Abstract
Mitral valve anatomy analysis is essential for cardiac assessment and surgical planning, yet robust segmentation remains challenging because annotations are scarce and clinical modalities are heterogeneous. Existing methods are often designed for one modality and use uniform pseudo-labeling rules without anatomical priors. We propose MVAA-Seg, a multi-stage framework for the MICCAI 2026 Mitral Valve Anatomy Analysis (MVAA) Challenge, covering computed tomography (CT), three-dimensional transesophageal echocardiography (3D TEE), and surgical video. For CT, a confidence-guided semi-supervised nnUNetv2 pipeline ranks pseudo-labels using prediction confidence and anatomical plausibility, exploiting unlabeled data while rejecting unreliable masks. For TEE, we use a 3D nnUNetv2 ResEnc M pipeline trained for 500 epochs on five folds and ensembled at inference. For video, we use UNet++ with an EfficientNet-B5 encoder. On the MVAA validation set, MVAA-Seg (Team: NPM) improves over the official baseline in all tasks, with DSC of 0.8585, 0.8543, and 0.8105 on CT, TEE, and video, respectively.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/MVAA_033.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=pv9UIXIAI0
BibTex
@InProceedings{ZhaLiy_MVAASeg_MICCAISAT2026,
author = { Zhang, Liyang AND Xu, Qing AND He, Xiangjian AND Chen, Zhen},
title = { { MVAA-Seg: A Multi-Stage Segmentation Framework for Mitral Valve Anatomy Analysis } },
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}
}