Dear Haakon,
Perhaps your EPI images are too distorted to register to T1 at all, e.g.
the coregistration algorithm can never find a good rigid body match
because a 'good' match simply doesn't exist?
Especially with long readout trains or high field strengths EPI images
can be spatially distorted considerably, compared to T1 acquisition, due
to susceptibility artifacts. In that case, you could either try to
reduce your EPI readout train by using a higher bandwith or deploy
parallel acquisition (SENSE, GRAPPA).
Alternatively, you could try to normalize your EPI data directly to an
EPI template when using classical normalization. Then the non-linear
part of the transformation might compensate for the typical spatial
distortion. Or, when your EPI images are of sufficiently high contrast,
try to use unified segmentation on them (I only manage to do that with
some success for hi-resolution EPI, eg 2x2x2 mm or below).
Finally, when you can acquire phase maps on your scanner, you could try
to unwarp your EPI images using the phase maps in SPM5/SPM8 unwarping
(0-th order unwarping). When I'm informed well this will presently only
work for axial EPI acquisition with the phase encoding AP or PA.
Good luck,
Bas
--------------------------------------------------
Dr. S.F.W. Neggers
Division of Brain Research
Rudolf Magnus Institute for Neuroscience
Utrecht University Medical Center
Visiting : Heidelberglaan 100, 3584 CX Utrecht
Room B.01.1.03
Mail : Huispost B.01.206, P.O. Box 85500
3508 GA Utrecht, the Netherlands
Tel : +31 (0)88 7559609
Fax : +31 (0)88 7555443
E-mail : [log in to unmask]
Web : http://www.linkedin.com/pub/3/774/590
--------------------------------------------------
-----Oorspronkelijk bericht-----
Van: SPM (Statistical Parametric Mapping) [mailto:[log in to unmask]]
Namens Haakon Engen
Verzonden: maandag 19 januari 2009 11:15
Aan: [log in to unmask]
Onderwerp: [SPM] Coreg problems
Dear SPM'ers,
I realize questions as basic as how to get coregistration to work are
a few notches below what usually gets answered on this list, but I'm
really at my wits end here.
The problem is that 4 (out of 27) of my subjects plainly refuses to
have their T1 images coregistered to the mean EPI image, or any other
EPI image for that matter. The experiment totals 4 runs, and I've tried
to coregister the T1 to all of the mean images post- realignment, as
well as multiple ordinary, realigned, images. I've also tried to
coregister the EPI's to the T1, once again without success.
I've attached the coreg output files for all 4 subjects. As you can
see they all seem to be out of alignment in the Z- dimension, with the
T1 consisently lower than the EPI's, while otherwise seemingly in OK
registration.
I've tried fiddling around with the settings, changing objective
functions, seperation and tolerances without much success. I've also
tried using the coregister and reslice function, with interpolations
from trilinear up to 7th degree spline- again without success. This is
rather troublesome as this means that the normalization parameters
calculated by the unified segementation procedure are similarly shifted,
which results in "extracranial" activations, when overlaid on the
glassbrain.
Can anyone give me a clue as to why this happens, and how to fix or
work around it?
I'm on a Vista x64 workstation using Matlab R2007b, and SPM 8b r2598.
Kind regards,
Haakon G. Engen
Research assistant
University of Oslo
