[ITK-users] 3D multimodal registration

[Iyas Hamdan]

Hello all,

I’m working on 3D multimodal image registration and I have a problem I've
been struggling with for quite a while.

I have a 3D MRI volume (fixed volume) and a 3D CT volume (moving volume),
but those two volumes have different sizes (different dimensions and
different number of slices).

First of all I have a general question, how does the 3D registration in itk
work basically ? I mean does it take the first slice from the fixed volume
and try to find its correspondence in the moving volume ( by optimizing a
metric ) and then go on to the next slice and so on?
Or does it rather work with the whole volume and not slice by slice ?

Because for me, lets imagine I have 50 MRI slices and 150 CT ones for
example, but the field of view is different between both volumes. I mean
the first MRI slice corresponds ( visually ) to the 50th CT slice for
example so the first 49 CT slices don’t exist in the MRI volume.
 So when I apply my registration method ( affine transformation with the
mutual information as the metric ) the registration fails to give a good
result. It show no errors but the result volume is not well aligned with
the fixed one. Although the transformation ( from the outside let’s say )
is correct, meaning the result volume has the same size as the fixed one
and both volumes are on top of each other perfectly but the slices don’t
look alike at all when I try to show both volumes on top of each other.
Knowing that my initialization is good, and I tuned the parameters of the
optimizer ( regular step gradient descent) and always ended up with almost
the same result.
So is that a normal behavior of itk to fail when the field of view is
different ?

Because I tried to cut off the first 49 CT slices, and created a new volume
with the rest of the slices ( so that I had the same field of view with the
MRI volume ), and when I applied the registration between this new volume
and the MRI one, the registration worked perfectly and I had a very good
result.


So what’s wrong here ? and if the field of view is the problem how can I
fix that automatically ? because I can delete the extra slices from one
volume manually, but what I'm looking for is an automatic approach and I
couldn’t think of one.


Any hint would be really appreciated,
Thanks in advance,

Best regards,

Iyas
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://public.kitware.com/pipermail/insight-users/attachments/20140811/6d73f6be/attachment.html>