Dear Ged,
We are referring to the handedness of the real-world coordinate
system when talking about left-handed or right-handed. This is a
complicated issue and we try to present it in a way which is
accessible to both medical people, life scientists and physical
scientists.
It is a confusing issue mainly because of the potential terminology
that exists in radiology, psychology, neurology, etc, etc.
Even within a purely mathematical framework it is possible to define
things in different ways, and we are trying to be consistent with the
nifti
definition given in:
http://nifti.nimh.nih.gov/nifti-1/documentation/faq#Q15
The avg152T1 image is our standard space image and has well
defined real-world coordinates. It is these real-world coordinates
that define it as being right-handed. The transformation mapping
the voxel to real-world coordinates does in fact have a negative
determinant, and any image that maps to the standard coordinate
system with a negative determinant is considered "radiological"
by our definition, with "neurological" ones having a positive
determinant.
Personally, I would have chosen different conventions, but we have
to be consistent with the standard that was originally prepared and
that was the avg152T1 images (or the avg305 images) which were
prepared and shipped with the above conventions. Hence we are
sticking to them carefully, as introducing anything different at this
point would raise even more confusion which would be bad.
As for why we need different handed coordinates - again I would
love to force that every image in the world was only ever stored
one way around. However, this is just not how life is, and scanner
manufacturers, DICOM converters, and custom software are capable
of producing images in either handedness and so we need to be
able to deal with both cases. Keeping careful track of the handedness
and making sure that it is correct is something that needs careful
consideration every time new scanners, scanner software, reconstruction
or processing techniques are used. Only by verifying that things are
maintained correctly by all stages in the pipeline can the handedness
issue be confidently dealt with and the left from the right side
determined
from the images.
Note that before nifti, when we were dealing with Analyze images
only, we always assumed that they were stored in the same
handedness as the avg152 images and did not deal with the
alternative. This was because Analyze did not store enough information
to reliably determine the handedness. However, nifti images do store
sufficient information and it is perfectly legal to switch between them
(and scanner/reconstruction software is unlikely to be uniformly
consistent) so that is why we have to deal with this issue.
I hope that this clears things up a bit.
All the best,
Mark
On 6 Mar 2006, at 20:16, Ged Ridgway wrote:
> Hi,
>
> There seems to me to be a mistake in the FSL FAQ, it says
>
> "Our technical definition of neurological convention for image
> storage is that the coordinate system is left-handed"
> ...
> "The avg152T1 image is now our definition of radiological"
>
> Then when describing the avg152T1 system, the FAQ says that the mm
> x-coordinate is opposite to the voxel x-index. This seems to imply
> that avg152T1 is left-handed (the transformation from voxel indices
> to world coords would have negative determinant in order to flip x
> as described here) so should the first sentence quoted above read
> neurological is right-handed? Assuming that they aren't both left-
> handed systems?
>
> Thanks,
> [something of a rant follows]
>
> It also seems to me that regardless of whether one wants screen-
> left to be patient-left or patient-right, there should never be
> reflections (e.g. flipping of x-dimension). Why not simply rotate
> by 180deg? Consider scanning a phantom with a clockwise screw-
> thread, in my opinion all software (reconstruction, viewing, etc.)
> should preserve the handedness of the thread, i.e. the volume can
> be rotated to give your desired left-right coronal display, but
> shouldn't be reflected, since then inference of the handedness from
> multiple planes will be wrong.
>
> Please say if you think I am talking nonsense, and in particular if
> anyone can enlighten me as to the advantages of switching
> handedness between voxel and world coordinates (by which I don't
> mean whether coronal images are screen-left = patient-left, but
> rather the voxel coordinates and the world coordinates can no
> longer be rotated into alignment) I would be very interested to hear.
>
> Many thanks,
>
> Ged.
|