> the solution is much simpler, than you would expect. The description in
> Good et al. belongs to that, what SPM is doing internally, when you segment
> non-normalized images. When you answer the question 'Are they spatially
> normalised?-> no' , than SPM is doing a affine transformation to MNI-space,
> create the segments there and transform them back. So, you won't get an
> sn3d-file. (type 'help spm_segment' for further information)
>
> The next step is indeed the transformation of this GM segment to a GM
> template by using the nonlinear options. Only here you will get the
> sn3d-file, which you use for the original (T1-)image. So, this sn3d-file
> containes the full, nonlinear transformation from the original-space into
> the template-space.
>
> The further story is, that you have to segment this normalized (T1-)image a
> second time and the rest is statistics.
>
> May be, John has some further comments on that.
Not much (new) to add. The procedure is:
1) Segment the original images.
----------------------------
This involves an implicit registration of a template to the image. The
transformation that is estimated here is used to overlay the prior
probability maps, which assist the segmentation. Before doing this step,
you may need to reorient the images via Display. This is so that the affine
registration has better starting estimates. A search for the following
keywords:
reorient Display
at http://www.jiscmail.ac.uk/cgi-bin/wa.exe?S1=spm will give you all the hints
about this that you need.
2) Clean up the grey matter.
-------------------------
This is only for SPM99. In SPM2, this procedure is combined with the
segmentation. The seg1 and seg2 images are entered into the program, and
the result is a brain_*.img, which has values of 1 for brain, and 0 for
non-brain. The resulting brain_*.img is used to remove a few misclassified
voxels from the *_seg1.img file. This is done using ImCalc, selecting the
seg1_, seg2_, seg3_, abd brain_ images, entering an output filename, and the
following expression:
i1.*i4./(i1+i2+i3+eps)
3) Estimate warps from the GM.
---------------------------
The reason for the first segmentation and cleanup is so that a set of
spatial normalisation parameters can be estimated by matching GM with
GM. In SPM99, you would disable 'Mask brain when registering?', and
set the resolution for the spatially normalised images to be higher than
the current defaults. The template would be an image of grey mater, which
could be the one in the apriori directory, or it could be a "custom"
made template.
Apply these warps to the original T1 image. This gives you a spatially
normalised T1 image, which should have a resolution of about 1mm isotropic.
The original *_seg* and brain_* images could be deleted at this stage.
4) Segment the spatially normalised T1.
------------------------------------
Tell the segmentation routine that the image is spatially normalised, so
no affine registration is done.
5) Clean up the grey matter.
-------------------------
The same as step (2) above (only for SPM99).
6) "Modulate".
----------
Spatial normalisation expands and contracts some brain regions. Modulation
involves scaling by the amount of contraction, so that the total amount of
grey matter in the modulated GM remains the same as it would be in the
original images. In SPM99, this is done by a script that was sent to the
mailing list:
http://www.jiscmail.ac.uk/cgi-bin/wa.exe?A2=ind0007&L=spm&P=R16328&I=-1
7) Smooth.
-------
Best if you do this by about 12mm.
8) Stats.
------
The whole idea behind the pre-processing is that the t-tests are sensitised
to volumetric differences in the GM. Remember that your results could reflect
other differences among the images.
For stuff about experimental design, see:
http://www.jiscmail.ac.uk/cgi-bin/wa.exe?A2=ind0208&L=spm&P=R3815&I=-1
Your experimental design may include the use of "globals". See the following
for more info:
http://www.jiscmail.ac.uk/cgi-bin/wa.exe?A2=ind0305&L=spm&P=R170&I=-1
http://www.jiscmail.ac.uk/cgi-bin/wa.exe?A2=ind0302&L=spm&P=R54629&I=-1
http://www.jiscmail.ac.uk/cgi-bin/wa.exe?A2=ind0010&L=spm&P=R36678&I=-1
http://www.jiscmail.ac.uk/cgi-bin/wa.exe?A2=ind0304&L=spm&P=R19452&I=-1
http://www.jiscmail.ac.uk/cgi-bin/wa.exe?A2=ind0209&L=spm&P=R2575&I=-1
http://www.jiscmail.ac.uk/cgi-bin/wa.exe?A2=ind0202&L=spm&P=R6740&I=-1
For other miscellaneous bits and pieces about VBM, see:
http://www.jiscmail.ac.uk/cgi-bin/wa.exe?A2=ind0207&L=spm&P=R12913&I=-1
http://www.jiscmail.ac.uk/cgi-bin/wa.exe?A2=ind0305&L=spm&P=R7159&I=-1
For serial scans, you may want to look at:
http://www.jiscmail.ac.uk/cgi-bin/wa.exe?A2=ind0304&L=spm&P=R34932&I=-1
http://www.jiscmail.ac.uk/cgi-bin/wa.exe?A2=ind0304&L=spm&P=R33596&I=-1
http://www.jiscmail.ac.uk/cgi-bin/wa.exe?A2=ind0304&L=spm&P=R20082&I=-1
http://www.jiscmail.ac.uk/cgi-bin/wa.exe?A2=ind0304&L=spm&P=R18761&I=-1
http://www.jiscmail.ac.uk/cgi-bin/wa.exe?A2=ind0304&L=spm&P=R12412&I=-1
http://www.jiscmail.ac.uk/cgi-bin/wa.exe?A2=ind0207&L=spm&P=R23655&I=-1
http://www.jiscmail.ac.uk/cgi-bin/wa.exe?A2=ind0207&L=spm&P=R366&I=-1
For symmetry studies, see:
http://www.jiscmail.ac.uk/cgi-bin/wa.exe?A2=ind0204&L=spm&P=R9464&I=-1
http://www.jiscmail.ac.uk/cgi-bin/wa.exe?A2=ind0012&L=spm&P=R10670&I=-1
Best regards,
-John
> > If I first affine-normalize a person's MR image to the T1 template, write
> > out the resulting image, segment it, then nonlinearly normalize the
> > resulting GM image to the GM template, that produces 2 different
> > _sn3d.mat files.
> > The first represents affine registration from original brain image to
> > template, and has no nonlinear "Transform" entries.
> > The second has nonlinear info present but of course has (nearly) an
> > identity matrix stored in "Affine".
> > If I then try to "modulate" the fully normalized GM image i.e. multiply
> > each voxel by the Jacobian determinant, I have to choose one of the
> > _sn3d.mat files, and neither by itself is meaningful.
> >
> > I must not understand what is meant by "native space", since it appears
> > to apply to affine-normalized images rather than to raw images. In what I
> > would call "native space," the person's MR image and the ICBM template
> > are not at all registered, so the a priori maps are useless for
> > segmentation.
> >
> > When people on this listserv have said "apply the deformations from GM ->
> > GM template to the original MR image", how is this possible to do in SPM?
> > The second _sn3d.mat file can't apply to the original MR image since it's
> > missing the affine info. Is there a way of composing the info in the 2
> > _sn3d.mat files, analogous to multiplying 4x4 matrices, that would result
> > in a single _sn3d.mat file that represents the full normalization from
> > original MRI to template?
> >
> > Is there any written description of the actual SPM99 functions used to do
> > this? I have spent a day or so searching the SPM archives and I found
> > things piecemeal only. Has anyone made a "VBM for dummies" help guide?
--
Dr John Ashburner.
Functional Imaging Lab., 12 Queen Square, London WC1N 3BG, UK.
tel: +44 (0)20 78337491 or +44 (0)20 78373611 x4381
fax: +44 (0)20 78131420 http://www.fil.ion.ucl.ac.uk/~john
|
