Surface Based Atlas Matching of the Brain Using Deformable Surfaces and Volumetric Finite Elements

The automatic identification and localization of structures in magnetic resonance (MR) brain images are a major part of the processing work for the neuroradiologist in numerous clinical applications, such as functional mapping and surgical planning. To aid in this task, a considerable amount of research has been directed toward the development of 3D standardized atlases of the human brain (e.g. [5]). These provide an invariant reference system and the possibility of template matching, allowing anatomical and functional structures in new scans to be identified and analyzed. There are mainly two types of methods for doing deformable atlas matching : surface-based and volume based methods. Surface based methods deform key surfaces of the atlas onto the target image and interpolate the surface displacement to obtain a fully volumetric mapping (e.g. [3]). Volumetric methods compute a deformation field that minimizes a similarity criterion between the atlas and the target image under a given regularization constraint (e.g. elastic). To reach convergence, the computations must be done in a multi-resolution fashion (e.g. [1]). The main issue with both methods is the initialization of the volume or surfaces. In this paper, we propose a surface based method with automatic initialization of the deformable surfaces using a global parametric transformation. We use an elastic volumetric finite element (FE) deformation model to infer a volumetric deformation field from the obtained surface deformations. The method provides us with a fully automated volumetric mapping of the atlas onto a target image.