On validation of non-physical techniques for elastic image registration

Abstract

Non-physical techniques for elastic image registration such as different spline-based optimization methods are often applied in biomedical applications for image normalization w.r.t. non-rigid transformations. Since mechanical properties of biological structures to be registered are usually unknown, a "ground truth" validation of the results of image registration is not possible. This article presents a framework for the validation of elastic image registration techniques by a direct comparison of displacement fields vs analytical or numerical reference solutions of customizable boundary value problems. The proposed procedure enables an easy handling of material parameters, domain shapes and boundary conditions, and provides a flexible benchmark-tool for quantitative validation of elastic image registration algorithms.