, however, that some
of the issues my be more or less prominent in spm2/5 versus spm99 which
I used.
> We plan to use this for coregistration and
> normalization of our functional data (rigid body registration from
> individual T1 to individual EPI, then regular affine/non-linear
> transformation from individual T1 to the new template T1, applying the
> parameters to the EPI).
OK, although you could think about segmenting the T1 and matching the GM
partition to the EPI, which will reduce the influence of extracranial
tissue. In that case, you don't have to reslice, as the transformation
from the coregistration step will be taken into account when
normalizing. You could also just use my old pediatric GM partition as
the target.
> 1) 12-parameter affine transformation (with brain mask, as used by
> Wilke) from individual high-resolution T1 images to the MNI T1
> template.
Do you have a large age range in your group? If so, 30 subjects may not
be enough; if not, you could try omitting the brain mask, especially if
you use the segmented GM.
> 2) Realigning using a coregister and reslice, creating a mean image of
> all the individual normalized anatomicals, and smoothing that to create
> the template.
OK.
> Wilke also discussed using an intensity nonuniformity correction.
I did :) Actually, you should always do that for high-field data (which
mine was) and you should explore the effects for 1.5 T data. The bias
correction will remove slowly varying intensity fluctuations which are a
combination of B0 inhomogeneity and the effect of the head in the coil,
so they will be unique (although similar) in each subject. Therefore, as
you don't want the normalization algorithm matching the inhomogeneities
but the brains, I would remove them.
> Also, is it problematic for brain averaging if the
> mean intensities of the images are not identical?
Another issue you avoid by using GM ;)
> Another thing is that we would like to be able to report the results
> in Talairach space.
I must say I am biased in this regard but I never understood why this is
considered so important. There are several reasons why I humbly think
that this is neither necessary nor meaningful in your case: one, spm
uses MNI space which is not Talairach space, and even using Matthew's
script only gets you close; two, you are creating your own template
which will be even further away from a single female 60-year-old brain
brain that spent some time in formaline and had its cerebellum chopped
off before being selected as a reference; three, considering the moderen
studies on neuroanatomical variability of even primary visual cortex and
other "distinct" brain areas (see the great work of Amunts, Zilles and,
more recently, Eickhoff), I have not much faith in fixed coordinates
meaning the same thing across individuals; and four, you are dealing
with children, whose brain undergoes tremendous changes as part of
maturation, so I am even less inclined to attach a meaning to such
numbers in these circumstances.
> This could be as simple as an affine/non-linear
> transform from the new template to the MNI T1 template
It could be as simple as taking a neuroanatomiy book and identifying the
regions :)
All the best, and sorry if I preached :)
Marko
--
=====================================================================
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
=====================================================================
|