Image registration has an increasing role in the rapidly advancing field of diagnostic imaging. It is for example used  in periodic comparisons of patient scans, the combination of types of MRI scans, or to fit existing segmentations and atlases to new patient data.

We develop and have developed non-rigid registration methods which do not depend on landmarks.

B-spline grid registration

One of the registration methods we developed is based on work done by Ruekert et al. It consist of an free form deformation (FFD) grid of 1D B-splines which is used to transform an image or 3d volume .The Control points of the grid are moved to transform an image, and a similarity measure between a target image and transformed image is used to determine if the registration improves. Future research will focus on including landmark information (SIFT).

Figure 1, Example of a b-spline transformation grid­, in which the red control points which transform the image will be ­moved by an optimization algorithm.	Figure 2, Example of b-spline registration of a spherical deformed lena onto the normal lena image.

 

The B-spline registration Matlab Code can be found on the Mathworks website 

Demon registration

A much faster registration method is demon registration which is a kind of fluid registration. We implement­ed a multi scale (steepest) gradient based demon registration algorithm. Future research will focus on higher order optimization methods.

The velo­city of each pixel depends on the difference between ­the moving and static image, and the edges (gradients) in both images. The equation below is used, which is smoothed with a gaussian filter.

­

Figure 3, The pixel velocity equation.

Figure 4, Multi-level demon Registration of spherical deformed lena image onto normal lena image.

 

The Demon registration Matlab Code can be found on the Mathworks website 

 

People involved:

• Dirk-Jan Kroon
  
• Kees Slump

 

Last modified: 2012-01-13 (14:12) by Almar Klein