Abstract
Mitral valve anatomy analysis requires robust interpretation of heterogeneous imaging data, including cardiac computed tomography (CT), three-dimensional transesophageal echocardiography (3D TEE), and intraoperative surgical video frames. In this report, we describe our solution for the MVAA 2026 challenge, which formulates all three modalities as segmentation tasks. Our solution follows a task-specific design. For cardiac CT and 3D TEE, we use nnU-Net v2 as a self-configuring 3D segmentation framework and train five-fold ensembles with task-specific preprocessing, spacing, patch size, normalization, and inference settings determined from the training data. The final CT model further incorporates 100 high-confidence pseudo-labeled CT volumes through self-training. For surgical video frames, we use an ensemble of five 2D UNet++ models with ResNet34, EfficientNet-B4, and EfficientNet-B5 ImageNet-pretrained encoders. At inference time, branch probabilities are resized to the original frame resolution, combined with fixed weights, and binarized using a threshold of 0.40 with test-time augmentation. On the organizer-side online validation server, the v35 submission achieves DSC/HD/ASD of 0.8578/4.6716/0.2741 for Task 1, 0.8463/11.1969/0.6406 for Task 2, and 0.8049/73.7878/12.2489 for Task 3. In the Docker final evaluation, the same v35 submission achieves mean DSC/HD/ASD of 0.8382/6.2431/0.5575 for Task 1, 0.8581/9.2974/0.4942 for Task 2, and 0.7602/494.1503/381.5384 for Task 3. The final inference workflow fixes checkpoints, preprocessing, thresholds, output naming rules, and packaging for reproducible organizer-side evaluation; code is available at https://github.com/ye-yang-ai/mvaa-competition-code.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/MVAA_053.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=e5B6MJKrR9
BibTex
@InProceedings{YanYe_TaskSpecific_MICCAISAT2026,
author = { Yang, Ye AND Zhang, Jiansong AND Wu, Yongji AND Yang, Shilong AND Shen, Linlin},
title = { { Task-Specific Multimodal Mitral Valve Segmentation for MVAA 2026 } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17269},
month = {pending},
page = {pending}
}