Dear Brad,
Thanks for sending your data.
It is quite tricky to get very good looking results out of this due
to the relatively large amount of signal loss and the high level of
ghosting. However, both the FEAT results and the direct
FUGUE results looked pretty good to me. One thing that really
helps when looking at the output is to mask the unwarped output
image by the fmap mask (from the undistorted - UD - space).
Note that you need to create a binary mask using avwmaths++
on an image such as FM_UD_fmap_mag_brain as the "sigloss"
images are really only there for registration purposes, and are
not binary either.
I agree that the registration in FEAT is not doing well, due mainly
to the large distortion, signal loss and ghosting present. However,
the predominant change is rotation, and the -12 in the matrix is
largely to compensate for rotations about a non-zero centre and
does not represent a simple translation. Note that it is quite difficult
to directly interpret the matrix entries in simple ways because of the
coupling between rotation and translation. Have a look at the differences
in the images (pre and post transformation) by flicking between them in
FSLVIEW. That shows the rotation quite clearly.
In the end I think that the best thing is just to apply FUGUE directly
and bypass the registration phase in this case. Your fieldmap and
EPI seem well registered to begin with so that should be fine. The
main thing you should do in this case is specific a mask for both the
fugue call and in a post-processing step to mask out the unwanted
portions of your unwarped EPI (which is full of ghost and chunks
of the image that have been shifted by a large amount). If you do this
then you should find that the outer cortex, ventricles, mid-line, etc
all line up very well. This will not make the inferior slices look good
though, as they are too full of signal loss, although the structures that
are visible look fine.
I hope this helps.
All the best,
Mark
Bradley Buchsbaum wrote:
> Hi,
>
> After playing around with Predlude and Fugue using the online
> documentation (and having only a little success) I decided to try
> using the Feat front-end. This produced much better results, probably
> because Feat does a number of things that are not discussed in the
> tutorial (e.g. masking with "sigloss", etc.). While the unwarping
> works much better within Feat there
> is still some disagreement between the anatomical images and the
> unwarped epis. One issue has to do with the alignment of the field
> map to the distorted EPI, which produces the transformation matrix
> FM_2_EF.mat. In our case the phasemap images are acquired just before
> the EPI sequence, so we expect these images to already be well
> aligned. But FM_2_EF.mat has a z-shift parameter of -12.0092, which is
> an implausible shift (this is a general problem we have using
> automated registration algorithms with our 4T epi data). Indeed, if I
> rerun FUGUE using the unregistered fieldmap, the results improve
> dramatically.
>
> I guess the answer is that I need to create my own script without the
> fieldmap-to-epi registration step, but the Feat script is quite
> elaborate and I'd like to avoid this.
>
> any suggestions?
>
> regards,
>
> Brad Buchsbaum
>
>
>
>
|