> >In SPM2, the defaults.analyze.flip is only used when working out the
> >orientations of images without a .mat file (or a .mat file created with
> >SPM99).  If there is a .mat file, then defaults.analyze.flip, the origin
> >field, and the voxel sizes are all ignored.
> >
> >If you use the DICOM conversion in SPM2, then images should be created
> > with a .mat file (unless they are exactly axial and the origin is at the
> > exact centre of a voxel).  The conversion routine will take the value of
> > the flip in order to figure out whether the order that the voxels are
> > stored in should be left- or right- handed.  The values in the .mat file
> > are set so that images are displayed using a right-handed system.
>
> I'm still confused.
>
> For example, the data for the FIAC contest is in SPM ANALYZE format, with
> accompanying .mat files.  According to what I read above, and what I read
> in the SPM Data format page, that should clearly define all three axes'
> orientations.
>
> And yet, for some reason the FIAC contest description then contains the
> following
> sentence:"With these 'mat' files, the images are considered in the
> radiological
> convention (Left is Right)."

I didn't write this documentation - so I can't say.  You will never hear me
refer to files being in radiological or neurological orientation, as this
terminology only refers to the way that an axial or coronal slice is veiewed.
I'm not actually sure whether a coronal  image with the most inferior part of
the brain displayed at the top of the image would be "radiological" or
"neurological" if the right side of the brain was on the right.

The notation that should be used for volumes is right- or left-handed, and a
distinction should be made with respect to how the data are stored, and
whether there is an implicit flip in the voxel-to-world mapping of the
images.

>
> Would the average SPM user, despite SPM's clear definition of axis
> orders and
> orientations via the mat file, assume this means that instead of
> L-R,P-A,I-S axes, you actually have R-L,P-A,I-S axes after applying the
> transformation in the matrix?

SPM2/5 assumes that people get their data in the correct orientation before
doing anything with it.  The only time that SPM knows for sure what the
orientations are is when the DICOM toolbox was used for converting the data.
If you use some other package to convert the data, then make sure you know
what it does.

>
> We've already asked JBP for clarification specifically about this
> contest data,
> but I see this terminology a lot when I'm supporting SPM users, and
> would like
> to know how it is intended to be.  Can you offer any additional
> clarification?

Ask about the order that the voxels are stored.  You can consider a volume as
several planes of data stacked together, and each plane being several rows
stacked together.  It is useful to clarify the direction as you traverse from
plane 1 to plane Nz, row 1 to row Ny, and voxels in a row going from 1 to Nx.

Given this knowledge,  it is possible to determine whether the voxels
themselves are stored using a left- or right-handed system.  If they are
stored left-handed, then the voxel-to-world mapping should have a negative
determinant.  All SPM2 and SPM5b needs is that the user ensures that the mm
(or world) coordinates that SPM reports are within a right-handed system.

Best regards,
-John