Matt,
On Tuesday 31 March 2009 22:11:12 Matt Glasser wrote:
> It is, of course, always prudent to evaluate the output of the eddy_correct
> script, I too have seen situations where it makes things worse.
If you have such a data at hand, could you try using "-cost normmi" option in
the flirt command (choosing normalized mutual information cost function
instead of the default correlation ration) in eddy current script to see if
the realignment would be any better?
Thanks,
Martin
>
> Peace,
>
> Matt.
> -----Original Message-----
> From: FSL - FMRIB's Software Library [mailto:[log in to unmask]] On Behalf
> Of Pablo Velasco
> Sent: Tuesday, March 31, 2009 1:18 PM
> To: [log in to unmask]
> Subject: Re: [FSL] Donating scripts for rotating bvecs after ecc
>
> Yes, Matt: in principle you could try to do a first pass with 6 parameters
> and apply those
> rotations to the gradient directions. The problem I see with it is that
> since the SNR of the
> diffusion-weighted images is so low, and the contrast is so different from
> the b=0 image,
> you would have to make sure the motion correction does the right thing
> (i.e., I wouldn't
> just script it and think that it will do the right thing).
>
> Regarding whether or not it will make a difference or not, it depends on
> how much your
> subject moved: I would bet that if your subject tipped his/her head a few
> degrees, it will
> make a significant difference in the fibertracking results (maybe for the
> FA it won't
> matter too much).
>
> -Pablo
>
> On Tue, 31 Mar 2009 12:45:28 -0500, Matt Glasser <[log in to unmask]> wrote:
> >Ah, I see the difficulty. Presumably one could run a 6 parameter motion
> >correction step, apply all angular transforms to the gradient table, and
> >then a 12 parameter eddy current correction where no transforms are
> > applied (my understanding of eddy currents is that they are mostly global
> > image skews and scales, or translations)? Unfortunately such processing
> > would take 2x as long and still probably not make any difference in the
> > end.
> >
> >Peace,
> >
> >Matt.
> >
> >-----Original Message-----
> >From: FSL - FMRIB's Software Library [mailto:[log in to unmask]] On Behalf
> >Of Pablo Velasco
> >Sent: Tuesday, March 31, 2009 12:03 PM
> >To: [log in to unmask]
> >Subject: Re: [FSL] Donating scripts for rotating bvecs after ecc
> >
> >Hi Matt,
> >
> >You are correct: the diffusion gradients are applied in the undistorted
> >space, so correcting
> >for the EPI distortions brings the images into "alignment" with the actual
> >directions along
> >which the diffusion gradient were applied. However, the problem is that
> >eddy_correct not
> >only corrects for eddy-currents and other EPI-related image distortions,
>
> but
>
> >also for
> >subject motion. It is the actual physical rotation of the head what calls
> >for a rotation of
> >the gradient directions, and there is no way to separate both effects
> >(image distortions
> >and rigid-body rotations) in eddy_correct.
> >
> >-Pablo
> >
> >On Tue, 31 Mar 2009 11:01:49 -0500, Matt Glasser <[log in to unmask]> wrote:
> >>Tim has told me in the past that the diffusion gradients are applied in
> >>undistorted space, and thus when you correct for EPI distortions you are
> >>actually making the data fit the applied gradients better. Not sure if
> >
> >that
> >
> >>applies in this situation, but it would seem to.
> >>
> >>Peace,
> >>
> >>Matt.
> >>
> >>-----Original Message-----
> >>From: FSL - FMRIB's Software Library [mailto:[log in to unmask]] On
> >> Behalf Of Pablo Velasco
> >>Sent: Tuesday, March 31, 2009 4:57 AM
> >>To: [log in to unmask]
> >>Subject: Re: [FSL] Donating scripts for rotating bvecs after ecc
> >>
> >>Dear Martin,
> >>
> >>If the subject moves, rotating the bvecs is the correct thing to do,
> >> since the rotated directions are those along which the diffusion
> >> encoding gradients were actually applied. In that sense, it can't hurt
> >> you.
> >>
> >>However, eddy_correct doesn't correct just for motion, but also for
> >>eddy-current image distortions. The distortion correction is an image
> >>artifact, not actual motion, so the diffusion gradients will still be
> >
> >acting
> >
> >>along the same direction with respect to the subject's brain. Therefore,
> >>you should only be correcting the bvecs for the motion component.
> >>Unfortunately, eddy_correct doesn't distinguish between them, so it is
> >>impossible to differentiate.
> >>
> >>Best,
> >>
> >>-Pablo
|