Hi Doug et al,
interesting point, Doug. I think Tim was referring to the
interpolation/reslicing required for motion correction and how it might
smooth the volumes interpolated.
I don't want to totally confuse the thread, but what if the motion
correction is not applied...
Is it possible to just estimate the signal change due to motion without
correcting it, so you just use it as a covariate in analysis? I think this
can be done with melodic (although you have to be careful to accurately
identify components related to motion). I guess, in this case, it wouldn't
matter whether it was one run or many runs.
Note that this covariate should be the amount of signal change related to
motion, not the geometric motion correction matrix. The motion correction
matrix just contains translations/rotations, not necessarily a perfect model
of the actual signal change at each voxel. I've never really understood
using "motion correction parameters" as covariates if the motion correction
is already done, because they would no longer represent the data time
series. (By "motion correction parameters" I mean the motion correction
matrix of rotations/translations/scaling.) Furthermore, even if they are
used as covariates on the raw time-series, they are not a model of the
signal change due to motion, but rather they are an abstract geometrical
derivation from the signal change due to motion (containing
translations/rotations).
I hope this makes sense.
Take care, Darren
----- Original Message -----
From: "Doug Greve" <[log in to unmask]>
To: <[log in to unmask]>
Sent: Saturday, December 21, 2002 4:06 AM
Subject: Re: mcflirt for multiple runs
> Has anyone ever looked at how much this really affects the smoothness?
> In principle, if someone moved by exactly one voxel, the smoothness
> would not change because the volume would just be shifted. Ie, what will
> really affect the smoothness is the modulus of the shift (relative to
> the voxel size) and not the magnitude of the shift itself, and it seems
> reasonable that the modulus between runs might not be that different
> than the modulus within runs. Does this make sense?
>
> doug
>
> Tim Behrens wrote:
> >
> > Hi - just to add a couple of points.
> > As Christian suggests, the motion correction will be less accurate, but
it
> > will also introduce another problem. That is, the different runs will be
> > shifted/rotated by dramatically different amounts in order to align
them,
> > and therefore will be subject to very different amounts of interpolation
> > related smoothness. Hence the smoothness will vary through time in a
step
> > like fashion (being roughly stationary through a single run but
changeing
> > between runs). As all the stats in Feat assume that the smoothness is
> > stationary, this will introduce errors into the final stats ...
> >
> > cheers
> > Tim
> >
> > On Fri, 20 Dec 2002, Christian Beckmann wrote:
> >
> > > Hi,
> > >
> > > I completely agree with Joe - concatenating the data is probably not
> > > what you want since it will result in suboptimal motion-correction.
The
> > > reason is that (for 'typical' fmri experiments) the amount of motion
> > > within one run is considerably smaller than between runs. When data is
> > > concatenated mcflirt will register all the low-res fmri volumes to one
> > > reference image (by default the middle one) which for all volumes from
> > > the different runs other than the one that the ref. image belongs to
> > > will make motion-correction less accurate.
> > > If you really want to concatenate the data you'll probably have to do
> > > some clever shell scripting. Usin avwroi instead of avwsplit will make
> > > this a bit easier for you though...
> > >
> > > cheers,
> > > Christian
> > >
> > > Joe Devlin wrote:
> > > > Hi Russ,
> > > >
> > > >> hi all - I'm trying to motion-correct fMRI data collected over
multiple
> > > >> runs, and it's not clear how to do this with mcflirt at the command
> > > >> line. Obviously I could do this using some combination of avwmerge
and
> > > >> avwsplit but that seems rather awkward, particularly because I
would
> > > >> have to reconstruct the numbering within the individual runs after
> > > >> avwsplitting the merged motion corrected image. Any suggestions
would
> > > >> be greatly appreciated.
> > > >
> > > >
> > > > I know this isn't really an answer but I'm wondering why you want to
motion
> > > > correct more than one run at a time? I typically motion-correct the
runs
> > > > individually, analyse them individually at the first level, and
enter them
> > > > into the second level as repeated measures now that Feat5 offers
that
> > > > option. I've found it works pretty well.
> > > >
> > > > Cheers,
> > > > Joe
> > > >
> > > > Joseph Devlin, Ph. D.
> > > > FMRIB, Dept. of Clinical Neurology
> > > > University of Oxford
> > > > John Radcliffe Hospital
> > > > Headley Way, Headington
> > > > Oxford OX3 9DU, U.K.
> > > > Phone: +44 (0)1865 222 738
> > > > Fax: +44 (0)1865 222 717
> > > > Email: [log in to unmask]
> > > >
> > > >
> > >
> > > --
> > > Christian F. Beckmann
> > > Address: Oxford University Centre for Functional
> > > Magnetic Resonance Imaging of the Brain,
> > > John Radcliffe Hospital, Headington, Oxford OX3 9DU, UK
> > > Email: [log in to unmask]
http://www.fmrib.ox.ac.uk/~beckmann/
> > > Phone: +44(0)1865 222782 Fax: +44(0)1865 222717 Mob: +44(0)7811
189123
> > >
> >
> > --
>
> --------------------------------------------------------------------------
-----
> > Tim Behrens
> > Centre for Functional MRI of the Brain
> > The John Radcliffe Hospital
> > Headley Way Oxford OX3 9DU
> > Oxford University
> > Work 01865 222782
> > Mobile 07980 884537
>
> --------------------------------------------------------------------------
-----
>
> --
> Douglas N. Greve, Ph.D.
> MGH-NMR Center
> [log in to unmask]
> Phone Number: 617-724-2358
> Fax: 617-726-7422
>
|