Rigid Registration of Medical Images by Maximization of Mutual Information

Abstract

Multi-modal medical image registration is an important capability for surgical applications. The objective of registration is to obtain a spatial transformation from one image to another by which a similarity measure is optimized between the images. Recently, new types of solutions to the registration problem have emerged, based on information theory. In particular, the mutual information similarity measure has been used to register multi-modal medical images. In this article, a powerful, fully automated and highly accurate registration method developed at BrainLAB AG is described. Key components of image registration techniques are identified and a summary presented of the information-theoretical background that leads to the mutual information concept. In order to locate the maximum of this measure, a dedicated optimization method is presented. A derivative-free algorithm based on trust regions specially designed for this application is proposed. The resulting registration technique is implemented as part of BrainLAB’s commercial planning software packages BrainSCAN™ and iPlan®.