[Insight-users] tentative strategy for 2D-3D registration with multiple projections

[Jian Wu]

Dear ITK users:

I’m trying to implement an intensity-based 2D-3D registration method that 
aligns multiple 2D projection images with a 3D dataset. The example 
code—IntensityBased2D3DRegistration—in ITK can only handle single 
projection. I came upon with the following idea of modification. I’m 
wondering if this strategy is feasible and neat. Please let me know if you 
have any suggestions or comments. Thanks!

1) Create a new class itk::MultiImageToImageMetric based on 
itk::ImageToImageMetric. The itk::MultiImageToImageMetric has multiple fixed 
images and multiple interpolator pointers.

2) For each specific metric to be used, create a new metric class. For 
example, if we choose normalized correlation, we create a 
itk::NormalizedCorrelationMultiImageToImageMetric. Set its super class as 
itk::MultiImageToImageMetric and modify member functions GetValue(), 
GetDerivative(), and GetValueAndDerivation() of the new class.

3) Modify or create a new itk::ImageRegistrationMethod that can handle 
multiple fixed images and multiple interpolators.

In our application, we use itk::RayCastInterpolateImageFunction as the 
interpolator. Multiple interpolators are constructed by given different 
focal point positions according to the projection geometry.

I guess above scheme will work—but only for orthogonal projections. I still 
don’t know how to handle projections with arbitrary angles. For orthogonal 
projections, a 2D image can be represented using a 3D matrix by setting the 
size of one dimension to be 1. But how to represent a non-orthogonal 2D 
image? Do we have to use a 3D matrix and use so much memory space?

I really hate to modify fundamental ITK classes such as 
itk::MultiImageToImageMetric and itk::ImageRegistrationMethod. It is really 
a limitation of ITK in its original design? Or I’m just not smart enough to 
figure out a simple solution? If someone can give any suggestions, I’ll 
appreciate your help very much!

Jian Wu

Dept. of Radiation Oncology
Univ. of Florida