Abstract
Cardiac shape has been shown to contain prognostic information beyond standard scalar clinical measures. Yet, cine MRI representation learning often relies on image intensities alone. Therefore, we incorporate cardiac shape as a prior into Equivariant Neural Fields (ENFs), which represent cine MRI images as continuous, localized, and geometrically grounded latent pointclouds. To obtain cardiac shape representations, we introduce HeartSDF, a latent set signed distance field model that encodes segmentation-derived left ventricular and myocardial anatomy as continuous 3D geometry. We study two ways of using cardiac shape in ENFs: an explicit prior, where HeartSDF latents are added to the ENF representation, and an implicit prior, where reconstruction is focused on the cardiac region by utilizing segmentations during training and inference. Experiments on UK Biobank show that HeartSDF captures global cardiac geometry and that cardiac shape provides a useful prior for clinical endpoint classification. Both explicit and implicit priors improve ENF-based classification, demonstrating the value of incorporating anatomical structure into neural field based cine MRI representations.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/ShapeMI_031.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to Open Review
Open Review Page: https://openreview.net/forum?id=8DWIrSHZVd
BibTex
@InProceedings{BuiSac_Cardiac_MICCAISAT2026,
author = { Buijs, Sacha E. AND Tjong, Fleur V. Y. AND Bekkers, Erik J. AND Ben Haddou, Soufiane AND Kuipers, Thijs P.},
title = { { Cardiac Shape Priors for Equivariant Neural Field Representations of Cine MRI } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17259},
month = {pending},
page = {pending}
}