Dear Jonas,
I didn't realise that both the z *and* y directions had
small FOVs. Anyway, the problem then is definitely related
to the 8mm and 4mm sampling sizes assumed by mcflirt. With
your resolution and FOV this turns your dataset into a 22x5x3
voxel dataset at 8mm which isn't useful.
The way around it is to artificially change the original voxel
sizes to 4x3.92x4 mm and then run mcflirt. I did this by using
avwedithd to change the voxel size and then ran mcflirt and
results look good to me. The movie mode in fslview shows
very little change between beginning and end, unlike the
original data which shows a slow trend that "snaps back"
when the movie wraps around from last to first image.
I also checked the matrices and they are definitely not
non-zero.
All the best,
Mark
On 9 Dec 2004, at 10:42, Mark Jenkinson wrote:
> From: Jonas Larsson <[log in to unmask]>
>
> I have been using mcflirt with reasonable success on "standard" FMRI
> data
> (3x3x3 mm voxels, around 20 slices covering the occipital or parietal
> lobes). Recently, we have tried it on a data set with smaller voxels
> (1x1x1 mm) and a limited field of view. Despite many attempts, I have
> not
> succeeded in getting mcflirt to do any motion correction at all (the
> aligned output is identical to the input). I was wondering if you have
> any
> pointers as to what could be going wrong. The data is characterized by
> -small voxels
> -limited field of view (about 20 slices, i.e. 2 cm)
> -very poor gray-white contrast (though obviously the brain/non-brain
> contrast is good).
> -voxel values between 0-1 (I assume this has no bearing on the result
> but
> just in case)
>
> I have tried to modify the voxel size, varying the cost function,
> smoothing parameter, scaling (not sure what this parameter does?),
> using
> edge and/or gradient based estimation, but as far as I can tell mcflirt
> doesn't make any corrections whatsoever. I'm fairly confident this is
> not
> due to any typo or improper command-line calls, so there must be
> something
> special about this data set that mcflirt doesn't like. I would very
> much
> appreciate if you could suggest a strategy that might work.
>
> Thank you!
>
> best regards,
>
> Jonas Larsson
> Center for Neural Science/Dept of Psychology
> New York University
|