Geometry-Aware Cardiac MRI Representation Learning with Equivariant Neural Fields

Jesse Wiers, David R Wessels, Lukas Arts, Samuel Ruiperez-Campillo, Maarten Kolk, Fleur Tjong, Erik J Bekkers

Keywords: Geometric Deep Learning, Equivariant Neural Fields, Cardiac Risk Stratification

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 of-
fer 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 la-
tent 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 com-
petitively with ResNet50 and global CNF latents across several clinical endpoints. These
results position ENFs as a compact, interpretable, and geometry-aware alternative for car-
diac MRI representation learning

Primary Subject Area: Geometric Deep Learning

Secondary Subject Area: Application: Cardiology

Registration Requirement: Yes

Reproducibility: https://github.com/JesseWiers/enfs_for_cardiac

Visa & Travel: Yes

Read CFP & Author Instructions: Yes

Originality Policy: Yes

Single-blind & Not Under Review Elsewhere: Yes

LLM Policy: Yes

Midl Latex Submission Checklist: Ensure no LaTeX errors during compilation., Replace NNN with your OpenReview submission ID., Includes \documentclass{midl}, \jmlryear{2026}, \jmlrworkshop, \jmlrvolume, \editors, and correct \bibliography command., Did not override options of the hyperref package., Did not use the times package., Use the correct spelling and format, avoid Unicode characters, and use LaTeX equivalents instead., Any math in the title and abstract must be enclosed within $...$., Did not override the bibliography style defined in midl.cls and did not use \begin{thebibliography} directly to insert references., Avoid using \scalebox; use \resizebox when needed., Included all necessary figures and removed *unused* files in the zip archive., Removed special formatting, visual annotations, and highlights used during rebuttal., All special characters in the paper and .bib file use LaTeX commands (e.g., \'e for é)., No separate supplementary PDF uploads., Acknowledgements, references, and appendix must start after the main content.

Submission Number: 339