You can't rely on what the images look like on the screen.
You can only rely on how they are labeled in FSLView.
If there are no labels, then the orientation information isn't
set correctly and FSL is using some defaults which may or
may not be right. If things started as Analyze, then this will
be the case. If you are sure that they were stored in a
radiological fashion (and Analyze format, as it is used in
practice, does *not* guarantee this) then you should not
use fslswapdim like this. The warning is in there precisely
to highlight the fact that it is a dangerous operation and
can really muck up attempts to register later on.
If you do fslswapdim with -x z y then you preserve the
same "left-right orientation" in the image as in the original.
This means that you are doing something equivalent to
just rotating the image - something that flirt could also do.
If you didn't do this then you are rotating *and* reflecting
and it is the reflection which flirt can never do, and hence
you would never get the correct left-right orientation in the
You can use flirt with a large search space and hopefully
it will just register your images without requiring any use of
fslswapdim - if so, then this is the best option. If not, then
use fslswapdim with -x z y as this at least preserves things
as they were, but you should always check that things are
on the correct side after registration, since the original
left-right orientation (if imported from Analyze) might be wrong.
Especially if you are using MRI and CT images. Given
that you have some invasive markers for left and right, it
will be easy to check in your case.
Hope this helps.
All the best,
On 12 Aug 2009, at 10:49, Yannis Paloyelis wrote:
> Dear Mark,
> Thanks very much for your response and the clarifications. I
> appreciate your
> help! I have to admit that I find orientation issues extremely
> Could I just attempt to further clarify things a bit?
> Background: my scans were axial ANALYSE (.hdr/.img, radiological)
> and I
> converted them to NIFTI with fslchfiletype and to coronal
> orientation with
> fslswapdim x z y before submitting them to further processing.
> So you are saying that a feature (in fact some piece of neck which
> after removing skull) which appeared on the right part of the screen
> in the
> axial plane before the x z y conversion, after the x z y conversion it
> should appear on the left side of the screen? And to the extent that
> it does
> not, it means that left-right flipping may have occured? So if I
> apply -x z
> y then it would appear on the left as it should?
> Would this be the recommended way of getting from the coronal to the
> orientation in general? In case it is of help, I have attached a CT
> image with an implanted electrode on the patients left, where the
> coronal was transformed to axial using avwswapdim x z y and which
> in the same warning message. In the resulting axial image, the
> still appears on the patients left. Would this mean that left-right
> orientation was maintained? Would it inform at all our case?
> In the case of further processing in FSL, is your recommendation not
> to swap
> from coronal to axial, but to let FLIRT handle it by specifying the
> parameters apropriatelly?
> Thanks a million for your help -I would be lost without it!
> Kind regards,