Abstract
We present our solution to the MVAA 2026 challenge for mitral valve segmentation across cardiac CT, 3D TEE ultrasound, and surgical video, addressing each task’s dominant failure mode with a targeted innovation. For Task 1, Geometry-Aware Dual-Teacher learning (GA-DT) tackles extreme label scarcity by forming unlabeled targets from the gated agreement of an EMA teacher and a frozen peer, and adds a signed-distance dual task for distance-aware supervision. For Task 2, Opening-Plane Consistency (OPC) resolves fusion of the two leaflets by rendering them en-face in a case-canonical opening-plane frame and enforcing a coaptation margin between their footprints; it is parameter-free and backbone-agnostic. For Task 3, Video Appearance-Conditioned Adaptation (VACA) counters cross-video appearance shift with a VideoFiLM modulation, driven by an appearance bank distilled from unlabeled videos, that specializes decoder features per video under agreement-gated consistency.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/MVAA_090.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=DLzCTiNzMG
BibTex
@InProceedings{HouWen_Mitral_MICCAISAT2026,
author = { Hou, Wenlong AND Wang, Juncheng AND Qin, Jing AND Wang, Shujun},
title = { { Mitral Valve Anatomy Segmentation for Multimodal Data } },
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}
}