[Insight-users] Problems with manual registration and DeformableMesh3DFilter

[Marcus Wallin]

Hi!

I'm having some problems doing manual registration followed by processing
using the DeformableMesh3DFilter (very similar to the example
deformablemodel1).

The entire process looks like this:

(1) The user imports a series of DICOM patient images and roughly aligns a
reference segmentation by placing it on the patient images. The
user-specified alignment can be represented as a translation in 3D where the
z component is the slice offset between the two images combined with the
slice-thicknesses. To the user, the patient images and reference mesh are
displayed in physical space coordinates. What I expect from this step is a
translation in 3D that maps the patient images and the reference images to
roughly the same physical coordinates, since this is a prerequisite for the
DeformableMesh3DFilter to function.

(2) The patient image is then pre-processed using the
GradientMagnitude/GradientMap filters.
(3) The origin of the patient image is set to the 3D-translation made by the
user using image->SetOrigin().

(4) The reference image is read and parsed into a mesh using MarchingCubes.

(5) The image from (2) and the mesh from (4) are fed into the
DeformableMesh3DFilter.

(6) The end result is displayed to the user.

The manual registration in step (1) is used as an attempt to improve
processing speed (I thought that a good manual registration was sufficient
for the deformable model algorithm, thus eliminating the need for a
resource-hungry automatic registration).

The end result from this process is not what I expect, however. The lower
part of the deformed mesh seems to be deformed (not correctly, but at lest
something has happened to it), but the upper part is not (when compared to
the reference mesh). I suspect that the image and mesh are not aligned
correctly in the physical space, but I'm not sure.

* Can someone please give some comments about this?
* Am I doing something fundamentally wrong here, or do I just need to tweak
the deformable model parameters somewhat?
* Is it possible to inspect the physical space coordinates that the
deformable mesh algorithm uses internally to confirm that the pixel spacing
and the pixel origin was set correctly?

Thanks in advance,
Marcus