Medical Image Registration Using a New Information Discrepancy Measure

Abstract

Function of Degree of Disagreement (FDOD), a new measure of information discrepancy, was proposed originally to quantify the discrepancy of multiple sequences. On the one hand, this function has been successfully used in many other fields recently. On the other hand, the Kullback-Leibler divergence (KLD)measure has made great success in multimodality image registration. Comparing these two measures, we find that the FDOD has some peculiar mathematical properties superior to the KLD measure. Motivated by these facts, in this contribution, we introduce the FDOD function to solve the (3-D) multimodality medical image registration problem. Furthermore, we propose a normalized version of the FDOD function which will be more suitable to image registration. Finally, we carried out many experiments to validate our methods. Our results illustrate that the proposed registration methods based on the FDOD function and the normalized FDOD function are feasible and competitive, and compared with the methods based on mutual information and normalized mutual information, the proposed normalized FDOD function performs best in most cases, obtaining subvoxel registration accuracy with higher speed and higher success rate.