Spatial Registration IOD is ill-defined - DICOM
This is a discussion on Spatial Registration IOD is ill-defined - DICOM ; The new Spatial Registration IOD is ill-defined. A registration matrix
transforms points in one frame of reference to a point in another frame
of reference. The Spatial Registration IOD doesn't specify which frame
is which.
The IOD has a Frame ...
-
Spatial Registration IOD is ill-defined
The new Spatial Registration IOD is ill-defined. A registration matrix
transforms points in one frame of reference to a point in another frame
of reference. The Spatial Registration IOD doesn't specify which frame
is which.
The IOD has a Frame Of Reference module, which we can call the FOR of
the registration. The Spatial Registration Module has a Registration
Sequence, each item of which contains a frame of reference (ignoring
the Reference Image Sequence for now) and a matrix sequence. Let's call
this second frame of reference the FOR of the item.
Section C.20.2.1.1 carefully describes how use the transformation
matrix to map a point in RCS-B to RCS-A. However, it nowhere states
which is frame A: is it the FOR-of-the-registration or the
FOR-of-the-item?
We are directed to PS 3.17 for more information. Unfortunately, Annex
O of PS 3.17 is inconsistent. The second paragraph on Page 151 states:
The Spatial Registration instance specifies a transformation from
its own local Reference Coordinate System to that of the
referenced image sets or atlas.
That makes it sound like the matrix maps from FOR-of-the-registration
to FOR-of-the-item. However, the first line of Page 154 says:
A Spatial Registration instance consists of one or more instances
of a Registration. Each Registration specifies a transformation
from the RCS of the Referenced Image Set, to the RCS of this
Spatial Registration instance (see PS 3.3) identified by the Frame
of Reference UID (0020,0052).
which is precisely the opposite; i.e. from FOR-of-the-item to
FOR-of-the-registration.
Which is it?
-Steve
-
Re: Spatial Registration IOD is ill-defined
Hi Steve,
The short answer is "FOR of the item to FOR of the registration".
This is taken from the "Frame of Reference Transformation Matrix"
description on page 805 which says "A 4x4 homogeneous transformation
matrix that registers the referenced images to the local RCS."
That being said, I agree that this isn't very precise and there seems
to be inconsistencies as you pointed out. Furthermore, since these
matrices are to be concatenated, they cannot all go from the references
images to the local RCS... We need to sort this out.
I'm hoping this can be addressed at the next meeting in a couple of
weeks, and we'll try to make things clearer.
Best,
Alex.
-
Re: Spatial Registration IOD is ill-defined
Hi Alex,
I'm happy to hear that there will shortly be an opportunity to address
this. I hope that the inconsistency in Part 17 can be addressed. But
more importantly, Part 3 should be made unambiguous.
Alexandre Guimond wrote:
> The short answer is "FOR of the item to FOR of the registration".
> This is taken from the "Frame of Reference Transformation Matrix"
> description on page 805 which says "A 4x4 homogeneous transformation
> matrix that registers the referenced images to the local RCS."
The problem for me is that many people treat "registers" as a symmetric
relation, almost as a synonym for "matches". So you cannot know which
way the mapping goes.
Once you start talking about a spatial transformation, however, you
most certainly can specify the frames of reference. Part 3 talks about
frames "A" and "B"; all the standard needs, IMHO is to specify which is
A and which is B.
> That being said, I agree that this isn't very precise and there seems
> to be inconsistencies as you pointed out. Furthermore, since these
> matrices are to be concatenated, they cannot all go from the references
> images to the local RCS... We need to sort this out.
Yes. So specify the identities of "A" and "B" for the CONCATENATED
matrix.
Regards,
Steve
