Shape comparison of the hippocampus using a multiresolution representation and ICP normalization

Abstract

It is known that deformation of the hippocampus shape is involved with several neurological diseases. In this paper, we propose a hybrid shape representation scheme, which consists of multiresolution skeletons, voxels and meshes for the shape analysis of the hippocampus. Initially, a hippocampal surface model is reconstructed from MRI and then it is placed into a canonical coordinate system, where the position, orientation and scaling are normalized. From the voxel representation of the hippocampus, multiresolution skeletons are extracted and Iterative Closest Point normalization is carried out. Then the shape similarity of two hippocampal models is computed with a hierarchical fashion. In addition, we have implemented a neural network based classifier to discriminate whether a hippocampal model is normal or not. Results indicate that the proposed hybrid representation and the skeleton-based normalization using ICP are very effective in 3D shape analysis of the hippocampus.