Abstract
We describe our submission to the MICCAI 2026 Universal Multi-Sequence, Multi-Center and Multi-View CMR Segmentation Challenge. The challenge couples two imaging sequences (Cine and Late Gadolinium Enhancement), three to four acquisition views per sequence, a heterogeneous label set that spans the ventricles, myocardium, atria, and myocardial scar, and two clinical estimates (ejection fraction and scar mass). Rather than force one network to absorb every view and label configuration, we build per-view specialist models and route each input to its matching specialist. We developed two variants of this idea and compared them. The first uses nnU-Net v2 with a residual-encoder backbone, cross-vendor intensity augmentation, and a scar-focused cascade with an intensity-based refinement step. The second replaces the backbone with MedNeXt, splits the network configuration by view geometry (3D for the volumetric short-axis stacks, 2D for the sparse long-axis views), trains all annotated Cine phases as a single volume, and preserves the model’s raw scar prediction instead of refining it. The MedNeXt variant produced our stronger result and is the submitted system. It reached an overall validation score of 0.67 with a Cine ejection-fraction correlation of 0.94 obtained by slice-information-guided temporal unpacking, where the intensity-based scar refinement proposed in the first variant, when measured against ground-truth Late Gadolinium Enhancement, was found to remove between 50 and 96 percent of true scar and was therefore disabled. We report five-fold cross-validation and held-out validation results, together with the scar-refinement ablation that motivated the final design.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/CMRSeg_032.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: https://papers.miccai.org/miccai-2026-sat/supp/CMRSeg_032_supp.pdf
Link to Open Review
Open Review Page: https://openreview.net/forum?id=tLNJunQ6aI
BibTex
@InProceedings{SidSaa_PerView_MICCAISAT2026,
author = { Siddiqui, Saaim AND Naqvi, Syed Haider Abbas AND Ankenbrand, Markus},
title = { { Per-View Specialist Networks for Multi-Sequence, Multi-View Cardiac MRI Segmentation and Clinical Quantification: A Comparison of nnU-Net and MedNeXt Backbones } },
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}
}