Abstract
Accurate mitral valve segmentation across diverse imaging modalities is essential for planning cardiac interventions like Transcatheter Edge-to-Edge Repair (TEER). However, state-of-the-art models struggle to generalize due to modality-specific artifacts, including severe acoustic shadowing in 3D Transesophageal Echocardiography (TEE) and specular reflections in 2D endoscopic images. To overcome these challenges, we propose a modality-tailored segmentation pipeline. Rather than applying a uniform architecture, we deploy specialized models customized to each modality’s physical properties: a semi-supervised Swin UNETR for high-resolution 3D CT, a noise-resilient 3D U-Net optimized with a hybrid Dice-Focal Loss for speckle-prone 3D TEE, and a dense UNet++ with an EfficientNet backbone for 2D surgical captures. Furthermore, we introduce zero-parameter, anatomy-aware post-processing to drastically suppress remote false-positive artifacts, specifically employing 3D connected-component filtering for volumetric continuity and 2D safe microdenoising for disjoint leaflet projections. Comprehensive evaluations demonstrate that our modality-tailored pipeline consistently outperforms the official benchmark baseline across all tasks. Specifically, our specialized networks drive the Dice Similarity Coefficient (DSC) from 0.618 to 0.789 on 3D CT, 0.745 to 0.747 on speckle-prone 3D TEE, and 0.644 to 0.774 on 2D static screenshots.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/MVAA_057.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=hrXv5uIqM0
BibTex
@InProceedings{LiMei_ModalityTailored_MICCAISAT2026,
author = { Li, Meijing AND Zhang, Zhihan AND Li, Chenxi AND Zhai, Ruichao AND Liu, Xinyang AND Li, Haodong AND He, Mengqi AND Sun, Yulin AND He, Runnan AND Liu, Xiuyun AND Ming, Dong},
title = { { Modality-Tailored Mitral Valve Segmentation: Custom Architectures and Robust Noise Suppression } },
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}
}