[Insight-users] ITK physical space & Image space

[Rick Shilling]

Hi-

I first wanted to validate a few things regarding Image to physical coordinate transformations in ITK.  All transformations are assumed to be 3-D rotation then translation.  Here's my approach:

We have two scans each with different MRI scanning orientations:  sagittal & coronal.  This means that in image/pixel space,denoted as 'If' & 'Im' for image fixed & Image moving, they are not closely registered.  In fact, unit vectors are permuted & polarity is reversed in the z axis ( right handed to left handed coordinate system change).  

However they have their own anatomical space (denoted as Af & Am for anatomical fixed and anatomical moving) where the two volumes are almost registered.   This means the transformation between Am to Af is close to the identity.  Note anatomical space is not physical space and I have the transformations:  If <-> Af,  Im<->Am.  

I assume the transformation between image space and ITK physical space (denoted as Pf & Pm) is represented by a diagonal matrix whose diagonal entries are the anatomical pixel sizes.  This means image/Pixel directions are the same as physical directions for each volume.  Both translations are set to the zero vector.  

In summary, the 6 Spaces are:
fixed               moving
-----------------------------
Af                Am      |       Anatomical
If                       Im        |     Image/Pixel
Pf                     Pm     |     ITK Physical

And we have the exact transformations:
Af   <-> If   <-> Pf
Am <-> Im <-> Pm


I want to give ITK the best initial estimate possible between physical spaces as a starting point to ITK registration: Pm->Pf.  Since we know in anatomical space the images are initially closely registered, we want to use that information (Am->Af=Identity) & the exact transformations as the way to compute that transformation by composing:
Pm->Im->Am->Af->If->Pf.

In our case we compute & set Pm->Pf = 
[0 0 -1; 
 0 1 0; 
-1 0 0] 
with Translation offset=
[298;34 223];

Though ITK reports:

itk::ExceptionObject (00F290C8) Location: "unknown" File: C:\InsightToolkit-2.6.0\Code\Algorithms\itkMutualInformationImageToImageMetric.txx
Line: 193 Description: itk::ERROR: MutualInformationImageToImageMetric(01262BA0): All the sampled point  mapped to outside of the moving image

I looked at where image/Pixel space boundaries map to in physical space for each and indeed, there is overlap.  I'm wondering if new problems are created if there is a polarity change in the physical moving to physical fixed tranformation that the programmer should know about?  Is there anything to possibly debug that you may know about?  Thanks for your help.

Rick
		
---------------------------------
Ring'em or ping'em. Make  PC-to-phone calls as low as 1¢/min with Yahoo! Messenger with Voice.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://public.kitware.com/pipermail/insight-users/attachments/20060601/fef26b12/attachment.htm