It isn't just signal compression.  If you compare a distorted to undistorted
dataset you will see that there are geometric effects outside of the region
with dephasing, compression, and rarefaction.  The correct way to deal with
this is to undistort the data and then use only 6 parameters to register the
BOLD to the T1.  The way your processing stream probably works is that it
uses 12 parameters and probably partially compensates for the distortion,
however your registration in other brain areas will also be off because you
are doing a linear correction of a nonlinear phenomenon.  

It matters most if you care about having exact alignment between your BOLD
data and T1, for example if you are going to do surface analysis.  

Peace,

Matt.

-----Original Message-----
From: FSL - FMRIB's Software Library [mailto:[log in to unmask]] On Behalf
Of Michael Harms
Sent: Thursday, October 14, 2010 4:53 PM
To: [log in to unmask]
Subject: Re: [FSL] effectiveness of field correction as a function of phase
encoding direction

Thanks for the tutorial.  So, as a practical matter, does one get
anywhere near the same benefit in correcting BOLD problems in the
frontal cortex when applying field map correction to data collected with
A>>P phase encoding, vs. P>>A phase encoding?

We have a bunch of BOLD data collected with A>>P phase encoding on a
Siemens Trio, for which we also collected a field map in that session.
However, we have yet to work the field map correction into our actual
processing protocol. Now I'm wondering if we'll really get much benefit
of applying a field map correction to that data given the rather
pronounced signal compression that you can get in frontal cortex when
using A>>P phase encoding.

thanks,
-MH

On Thu, 2010-10-14 at 17:24 +0100, Jesper Andersson wrote:
> Dear Michael,
> 
> > Thanks Matt.  But I'm still a little bit puzzled about the "limits" of
> > field mapping correction.  You wrote that it is good at "geometrical
> > distortion correction, but less good at correction of intensities from
> > signal stretching or compression".  Aren't those just variations  
> > along a
> > continuum of the same underlying phenomenon?
> 
> partially they are, and partially they are not. For DTI (spin-echo  
> EPI) the intensity distortions have a simple relationship to the  
> geometric distortions (e.g. if you take two voxels worth of signal and  
> squeeze into on voxel you will observe double the intensity).
> 
> For those scans it would in principle be possible to exactly restore  
> the true images. BUT, because the images are discrete we end up in  
> situations where the intensity from several "object voxels" is  
> squeezed into one image voxel, and in those cases we do not have the  
> information that we need. It would be equivalent to try to deduce a  
> sample from the sample mean. A way around that, as has been suggested  
> in this thread, is to collect data with different traversals of k- 
> space (e.g. phase-encoding ant-post and post-ant) which means that for  
> each area that has been squashed in one acquisition the corresponding  
> area will have been stretched in the other acquisition. The  
> information from the two scans can then be combined to yield a  
> reasonable estimate of the true intensity values. We currently have a  
> internal release of software for doing that, and will release it as a  
> part of FSL when we are happy with how it works.
> 
> For fmri (gradient echo EPI) the situation is much more tricky. The  
> stretching/compression of intensities is still present, but we now  
> also have signal loss due to de-phasing within the voxel. This de- 
> phasing is a highly non-linear function of the partial derivatives of  
> the field in the phase-encode and slice-select directions and there is  
> currently no method available for estimating it.
> 
> Hope this helped clarifying things a little.
> 
> Jesper
> 
> 
> 
> 
> >
> > thanks,
> > -MH
> >
> > On Wed, 2010-10-13 at 11:44 -0500, Matt Glasser wrote:
> >> Hi Michael,
> >>
> >> It is certainly true that once signal from multiple voxels has been
> >> compressed into a single voxel, it is hard to put it back with a  
> >> field map
> >> alone.  For diffusion data, phase up / phase down corrections have  
> >> a better
> >> chance of dealing with this issue, as you have the data both  
> >> compressed and
> >> rarified and can figure out what the undistorted image would have  
> >> looked
> >> like.  It is not possible to do this with BOLD data, as you can't  
> >> acquire a
> >> run and then a second run with the phase encode direction flipped  
> >> and unwarp
> >> and average them (which is what happens with the diffusion data).   
> >> That
> >> being said, perhaps if you acquired the slices of phase up and  
> >> phase down
> >> interleaved with one of these faster TR sequences you could pull  
> >> off phase
> >> up / phase down correction on BOLD data (though you would want to  
> >> run slice
> >> timing correction first).
> >>
> >> If I recall correctly from a conversation Jesper and I had, R/L (or  
> >> L/R)
> >> phase encode direction actually have the least severe distortions,  
> >> but
> >> people don't like them because the distortions are not  
> >> symmetrical.  If you
> >> were most interested in stuff in the orbitofrontal region, it might  
> >> make
> >> sense to use a phase encode that stretches rather than compresses  
> >> this
> >> region (though of course other regions would be compressed instead).
> >>
> >> The field map is good at geometrical distortion correction, but  
> >> less good at
> >> correction of intensities from signal stretching or compression (or  
> >> outright
> >> loss).
> >>
> >> Peace,
> >>
> >> Matt.
> >>
> >> -----Original Message-----
> >> From: FSL - FMRIB's Software Library [mailto:[log in to unmask]] On  
> >> Behalf
> >> Of Michael Harms
> >> Sent: Wednesday, October 13, 2010 11:25 AM
> >> To: [log in to unmask]
> >> Subject: [FSL] effectiveness of field correction as a function of  
> >> phase
> >> encoding direction
> >>
> >> Hello,
> >> Is the effectiveness of field correction for compensating for B0
> >> distortions at least partially dependent on the phase encoding  
> >> direction
> >> used to acquire the BOLD or DTI data that is to be corrected?
> >>
> >> To use Siemen's parameter lingo, a "P>>A" phase encoding direction
> >> stretches signal from frontal cortex outward (anteriorly, into empty
> >> space), whereas "A>>P" (which is unfortunately Siemen's default)
> >> compresses signal from frontal cortex (i.e., moves frontal signal
> >> posteriorly, onto existing brain).  It seems that once signal from
> >> multiple voxels is compressed into 1 voxel that not much correction
> >> would be possible...  Is that indeed the case?
> >>
> >> thanks,
> >> -MH
> >>
> >>
> >