2011 Winter Project Week:LandmarkRegularization

• 

  Reference image

• 

  Test image

• 

  Result of automatic registration

• 

  Enhancement with one landmark

• 

  Enhancement with two landmarks

• 

  Enhancement with three landmarks

Key Investigators

• MGH: Nadya Shusharina, Gregory Sharp

Objective

We are introducing a new method for local enhancement of image registration. The method is intended to make rapid, interactive corrections of local registration failures with a small number of mouse clicks. We use Gaussian radial basis functions (RBFs) to define a vector field from point landmarks, and applies regularization based on the vector field second order derivative.

Approach, Plan

Our approach is based on the fact that the Gaussian RBF has infinite support, but the influence of each RBF is localized, making this method well suited for local corrections. In addition, Gaussian RBFs have a distinct advantage over competing approaches such as Wendland functions, because the regularized vector field can be solved exactly with a simple equation. Algorithm for the landmark-based registration has been implemented as a part of out in-house software Plastimatch. We have validated the method on 10 large landmark sets. This work has been submitted for publication to IEEE Transactions on Medical Imaging.

Our plan for the project week is to implement our registration algorithm as a Slicer command line module.

Progress

Delivery Mechanism

This work will be delivered to the NA-MIC Kit as a (please select the appropriate options by noting YES against them below)

1. ITK Module
2. Slicer Module
   1. Built-in
   2. Extension -- commandline YES
   3. Extension -- loadable
3. Other (Please specify)

References

• DIR-LAB. http://www.dir-lab.com
• Plastimatch. http://plastimatch.org