Curvature-based Penalty for Anatomical and Functional MR Human Spine Image Registration

This paper describes an application of image registration. The method is based on an efficient implementation of the curvature registration. This non-rigid registration allows us to find best geometric correspondence between two images. The goal is to register anatomical and functional spine images of the same patient to localize functionality in anatomical images. Most of previous experiments have been tested on brain images and it is the first time that the variational method has been used to register spine images. Registration results are compared with those of MIRT toolbox using two kinds of similarity measures; mutual information (MI) and correlation ratio (CR). MIRT is a Matlab software package for 2D and 3D non-rigid image registration. The model of transformation is parametric and based on B-spline method. Superior results have been achieved compared to the results of MIRT.