Abstract
Multi-modal perioperative mitral valve segmentation is critical for cardiac surgical planning, yet it faces challenges including blurred tissue boundaries, heavy occlusion, and substantial domain discrepancies among cardiac CT, 3D TEE and surgical endoscopy. This work proposes a hybrid-supervised multi-task framework equipped with modality-adaptive subnetworks for the three imaging modalities. We deploy the UniMatch EMA teacher-student semi-supervised paradigm for CT and endoscopic frames, whereas Vista3D SegNet is trained in a fully supervised manner for 3D TEE segmentation. DINOv2-DPT serves as the feature extraction backbone for intraoperative endoscopic frames. Multi-scale feature refinement and boundary-aware compound loss are introduced to mitigate jagged contours and incomplete segmented regions. Quantitative experiments in terms of DSC, HD and ASD demonstrate steady performance improvements across all modalities, and qualitative visualization verifies that our framework generates smooth, anatomically consistent mitral valve outlines. The code is publicly available at \url{https://github.com/Daheilou/MICCAI2026_MVAA}.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/MVAA_084.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=IymxMulVFi
BibTex
@InProceedings{QiYue_HybridSupervised_MICCAISAT2026,
author = { Qi, Yue AND Li, Zheng AND Yu, Junxuan AND Yang, Xin},
title = { { Hybrid-Supervised Multi-Task Network for Multi-Modal Perioperative Mitral 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}
}