An adaptive irregular grid approach using SIFT features for elastic medical image registration

Abstract

Elastic image registration is an active field of current research. In contrast to B-spline transformations, defined on a regular grid of control points, we consider physics-based transformations defined on irregular grids. With such control point arrangements, the required nonlinear behaviour can be described with less parameters. We combine this transformation model with the SIFT algorithm for identifying prominent image structures that serve well as initial control point positions. Medical applications from different modalities show that with this intelligent control point initialization the number of required control points can be further reduced, which significantly speeds up the registration process.