> If I want to coregister a neurological image (say an EPI) to a radiological
> image (say a T1), can I just flip the latter by zooming in x by -1 and
> reorienting? (Ie, will that method of flipping be seen by the
> coregistration program?)
That would work.
>
> Also, is the coregistration algorithm symmetric in its action, ie is the
> linear transform that coregisters A to B the inverse of that that
> coregisters B to A?
The original SPM99 versions (the default version and the MI version) were
both asymmetric in their behaviour, but the newer MI version (from the
updates directory in our ftp site) should behave in a more symmetric way,
giving more internally consistant results.
The original MI code sampled one of the images on a regular lattice
without needing to interpolate it, whereas the other image was interpolated.
This meant that the optimisation may be more prone to local minima if the
images were registered one way rather than the other. The MI cost function
itself is perfectly symmetric, but the generation of the joint histograms
was not.
The default coregistration algorithm in SPM99 involved simultaneously
matching GM with GM and WM with WM. This was coded so that only the
GM and WM gradients of one of the images featured in the computation.
In the new version of the MI registration, I have dithered the sampling
of the reference image, which seems to reduce the local minima problem.
It still isn't 100% symmetric, but it is close. Slightly more symmetry
can be achieved by commenting out lines 69-76 and uncommenting lines
78-94 of spm_hist2.c 2.9.
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
|