VTK CISG Registration Toolkit An Open Source Software Package for Affine and Non-rigid Registration of Single- and Multi
Voxel-based image registration using Normalised Mutual Information (NMI) has been shown to register single- and multi-modal 3D images accurately without any user interaction [1,5,4]. Our group has proposed both an affine and a non-rigid registration algor
- PDF / 1,162,389 Bytes
- 4 Pages / 439.32 x 666.12 pts Page_size
- 23 Downloads / 177 Views
T. Hartkens, D. Rueckert*, lA. Schnabel, DJ. Hawkes, D.L.G. Hill Computational Imaging Sciences Group, King's College London, London, UK Zusammenfassung. Voxel-based image registration using Normalised Mutual Information (NMI) has been shown to register single- and multi-modal 3D images accurately without any user interaction [1,5,4]. Our group has proposed both an affine and a non-rigid registration algorithm based on NMI, and has validated these algorithms on a range of medical applications like brain. breast, and cardiac data [7.8.6.2.3]. We present a publicly available software package that incorporates these algorithms in a user-friendly command-line and graphical interface including a visualisation tool for 3D image pairs in order to analyse registration results. Beside the pure usage of registration algorithms, the software can be easily adjusted to specific environments (e.g. including other image file formats) and can be modified for specific applications.
1 Method We present a software package that provides an affine and a non-rigid voxel-based image registration algorithm. The affine registration algorithm optimises an affine transformation iteratively by maximising the normalised mutual information (NMI) [9] of the joint probability distribution of two images using a multi-resolution approach [8] (Fig.l). The optimisation process can either be restricted to a selected number of degrees of freedoms (e.g. only rigid parameters) or be initiated to optimise up to 12 degrees of freedom. While the affine registration captures only the global motion between an images pair, the non-rigid registration models local motion using a free-form deformation (FFD) model based on B-splines (Fig.2). The non-rigid registration algorithm defines a regular grid of B-spline control points and deforms the underlying image by moving these control points [6]. Again, NMI is used as a measure of the similarity of the image pair. Both algorithms are implemented using the free and widely-used visualisation software VTK by Kitware (http://www.kitware.comlvtk) a powerful visualisation pipeline for interactive image display.
2 Result Our software package provides command-line programs for off-line affine and nonrigid registration of 3D single- or multi-modal images. The resulting transformations are * now with Visual Information Processing, Department of Computing, Imperial College London,
London, UK
M. Meiler et al. (eds.), Bildverarbeitung für die Medizin 2002 © Springer-Verlag Berlin Heidelberg 2002
410
Abb. 1. Rigid registration of 3D CT and 3D MR image The subfigure in the middle shows the contour of the unregistered CT image (left) overlaid on tl reference MR image. The contour of the registered CT image is overlaid in the right subfigure.
stored in a file for use of an image transformation tool for interpolating the registration result. The function used in the command-line programs are also incorporated in a graphical interface (see Fig. 3) which displays orthogonal views of 3D image pairs and offers an interactive inte
Data Loading...
