Abstract
Cardiovascular magnetic resonance (CMR) imaging is the reference standard for quantifying cardiac anatomy and function, but clinical interpretation is typically based on scalar measurements that only summarize cardiac remodeling. We propose a neural implicit cardiac atlas for short-axis cine CMR in which each subject is represented by a compact latent embedding and decoded by a shared coordinate-based implicit function into image intensity and cardiac structure occupancy. In 201 subjects, including healthy controls and subjects with left ventricular dilation or hypertrophy, we compared a reconstruction-only baseline with embedding-structure regularized variants that align pairwise embedding distances with pairwise phenotype distances. Held-out embeddings were inferred by test-time adaptation and evaluated for clinical readout, neighborhood retrieval, pairwise distance preservation, semantic phenotype directions, and atlas prototype synthesis. The learned embeddings encoded clinically meaningful variation: linear probes recovered disease status, sex, age, body surface area, ventricular volumes, and myocardial mass; nearest-neighbor retrieval identified morphologically similar subjects; and semantic directions captured interpretable variation related to hypertrophy, dilation, chamber size, and sex. Decoded group-average embeddings pro- duced atlas prototypes. These results indicate that subject embeddings in neural implicit cardiac atlases can serve as compact, clinically readable population coordinates for car- diac morphology rather than reconstruction codes alone.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/STACOM2026_027.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/profile?id=~Thomas_Hadler1
BibTex
@InProceedings{HadTho_ANeural_MICCAISAT2026,
author = { Hadler, Thomas AND Schulz-Menger, Jeanette},
title = { { A Neural Implicit Cardiac Atlas with Clinically Readable Subject Embeddings } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17267},
month = {pending},
page = {pending}
}