This talk addresses the problem of 3D and 4D medical image segmentation based on deformable models. Surfaces represented by active meshes are used to model anatomical structures. We introduce the simplex mesh formalism that is adapted to define local regularizing constraints including shape constraints. The prior shape information makes the 3D reconstruction process more robust especially in area where image data are noisy or lacking. A restriction of the surface model space of deformations similarily aims atimproving the segmentation robustness to noise and outliers.

Implicit surface representations (level sets) have encountered some success for their capability to perform topology changes. We implemented an efficient topology adaptation algorithm for our explicit representation. The algorithm results are compared to the level sets method.

An active model deforms according to a data term to match the boundaries extracted from an image. Different strategies for computing the data attraction term adapted to various 3D imaging modalities are investigated. For instance, region based and intensity-profile based data forces are used for monomodal segmentation and multimodal image registration.

Finally, our deformable modelling framework is extended to deal with time sequences of 3D images (4D images). The introduction of time continuity constraints allows to improve the deformation process. Segmentation examples on different sequences of the heart left ventricle are given.