Reconstructing complex structures from planar cross-sections is a challenging problem, with wide-reaching applications in medical imaging, manufacturing, and topography. Out-of-the-box point cloud reconstruction methods can often fail due to the data sparsity between slicing planes, while current bespoke methods struggle to reconstruct thin geometric structures and preserve topological continuity. This is important for medical applications where thin vessel structures are present in CT and MRI scans.
This paper introduces CrossSDF, a novel approach for extracting a 3D signed distance field from 2D signed distances generated from planar contours. Our approach makes the training of neural SDFs contour-aware by using losses designed for the case where geometry is known within 2D slices. Our results demonstrate a significant improvement over existing methods, effectively reconstructing thin structures and producing accurate 3D models without the interpolation artifacts or over-smoothing of prior approaches.
Our approach integrates three key innovations: adaptive sampling focused on thin structures, a hybrid encoding combining hash-grid and Fourier features for detail preservation, and a novel symmetric difference loss that maintains topological consistency between slices. This comprehensive approach enables robust reconstruction of complex geometries while preserving fine structural details.
Visualization of CrossSDF's reconstruction process, showing convergence from initial predictions to final detailed models for both synthetic and medical structures.
Reconstruction results on thin structures featuring the Heart (top row) and Pulmonary (bottom row) for various methods. Results are presented using both input-aligned and non-aligned planes, displayed on the ground truth meshes.
@article{tomandsal2024_crosssdf,
title={CrossSDF: 3D Reconstruction of Thin Structures from Cross-Sections},
author={Walker, Thomas and Esposito, Salvatore and Rebain, Daniel and Vaxman, Amir and Onken, Arno and Li, Changjian and Mac Aodha, Oisin},
journal={arXiv preprint arXiv:2407.06938},
year={2024}
}