Dear Chris,
I think that the simplest way around this is to normalize the structural MRI
image for each subject (assuming you have one). Before doing that, you
coregister the structural MRI to the mean functional image (moving the
structural image into 'functional' space), then normalize the structural image
to the (T1 presumably) template (you don't need to use the r* image created by
coregistration, as the coregistration transformation is added to the structural
MRI .mat file). The resultant *sn3d.mat file can be used to normalize the
individual functional scans.
Cheers,
Alex.
> MIME-version: 1.0
> Content-transfer-encoding: 7BIT
> Date: Mon, 4 Dec 2000 16:18:22 +0100
> From: [log in to unmask]
> Subject: normalization of few slices
> To: [log in to unmask]
>
> hi spm
>
> we have been experimenting with the normalization tool for fMRI in situations
> where there are a reduced number of slices available. when we
> attempt to normalize 4 realigned slices with 7mm voxels on the z
> axis, the output images invariably result rotated by several degrees,
> and the modelled activations, although roughly in the expected area,
> seem to be out of place. This occurs despite the fact that we have
> used "header edit" and "check reg" to ensure that the origin and
> image dimensions are equivalent (by visual inspection) to those of the EPI.img
> template. The error message: "warning: field of view is too small to
> allow linear registration" (or something similar) appears during the
> normalization process.
>
> Other than taking more slices with better resolution, is there a way
> to correct this error? We have experimented using a bounding box
> appropriate to the acquired slices, but the images remain rotated
> within the confines of the bounding box. We also cut the internet SPM
> example images (the fMRI auditory paradigm) to
> roughly like ours (14 slices x 2mm voxels, including AC-PC slice) and tried to
normalize
> these, but we get the same error message, and the same rotation of
> the output images.
>
> We would like to know what
> is the minimum number of slices necessary to acheive adequate
> normalization with SPM, with the AC-PC origin included for
> appropriate registration with the template image. Is there a
> reasonable minimum? Or is it that SPM should theoretically be able
> to normalize any number of slices, and we are simply doing something
> wrong? We have looked at past correspondence on this list, but have
> not found any references to this problem.
>
> Thanks in advance for any help.
> Chris Summerfield
>
> dept psiquiatria i psicobiologia clinica
> universitat de barcelona
|