Deformable models are used for the segmentation of structures in 3D images. The basic principle of such methods consists of the adaptation of flexible meshes to the image. However, the simultaneous segmentation of multiple or composed objects often causes problems in that spatial relationships between the objects are violated, or that meshes are intersecting each other. According to the present invention, a priori knowledge about spatial relationships between objects is introduced into the shaped model. This allows to maintain spatial relationships between the objects and to avoid intersecting meshes.