Dear list,
I would like to clarify my thoughts as to the space in which the
Jacobians produced by DARTEL are in, the issue being interoperation
with MNI space. In my data, they seem to be in the space of
Template_6.nii. This is what one would expect since in the step to
produce them, one only needs to provide the flow fields u_.*
(registration to MNI space would require access to Template_6.nii to
compute the transform).
I therefore inferred one needs to apply the tranform in
Template_6_2mni.mat to the Jacobian images to bring them in MNI space
(or the inverse to bring anything else from MNI to the space of the
Jacobian images).
Looking at the Mailing list, this issue comes up in a mail in 2009
(somewhat puzzingly in only one message, at least as far as I could
find). At the time, it was discussed use of the "sn.mat file of the
affine transform from DARTEL-Template to MNI", but I do not have that
file now (I am not clear if I can get one or the process has changed
since then).
In that thread, there is an interesting discussion about the right
strategies for interpolating Jacobian images. I am thinking this might
be a reason why these images are not brought into MNI space by default
in the toolbox. I would be personally happy with a preliminary
log-transform since this is the way one would go about these data in a
linear model (which implies interpolation). This would be at least
conceptually consistent. By the same token, one would log-transform
before applying any smoothing step.
In the list, Jacobians are occasionally discussed in connection with
the creation of modulated warped GM maps. An issue coming up in these
few threads is whether the Jacobian images include any affine
transform for global head size, or only the nonlinear component
(answer: it is the nonlinear component).
I am not sure if this applies to the Jacobian images produced by the
DARTEL toolbox. More precisely, the issue is the extent to which the
mean within-brain value from the Jacobian images (some kind of mean,
for example the geometric mean) would be expected to correlate with
the brain volume computed from the segmented images. In my data, it
does, implying that the Jacobian includes global volume changes. Am I
wrong here? (By the way, the best results are with the arithmetic, not
the geometric mean, suggesting that perhaps one does not need the
log-transform mentioned above).
Best wishes,
Roberto Viviani
University of Innsbruck
|