Abstract
Mitral-valve assessment guides diagnosis, planning, and intraoperative care for structural heart disease across three visually distinct modalities: cardiac CT for valve geometry, 3D transesophageal echocardiography (TEE) for volumetric function, and surgical video for leaflet-motion confirmation. The MVAA 2026 challenge asks one team to segment the mitral valve in all three. We present a unified solution matched to each data regime: for CT, an nnU-Netv2 with confidence-gated self-training over 1040 unlabeled volumes; for TEE, a single supervised nnU-Netv2 on all 105 labeled cases; for video, a two-model ensemble with surgical and ImageNet encoder pretraining, mean-teacher self-training, and test-time augmentation. Each task respects a single-model-per-task budget of $\leq$10,s/case on a 12,GB GPU. On the hidden validation sets we obtain Dice scores of 0.861 (CT), 0.845 (TEE), and 0.814 (video), an 8–29% relative DSC gain over the reproduced organizer baseline on every task. We further contribute a cross-modal analysis of what transfers, including internal gains that failed on held-out data.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/MVAA_097.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: https://papers.miccai.org/miccai-2026-sat/supp/MVAA_097_supp.pdf
Link to Open Review
Open Review Page: https://openreview.net/forum?id=XUOT3HIXlq
BibTex
@InProceedings{AgrHar_MitralValve_MICCAISAT2026,
author = { Agrawal, Harshit},
title = { { Mitral-Valve Segmentation Across CT, TEE, and Surgical Video } },
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}
}