> *John Asburner wrote:
> >It was removed as I consider it to be too dangerous.  Too many people used
> >it to change the origin or voxel sizes of their .hdr files, and wondered
> > why these appeared to remain the same (because of the .mat file).  This
> > is likely to cause even more problems with SPM2b, as the data are (for
> > most people) implicitly flipped.
>
> When the AC line is set at the origin in the HDR file as the first step in
> preprocessing, i.e. when no .mat files have been created, no danger is
> present.

If you use the DICOM toolbox pulldown, then .mat files are created for the
original images.  These encode the orientation of the images, so data
acquired in even saggital or coronal orientations will be viewed in SPM2b
as transverse.

> In general, i thought that it should be wise to set the origin of all
> images to obtain similar starting points for coreg-realign-normalize steps,
> whose underlying minimilization-algorithms could otherwise get stuck in
> local minima.

For those of you who are just starting out with SPM, the registration
routines simply do a local optimisation.  This involves starting
with a guess about how to spatially transform one of the images, and
updating the guess at each iteration until the measure of how well the
images are matched stops improving.  If the initial guess is not very
good, then this can go wrong.  This is why the images should be roughly
aligned beforehand.  The DICOM toolbox usually produces .mat files that
are close enough for SPM2b to use as starting estimates for most
registration procedures.

An alternative strategy would be to use a glogal search method, which
involves trying all combinations of possible parameters.  A full global
search would be extremely slow, but it is possible to use a good
compromise between a local optimisation and a global search.  When
properly implemented, such algorithms are able to always find a good
solution no-matter what the starting estimates are.  See e.g.
        http://www.fmrib.ox.ac.uk/analysis/research/flirt/

>
> On the other hand, mathematically it is similar to obtain starting points
> for preprocessing steps by means of a translation, which is saved in a .mat
> file, and hence setting the AC point to the origin is not longer necessary.
>
> It is kind of naieve to expect that everyone is knowing what he or she's
> doing, and I understand the dangers of changing origins in the header file
> when one doesnt know what is computationally going on.
>
> So my main question is mainly a practical one: is there still the need to
> alter the origin in the Analyze-header at the AC line in spm2b? Or can one
> now roughly pre-register the images with a translation matrix and leave the
> header file unchanged?

This can be done via their .mat files.  If you have lots of images of different
subjects that were all acquired roughly in the same position and orientation,
then you can display one image using the <Display> button, work out how best to
rotate and translate the image, and then apply the same transformation to
all of them using the "reorient..." button.  Not only would this put the
origins in roughly the correct place, but it also allows rotations to be
included.

By the way, the initial affine registration part of the spatial normalisation
is now more robust to poorer starting estimates (when a single template image
is used).  This means that the origin does not need to be as accurate as it
did in SPM99.

Best regards,
-John

--
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