[Insight-users] Influence of background on Registration metric & Choice of transform VersorRigid3DTransform / QuaternionRigidTransform / Euler3DTransform

[Axel Martinez Möller]

 Hi all,

 I am doing a 3D image registration between a PET and a CT image, both of
size 128x128x47 voxels, each voxel being 5x5x3 mm. I have encountered two
issues:

1) Approximately 60 to 80% of my images is background (the patient occupies
only a small part in the center of the field of view). I am afraid that this
is having a bad influence on the registration: there are more voxels to
interpolate, and the sampling for the metric takes into account much voxels
which are irrelevant. Are my assumptions right? If so, do you know how to
remove the influence of the background?

2) Until now I have tested only the TranslationTransform, until I managed to
tune the parameters and make it work. I want to switch now to a translation
+ rotation transform. Going through the manual I am quite confused about the
3 transforms which refer all to a 3D rigid registration:
*VersorRigid3DTransform
*QuaternionRigidTransform
*Euler3DTransform
My first tests were a bit frustrating. Can you give me some advice? Which
combination Transform / Optimizer should perform better? Which should be
avoided?

 Thanks,

  Axel