Model-Based Elastic Registration



This project presents a new approach for the non-rigid registration of medical data. The proposed technique can also be applied to model-based segmentation.

Affine Registration

Affine registration is the whole family of transformations including rigid, skew or stretching that can be represented by a 3×4 matrix. This family of transformations is known as 6 to 12 degrees of freedom (DoF) in the 3D case.

Non-Rigid Registration

Non-rigid registration or free-form deformations can be achieved using, for example, B-Spline techniques. In this case, the number of degrees of freedom depends on its control grid density in which case the number of DoF can dramatically increase.

Left: medical data before registration. Right: deformable model after non-rigid deformation during segmentation process. Note that local registration is still not perfect. Active contours should be used to find the exact position of the data.

Proposed Solution

The global registration process first starts with coarse registration, including a rigid followed by an affine transformation. The intermediate result is refined using a non-rigid deformation based on a deformable model technique. The model consists of a simplified skeleton (medial axis) connected to the skin (surface points) and deformation control points. Non-rigid raw deformations can be achieved by modifying the control points positions and strengths. Finally, the local deformation is achieved a free form motion based on active contours and solved using dynamic programming.


Efficient registration is one of the key issues to validate the preoperative information during the surgical process. This project is one module of the activities lead by the EPFL-VRAI group within the PNR Co-Me project.