Hi Shary, everyone,
[log in to unmask] wrote:
> I cannot find answer to my question regarding follwup studies in SPM5.
> Do we need any addups for TBM under SPM5?
I think everything you need is there in SPM5. The basic procedure of
Whitford et al '06 [1] (based on Kipps et al '05 [2], and basically
the same idea as Scahill et al '02 [3]. See also Chetalat et al '05
[4]) is to use high dimensional warping within-subject (over time)
and SPM's usual spatial normalisation to the template.
Each Jacobian image from each within-subject warping is transformed
with the same normalisation parameters derived to map the
corresponding within-subject target (could be baseline or follow-up)
to the template. In [3] the warped Jacobians were separated into
expansion and contraction; in [1,2] they were used to modulate
(multiply) the normalised tissue segmentations.
[1] http://dx.doi.org/10.1016/j.neuroimage.2006.03.041
[2] http://dx.doi.org/10.1136/jnnp.2004.047993
[3] http://www.pnas.org/cgi/doi/10.1073/pnas.052587399
[4] http://dx.doi.org/10.1016/j.neuroimage.2005.05.015
In practical terms, perform the usual normalisation with SPM5's
Unified Segmentation. Use the HDW toolbox for the high dimensional
within-subject warping, as follows: (Shary, hopefully, this answers
your further email)
In the menubar of the large graphics window, select:
Tasks -> Tools -> High-Dimensional Warping
you should then see a familiar-style GUI to change the bias and
warping options (I can't help you decide what values to use here, but
I'd guess that the defaults are okay), and select the images. Just add
"new subject" as often as you wish, selecting reference (aka target)
and moving (aka source) images, then (after optionally saving the
job), click run and sit back for quite some time...
When that finishes you will have, in the same directory as the
moving/source images, the deformation fields (prefixed with 'y_') and
their Jacobian determinant images (prefixed with 'jy_').
You then need to transform these jy_ images to standard space. Use
SPM's "Normalise Write" GUI, select the saved blah_seg_sn.mat file
that corresponds to the reference/target image that you picked, and
the jy_blah image. [see also P.S. below...]
Here, you have the option of modulating the Jacobian image using the
template transformation or not. My view is that you should, but that
you should then be careful that if following the approach of [1,2] you
multiply by the unmodulated warped segmentations (e.g. wc1blah);
alternatively, you could transform without modulation, and then
multiply by the modulated (mwc1blah) segmentations.
I think it's important to avoid "double-counting" the modulation by
modulating when transforming and then multiplying by modulated images.
And I think it's important to account for both the within-subject
volume changes (in the Jacobian image) and the to-template volume
changes (with modulation), though I am open to argument here.
Hopefully, Shary, this answers your second question about whether SPM
has already modulated.
To multiply the warped Jacobian image by the segmentation, something
like this should be appropriate (untested, watch for typos!):
flags = {0, 0, spm_type('float32'), 1};
files = char({'mwjy_blah.nii', 'wc1blah.nii'});
spm_imcalc_ui(files, 'mwjy_c1_blah.nii', 'i1.*i2', flags);
Finally, smooth mwjy_c1_blah.nii, and you should be ready to perform
stats on it. Shout if some/all of that doesn't make sense or doesn't
seem to work. Good luck.
Lastly, it seems to me that this is all a bit tedious, and I am
therefore currently working on a simpler quicker-to-use GUI for this,
watch this space... (and please let me know if you, the reader, are
also working on something similar -- it would be nice not to duplicate
effort too much!)
Best of luck,
Ged.
P.S. I think an alternative to transforming the Jacobian images would
be to use "Tasks->Utils->Deformations" and make a "composition" of the
"imported sn" file and the y_ transformation file, then saving a new
Jacobian image from this (if possible). But I have not investigated
this approach. In any case, I think what I describe above matches
what's in the papers cited. Subject to me getting it wrong...
|