OK, I'm going to start with the fslswapdim - DON'T DO IT!
If you have a left-right flip, even if one plane looks OK it is
not safe. Do you have any way of deciding what way
around the image really is (say a vitamin capsule)?
If so, and you are getting labels in FSLView that are
correct, then that is fine. If not, I really would never
trust this. At any rate the before and after swapped image
should be different in left-right orientation, and it is probably
something that doesn't show up in the axial view. For instance,
swapping the inferior-superior direction of the MNI152 image
will change the left-right orientation, although it leaves the
axial views exactly as they were. It would be much safer to
also swap the direction of your left-right axis (e.g. if this is
the x axis, then just do -x z y in fslswapdim) as then it
will retain the original left-right orientation of the image.
The second thing is the FLIRT result - this is typical if you
do not use the full search space, which is easy to select
in the GUI, or if you use the command line you will want
-searchrx -180 180 -searchry -180 180 -searchrz -180 180
This enables it to search past the 90 degree rotation that
exists between scans like coronals and axials. Otherwise,
the correct transformation is on the edge of the limits used
by default (-90 to +90 degrees) and so sometime it finds it
correctly and sometimes it doesn't - often just depending on
whether it was a little bit more or less than this.
I hope this helps clarify things.
All the best,
On 11 Aug 2009, at 20:36, Yannis Paloyelis wrote:
> Dear FSL users,
> I used FLIRT to register some coronally-acquired T1s to the MNI
> which resulted in orientations problems for s o m e of the images
> (for some
> images the registered image appeared in the same orientation as the
> template, for the rest I got an interesting set of orientations. As
> images were acquired and converted with the same parameters, I would
> be very
> interested if anybody could explain why registration worked only for
> Therefore, I re-oriented the images to match axial acquisition, using:
> fslswapdim input_image x z y output_image.
> My question: I got a left-right warning:
> WARNING:: Flipping Left/Right orientation (as det < 0)
> However, an anatomical landmark appeared on the same side in the
> axial plane
> both before and after fslswap. Would I be safe to ignore the warning
> assume that left-right orientation is maintained as per original
> (My approach will be to use fslswap to get all the coronal analyze
> T1 images
> to the axial orientation before any further processing with FSL -
> bet, fast,
> flirt,etc. Would there be any other recommendations on that?).
> Many thanks for your invaluable help!