Abstract
Multi-view cine cardiac magnetic resonance (CMR) segmentation must accommodate two-chamber (2CH), four-chamber (4CH), and short-axis (SAX) acquisitions whose sampling organizations and label spaces differ. Existing 2D/3D ensembles are effective for a single volumetric task but do not directly resolve which representation should process each cine view or how to prevent cross-view interference. We present the Geometry-Aware Hybrid Network (GAHN), a deterministic view-routed framework that assigns 2CH/4CH to an anchor-corrected 2.5D branch and SAX to a separately pretrained 3D context branch. The long-axis branch uses centered temporal stacks, task conditioning, and uncertainty-guided residual correction; the two branches retain disjoint parameters and gradient paths. Internal development selection identified a synchronized epoch-4 checkpoint. Without parameter updates, this checkpoint achieved Dice scores of $0.8751$, $0.8946$, and $0.8544$ for 2CH, 4CH, and SAX, respectively (mean $0.8747$), on the separately released annotated CMR-MULTI validation cohort. The routed model has $16.13$M parameters and requires $2.67$ s per three-view patient. In a matched four-epoch development control, random rather than pretrained SAX initialization reduced SAX Dice from $0.8943$ to $0.4621$. The results support geometry-aware routing and stable pretrained context modeling, while motivating phase-explicit SAX representations and repeated evaluation.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/CMRSeg_051.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=DmXgofs3mY
BibTex
@InProceedings{ZhoQik_GeometryAware_MICCAISAT2026,
author = { Zhong, Qikai AND Lu, Jiawen AND Yu, Zixin AND Zhu, Renjun AND Guo, Qiaojuan AND Wu, Xiangjun},
title = { { Geometry-Aware View Routing with Hybrid 2.5D–3D Context for Multi-View Cine CMR Segmentation } },
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}
}