Dear John and Darren,
Thank you very much for your suggestions. I have been working with the
manual set origin of AC in MEDx, and using an EPI Template.
It seems to me as we only 12 coronal slices spm in MEDx is unable to have
sufficient data to proceed with an optimal transformation.
I have tried reducing the Number of X Basis Function, (or Y or Z or all of
them at once) and only managed to have minimal effect in preventing the
tilting of the coronal slices.
I was unable to find the starting estimates of the affine transformation.
Is spm starting estimates are the anterior upper corner of the field of
view (i,e the 12 coronal slices together) or it defaults to what
corresponds to the EPI T2 Template (which would have then larger
dimensions? Why is the transformation easier on 12 transverse slices then
on coronal ones?
We are trying to dowload the spm99b and try it through it.
Any additional comments would be greatly appreciated
Ziad Nahas, MD
--On Thu, Dec 2, 1999 10:57 AM +0000 John Ashburner
<[log in to unmask]> wrote:
> I've never used the MEDX implementation of SPM96, so I can't really say
> anything for definite. All I know is that spatial normalisation is
> more robust under SPM99b than under SPM96.
> There could be any number of reasons why things are not working properly:
> 1) What themplate are you matching to? A bold image will not register
> to a template image with a T1 contrast. Is there a template image
> under MEDX of a similar contrast to your own images?
> 2) Starting estimates for the affine registration may not be so good. If
> you have coronal images, then the spatial normalisation needs to know
> that they are coronal, and I would guess that this is done in MEDX
> by setting some starting estimates. Setting the origin field is also
> effectively partly setting the starting estimates. The affine
> registration begins with an initial estimate for the transformation,
> checks the sum of squared difference between the image and template
> and changes the estimate so that it should reduce the sum of squared
> difference. This continues until the sum of squared differences
> no longer decreases. If the starting estimates are poor, then the
> registration is likely to get cought in a local minimum.
> 3) Problems can arise when the images have a relatively small field of
> view, as there may not be enough information in the images to obtain
> a good match.
> 4) A few others that I can't currently remember. MEDX spatial
> normalisation is based on software that I wrote a long time ago. The
> same procedure using SPM99b would be much more robust....
> I would suggest that you initially work with only an affine registration.
> You may need to reduce the number of parameters from the default 12 for
> the registration to work well.
> Once you have the affine bit working OK, then try out different numbers of
> basis functions.
> Good luck,
> | > I have a set of 12 coronal slices of BOLD fMRI (covering the frontal
> lobes | > and extending posteriorly to the motor cortex). The Anterior
> Commissure | > (AC) point is centered at Slice #7. I am trying to
> transform these images | > into Tailarach space using spm (part of MEDx
> | >
> | > I am running into problems with the final position of the AC point
> | > ("origin") of my spatially normalized images. Even if I define the
> outbox | > boundaries accordingly with y dimensions extending the
> equivalent of 12 | > slices and centered around AC point, I find that
> after the images are | > normalized, the AC point is consistently shifted
> posteriorly 16 mm from | > where I had set it prior to normalization.
> What were coronal slices | > (acquired 90 degrees to the AC PC line) are
> now looking diagonal (tilted | > forward). The dimensions defined by the
> boundary box are correct. | >
> | >
> | > Has someone encountered such a problem, or should I assume that since
> the | > initial data does not cover the whole brain, then I should not
> expect an | > exact transformation?
> | >
> | > How to explain the tilting forward (a rotation of 15-20 degrees around
> | > the X axis)? I suspect that this is what is making the final AC point
> | > seems more posterior.
> | > Should I try to modify the Number Basis X (or X, or Z) Functions ?
> Several | > different combinations did not seems to make much difference.
> | > Or is the problem with Affine Parameters and Transforms?
Ziad Nahas, M.D.
Medical Director, Brain Stimulation Laboratory