Abstract
Cardiac MRI segmentation supports two clinical measure- ments: left-ventricular ejection fraction (LVEF) from cine imaging, and myocardial scar mass from late gadolinium enhancement (LGE) imaging. These are related but separate tasks, as cine and LGE differ substantially in geometry and class balance. We segment each view independently at native resolution with a per-view 2D U-Net (pretrained ResNet-34 encoder), then compute LVEF or scar mass deterministically from the mask rather than end-to-end. Four additional architectural modules, each targeting a specific failure mode (scar contrast, anatomical consistency, through-plane inconsistency, temporal context), tie or underperform this pretrained baseline. Transfer learning, aided substantially by test-time augmentation, is the sole intervention that improves performance on the most challenging class, scar. On top of this, a four-stage postpro- cessing system (filtering, adjacency constraints, calibration, recombina- tion) corrects systematic errors without retraining. On CMR-MULTI, the deployed pipeline reaches LVEF correlation 0.97 (mean absolute er- ror, MAE, 3.11 percentage points) and cuts LGE scar-mass relative error from 0.44 to 0.37; postprocessing lifts our leaderboard Overall score from 0.70 to 0.74.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/CMRSeg_079.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: https://papers.miccai.org/miccai-2026-sat/supp/CMRSeg_079_supp.pdf
Link to Open Review
Open Review Page: https://openreview.net/forum?id=yOOoGjuj0D
BibTex
@InProceedings{SinGag_NativeResolution_MICCAISAT2026,
author = { Singh, Gagneet AND Dukre, Adinath Madhavrao AND Razzak, Imran},
title = { { Native-Resolution Multi-Sequence Cardiac MRI Segmentation for Accurate LVEF and Scar Quantification } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17268},
month = {pending},
page = {pending}
}