DAMMER: Direct Adaptive Multi-Resolution MEsh Reconstruction From X-Ray Measurements

Abstract

X-ray computed tomography (XCT) reconstructs a scanned object using measured projection data, with the object typically represented on a voxel grid during the reconstruction process. However, since material interfaces typically do not align with voxel boundaries, a voxel representation inherently suffers from partial volume effects. This paper presents DAMMER: a method that reconstructs a multi-resolution triangle mesh to represent the attenuation values of a piecewise homogeneous object, often encountered in industrial CT, based on X-ray projection data of this object. DAMMER progressively reconstructs this mesh to match the object. For this, different homogeneous segments are created based on an agglomerative hierarchical clustering procedure, which targets a compact object description by optimizing a weighted sum of projection difference and number of edges between different segments. These segments are progressively optimized to match the homogeneous materials in the object. Simulation and real data experiments show that DAMMER generates significantly more accurate reconstructions compared to pixel grid reconstruction methods, outperforming conventional voxel-based methods in capturing the true geometry of complex material boundaries.