Abstract
The Mitral Valve Anatomy Analysis (MVAA) 2026 challenge requires segmenting a single valve across three very different modalities: a small labelled cardiac CT cohort, thin coapting leaflets in 3-D transesophageal echocardiography (TEE), and sparsely annotated surgical video. We build a separate system per modality and segment directly in the image grid rather than fitting a leaflet surface model, because the folded, coapting leaflet is hard to represent as a single surface from so few labelled cases, and we give the CT network an explicit descriptor of the leaflet’s thin-sheet shape through a 47-channel bank of multi-scale image derivatives and sheetness features. On the public validation set this CT model reached 0.8627 DSC, 4.18 mm $HD_{100}$, and 0.257 mm ASD, with a fifteen-model residual nnU-Net ensemble a close and more conservative reference at 0.8596. In TEE, conservative cleanup cut $HD_{100}$ from 12.23 to 10.49 mm at Dice parity (0.8363). For surgical video, a semi-supervised DINOv3 ensemble trained with Mean Teacher reached 0.7917 DSC (84.1 $HD_{100}$, 14.8 ASD, in pixels), and blending in a hard-negative fine-tuned member at 15%, which we submitted, reached 0.8174 (71.3, 12.6).
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/MVAA_099.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=f8PCDuY1eE
BibTex
@InProceedings{GazMat_GeometryAware_MICCAISAT2026,
author = { Gazda, Matej AND Paik, David},
title = { { Geometry-Aware and Semi-Supervised Ensembles for Multimodal Mitral Valve 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}
}