Paper: | SA-PM-PS1.1 |
Session: | Image Registration |
Time: | Saturday, April 8, 13:30 - 14:50 |
Presentation: |
Poster
|
Title: |
An Interactive Hybrid Non-Rigid Registration Framework for 3D Medical Images |
Authors: |
Antoine Azar; Siemens Corporate Research, Inc. | | |
| Chenyang Xu; Siemens Corporate Research, Inc. | | |
| Xavier Pennec; INRIA | | |
| Nicholas Ayache; INRIA | | |
Abstract: |
This paper proposes a new interactive hybrid non-rigid registration framework that combines any intensity-based algorithm with a feature-based component, using an iterative dual energy minimization. The resulting transformation combines both intensity-based and feature-based deformation fields. The feature matching exploits user-placed landmark pairs, and based on saliency and similarity measures, optimizes the correspondences in the neighborhood of each landmark. A dense feature-based deformation field is then generated using a thin-plate spline interpolation. Additionally, the framework allows user interactivity for live guidance of the algorithm in case of errors or inaccuracies. We present three experimental results of our hybrid approach on lung, pelvis and brain datasets, and show how in each case, the registration benefited from the hybrid approach as opposed to its intensity component alone. |