Overall Objectives
Research Program
Application Domains
Highlights of the Year
New Software and Platforms
New Results
Bilateral Contracts and Grants with Industry
Partnerships and Cooperations
XML PDF e-pub
PDF e-Pub

Section: New Results

Deformable group-wise registration using a physiological model: Application to diffusion-weighted MRI

Participants: Evgenios Kornaropoulos, Evangelia I. Zacharaki, Nikos Paragios (in collaboration with Centre Hospitalier Universitaire Henri-Mondor and Chang Gung Memorial Hospital)

In this contribution [2] we develop a novel group-wise deformable registration method for motion correction in Diffusion-Weighted MRI towards computing a more accurate Apparent Diffusion Coefficient parametric map (ADC map). Calculation of the ADC has been performed without motion correction in the previous studies. It is reported though that ADC is a parameter susceptible to artifacts, the most frequent of all being patient's motion and breathing, resulting in misregistration of the images obtained with different b-values. Being group-wise designed, the image registration method we propose has no need of choosing a reference template while in the same time it is computationally efficient. We aim at finding the optimal deformation fields of the diffusion-weighted (DW) images using a temporal constraint, related to the diffusion process, as well as a smoothness penalty on the deformations. To this end, we address the deformation fields estimation problem with an Markov Random Fields formulation, in which the latent variables are the deformations (B-spline polynomials) of the images. The latent variables are connected with the observations towards ensuring meaningful temporal correspondence among the DW images. They are also inter-connected in order to decrease the cost of pairwise comparisons between individual images. Linear programming and duality are used to determine the optimal solution of the problem. Finally, as an image similarity criterion in the MRF framework, we used a metric that was based on a physiological model describing the image acquisition process. Quantitative evaluation of the method was performed, in which it was compared against two state-of-the-art methods that use other modelling criteria, It outperformed both of them, while the ADC map derived by our method appeared to preserve structure, that was not observable by the other methods.