Non-rigid 2D-3D Medical Image Registration using Markov Random Fields
Enzo Ferrante, Nikos Paragios · HAL (Le Centre pour la Communication Scientifique Directe) · 2013
Abstract. The aim of this paper is to propose a novel mapping algorithm be-tween 2D images and a 3D volume seeking simultaneously a linear plane trans-formation and an in-plane dense deformation. We adopt a metric free locally over-parametrized graphical model that combines linear and deformable param-eters within a coupled formulation on a 5-dimensional space. Image similarity is encoded in singleton terms, while geometric linear consistency of the solu-tion (common/single plane) and in-plane deformations smoothness are modeled in a pair-wise term. The robustness of the method and its promising results with respect to the state of the art demonstrate the extreme potential of this approach.