Geometry-Aware Cardiac MRI Representation Learning with Equivariant Neural Fields
Abstract
Cardiac MRI encodes detailed geometric information, but standard deep learning models rely on grid-based encoders that emphasize texture rather than structure. Neural fields offer a continuous alternative, yet Conditional Neural Fields (CNFs) compress each subject into a single global latent, discarding spatial organization. We evaluate Equivariant Neural Fields (ENFs) for cardiac MRI, which replace the global latent with a geometry-aware latent point cloud. ENFs achieve competitive reconstruction quality with far fewer decoder parameters and produce latents that are local, anatomically meaningful, and robust to geometric transformations. For downstream prediction tasks, ENF latents perform competitively with ResNet50 and global CNF latents across several clinical endpoints. These results position ENFs as a compact, interpretable, and geometry-aware alternative for cardiac MRI representation learning.
Cite
Text
Wiers et al. "Geometry-Aware Cardiac MRI Representation Learning with Equivariant Neural Fields." Proceedings of The 9th International Conference on Medical Imaging with Deep Learning, 2026.Markdown
[Wiers et al. "Geometry-Aware Cardiac MRI Representation Learning with Equivariant Neural Fields." Proceedings of The 9th International Conference on Medical Imaging with Deep Learning, 2026.](https://mlanthology.org/midl/2026/wiers2026midl-geometryaware/)BibTeX
@inproceedings{wiers2026midl-geometryaware,
title = {{Geometry-Aware Cardiac MRI Representation Learning with Equivariant Neural Fields}},
author = {Wiers, Jesse L. and Wessels, David R. and Arts, Lukas P.A. and Ruiperez-Campillo, Samuel and Kolk, Maarten Z.H. and Tjong, Fleur V.Y. and Bekkers, Erik J.},
booktitle = {Proceedings of The 9th International Conference on Medical Imaging with Deep Learning},
year = {2026},
pages = {3641-3658},
volume = {315},
url = {https://mlanthology.org/midl/2026/wiers2026midl-geometryaware/}
}