Abstract
We describe a modality-specific method for the Mitral Valve Anatomy Analysis (MVAA) challenge, which spans cardiac CT, 3D transesophageal echocardiography (TEE), and surgical video segmentation. For CT, a four-source 3D ensemble supplies a base mask; signed-distance scores and multi-model votes then rank candidate voxels on connected one-voxel boundary shells only, and a tri-planar ConvNeXt-V2 FCMAE refines the result. For TEE, a 5-fold nnU-Net is blended with a MONAI U-Net/SegResNet ensemble before native-space, class-wise component filtering. For video, eight diverse segmenters are fused by confidence-adaptive weighting and sequentially refined by SAM~3, DINOv3, and an FCMAE. A noisy student then learns from confidence-filtered pseudo frames; teacher and student are temporally smoothed and blended before detector-guided cleanup. On the challenge test set the method attains mean Dice scores of 0.839, 0.857, and 0.833 on the three tasks. The tasks thus share a common principle of conservative probability fusion while retaining modality-specific training, geometry handling, and safeguards.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/MVAA_055.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=4EKSTQSAFF
BibTex
@InProceedings{ShiAsu_AModalitySpecific_MICCAISAT2026,
author = { Shin, Asuka AND Shigematsu, Kosuke AND Inoue, Kazuyuki},
title = { { A Modality-Specific Ensemble for Multimodal 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}
}