Abstract
We describe our entry to the 2026 Mitral Valve Anatomy Analysis (MVAA) Challenge for segmentation in 3D cardiac CT, 3D transesophageal echocardiography (TEE), and surgical video. For CT and TEE, modality-specific coarse-to-fine 3D U-Nets localize the valve in the full volume and then segment it at high resolution within the resulting region of interest. For surgical video frames, we distill dense DINOv3-L features into a compact PVTv2-B2 encoder and train a UPerNet segmentation model. Tasks1 and3 exploit unlabeled data through a common offline strategy: pseudo-labels are ranked, uncertainty-masked, and used for fold-wise retraining, whose models are retained only when they improve held-out validation performance. Final predictions use uniform-logit cross-validation ensembles with five folds for CT and TEE and six leave-one-recording-out folds for surgical video.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/MVAA_104.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=1tnAvijH70
BibTex
@InProceedings{FagEir_Multimodal_MICCAISAT2026,
author = { Fagerbakke, Eirik AND Stenhede, Elias AND Ranjbar, Arian},
title = { { Multimodal Mitral Valve Segmentation with Cross-Validation Ensembles and Pseudo-label Retraining } },
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}
}