Feature Preserving Volumetric Data Simplification for Application in Medical Imaging

Abstract

In this paper, we propose a new simplification algorithm to reduce the large amount of redundancy in 3D medical image datasets and generate a new representation in tetrahedral meshes with considerably lower storage requirements. In the proposed algorithm, we first apply level set segmentation to partition the volume data into several homogenous sub-regions. We consider the interior boundaries between sub-regions as contributing more to the significant visible features. Next we convert the regular grid data into a tetrahedral representation and simplify the irregular volume representation by iteratively removing tetrahedra without significantly altering the exterior boundary or interior field distribution features. Within each sub-region, field gradients, tetrahedral aspect ratio changes and variances of interior region values are further used so as to maintain features of the original dataset in regional interiors. We tested our algorithm on several 3D medical datasets. The promising results show that we reduce redundancy and yet preserve important features and structures present in the original data set for decimation rates up to 50%.