Anisotropic Adapted Meshes for Image Segmentation: Application to Three-Dimensional Medical Data

Abstract

This work focuses on a variational approach to image segmentation based on the Ambrosio--Tortorelli functional. We propose an efficient algorithm, which combines the functional minimization with a smart choice of the computational mesh. With this aim, we resort to an anisotropic mesh adaptation procedure driven by an a posteriori recovery-based error analysis. We apply the proposed algorithm to a computed tomography dataset of a fractured pelvis to create a virtual model of the anatomy. The result is verified against a semiautomatic segmentation carried out using the ITK-SNAP tool. Furthermore, a physical replica of the virtual model is produced by means of fused filament fabrication technology to assess the appropriateness of the proposed solution in terms of resolution-quality balance for three-dimensional printing production.