[Insight-users] 3D free form registration with ITK

[Luis Ibanez]

Todd Jensen wrote:
>>From: Ali Sadeghi 
> 
> 
>>Subject: [Insight-users] 3D free form registration with ITK
>>
>>I'm just trying to use ITK for a 3D free form registration (rigid + affine +
>>Bspline).  The registration was done good, but the algorithm corrupt the
>>first and last slices of the 3D image in the output of the registration,
>>most parts of these two slices were filled up with the background default
>>value. Has anyone else encountered with such a problem and if yes how could
>>you solve it?
> 
> 
> I don't think this is a corruption, but an artifact from the transformation/interpolation. The interpolator is not finding all the required pixels for the interpolation in the moving image for the location in the fixed space.  It probably needs a pixel that is at index -1 or n in the 3rd dimension (assuming you have n slices).
> 
> One way I found around this is to create a temporary moving image that is n+2 slices, paste the original moving image into 1 to n. But this results in partial volume artifacts from interpolation - another issue to solve.
> 
> Todd Jensen
> _____________________________________


Ali,

As Todd pointed out, this is not data corruption.

It is perfectly normal for the resample filter to
find that some pixels in the boundary of the image
map outside of the moving image domain.

In such cases, the Resample filter uses the
DefaultPixelValue() that you set in the filter.

It is a good practice to initially set this default
value to a number that will make these regions
very visible. In this way you become aware that the
pixels in question were assigned a "filler" value.

Once you have recognized this situation, you can
set the default value to a less noticeable value.
(which typically depends on the content of your
image).



     Regards,



         Luis