Abstract
Accurate atrial segmentation from cardiac MRI constitutes a crucial prerequisite for the quantitative analysis of atrial anatomy and structural remodeling. It requires capturing both low-frequency global structures for overall atrial cavities’ geometry and high-frequency local details for atrial walls and small protrusions. However, standard convolutions struggle to handle both simultaneously, while dilated convo- lutions, though enlarging the receptive field, attenuate high-frequency responses due to sparse sampling. Our frequency analysis reveals that different convolutional patterns exhibit distinct responses across frequency sub-bands. Motivated by this observation, we propose the Frequency-Decoupled Wavelet Network (FDW-Net), the first segmentation framework that integrates dilated convolutions with wavelet-based frequency decomposition. FDW-Net decomposes features into distinct sub-bands and assigns dedicated convolutional operators via the Frequency-Decoupled Wavelet Fusion (FDWF) modules, while the Channel Mixer (CM) mod- ules with the Local-Enhanced Channel Gating (LECG) units enhance boundary-sensitive representations. Experiments on the largest publicly available cardiac datasets (MBAS and LA) demonstrate that FDW-Net outperforms state-of-the-art CNN, Transformer, and frequency-aware mod- els. Code is available at https://github.com/greeyun/FDW-Net.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/STACOM2026_001.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/profile?id=~Yun_Gu2
BibTex
@InProceedings{GuYun_FDWNet_MICCAISAT2026,
author = { Gu, Yun AND Zhong, Xinyue AND Xu, Fangqiang AND Feng, Fan AND Gunawardhana, Malitha AND Zhao, Jichao},
title = { { FDW-Net: Frequency-Decoupled Wavelet Network for Atrial Segmentation in Cardiac MRI } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17267},
month = {pending},
page = {pending}
}