[Insight-developers] [GDCM] ITK Origin and coordinate system

[David Clunie]

Hi all

Just to clarify a few things with respect to DICOM and what
vendors do, without knowing enough about ITK or gdcm to
answer the context of the concern. Please excuse me if I
restate the obvious.

As has been pointed out in previous posts there is absolutely
no ambiguity about the DICOM attributes in this respect, nor
whether the vendors interpret them differently (they do not).

Specifically:

- all DICOM references are relative to a specific "frame of
reference" identified by a UID - all images that share that
FoR share the same, completely arbitrary, origin; the origin
might be the isocenter of the magnet, which is constant, but
the patient position relative to it is not; this is the
context in which all other DICOM attributes related to the
"patient coordinate system" should be interpreted; the
corollary is that different FoRs means coordinates and vectors
are not comparable

- the center of the top left hand voxel of a slice is defined
relative to the FoR-specific origin by an x,y,z tuple that
represents a translation of that location in mm from the offset;
the x,y and z directions are patient (not gantry) relative,
and their order is consistent and defined (see the standard)

- the direction of the rows and columns are defined relative
to the patient by unit vectors (direction cosines), again
the details are in the standard.

- the Image Position (Patient) TLHC location (which defines
the offset from the FoR origin) is completely independent of
the Image Orientation (Patient) unit vectors, which define
the plane of the slice (only)

Again, the x,y,z translation from the origin of the TLHC is
in the nominal patient relative FoR, NOT the plane specified
by the unit vectors that define the orientation of the slice.

Note that the use of "patient relative" is relatively imprecise,
in the sense that if the operator does not line the patient's
head foot axis exactly down the center line of the gantry or
table, then though the direction of the Z axis will nominally
be along the head-foot axis, but reproducible only within the
same nominal FoR, and even then, only as long as the patient
doesn't move without the operator re-landmarking.

All the vendors interpret this the same way, with the one notable
exception being arguments over whether the TLHC is the center
of the voxel or one edge of it (the standard has been clarified
to specify the center, whereas previously it was unspecified,
hence older implementations vary in this respect).

Hope this helps.

David