Hi Robert,
I'm afraid I don't know a particular FAQ for this. There is some
material in the FSL course, slides 117 and following of:
http://www.fmrib.ox.ac.uk/fslcourse/lectures/flirt_fugue/flirt_fugue_slides.pdf
You're on the right lines about voxel and world space differing.
Essentially, voxel-space is just a way of numbering the voxels, it's
fairly meaningless in terms of spatial location. World-space
coordinates aim to be spatially meaningful. They aren't just
translated from voxel-space though, i.e. there is not just an offset
along each axis, as you seem to assume.
Firstly, world-space takes into account the voxel dimensions, which is
especially important if they are not isotropic (i.e. the voxels are
cuboids rather than cubes), as well as the origin/translations. The
voxel-world mapping can also include rotations, e.g. for an oblique
acquisition the voxels won't be aligned with the desired anatomical
axes. It can even include skews of the axes, in a general 12
degree-of-freedom affine transformation (this is what the 12 parameter
matrix on slide 117 of the above allows for).
Registration, using e.g. FLIRT, creates a correspondence between the
world space of your image and the world space of e.g. the MNI
template. (often, images are resampled/interpolated so that their
voxel spaces correspond too).
More info on how MNI space is defined can be found here:
http://imaging.mrc-cbu.cam.ac.uk/imaging/MniTalairach
I think that's all I have time for, I hope it's of use.
Ged.
Robert Terwilliger wrote:
> Dear FSL,
>
> I am trying to determine a conversion (linear
> translation) from fslview's coordinate system to
> MNI152 coordinate space.
>
> My experience is that fslview, along with other
> viewers I have used, sets the origin (0,0,0) as
> follows:
>
> x = 0: leftmost location in the x direction (rightmost
> radiologically),
> y = 0: most posterior location,
> z = 0: most inferior location.
>
> In an attempt to associate a specific coordinate with
> the nearest anatomical structure, either in
> command-line programs or in websites I've found, it
> appears that (0,0,0) is somehow near the center (or
> perhaps THE center) of the brain itself.
>
> It appears that there are two coordinate systems: a
> voxel-based coordinate system and a so-called world
> coordinate system, that latter having the MNI brain
> centered at the origin.
>
> 1) Is there an FAQ or archived email discussion about
> this?
> 2) If not, what are the offsets along each axis to
> convert from one system to the other..?
> 3) Does FSL already have a command line tool that
> automates this translation?
>
> Many thanks and regards,
>
> Robert Terwilliger
> Laboratory of Neuro-Cognitive Development
> University of Pittsburgh
>
|