Abstract
Accurate segmentation of multiple left atrial structures is important for anatomical reconstruction in radiofrequency catheter ablation. In clinical practice, window width and level are commonly adjusted to enhance structural features in CT images. However, conventional hard windowing relies on manually specified window settings and clips intensity information outside the predefined range. The task is further challenged by limited annotated data and the need for accurate boundary delineation. In this paper, we propose AGSW-LAMS, an adaptive Gaussian soft-windowing framework for left atrial multi-structure segmentation. AGSW-LAMS replaces conventional hard windowing with multiple learnable Gaussian windows, whose center and scale parameters are jointly optimized with the segmentation network to generate structure-sensitive intensity representations. A Mean Teacher architecture is employed to exploit unlabeled data, while a boundary loss is incorporated to improve the delineation of complex anatomical structures. Continual test-time adaptation and connected-component post-processing are further applied during inference to improve prediction robustness. On the CARE 2026 left atrial multi-structure segmentation dataset, AGSW-LAMS achieves a DSC of 0.9549 and an HD of 9.2927 on the official test set.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/CARE_022.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=oQM0BPRPij
BibTex
@InProceedings{WanFan_AGSWLAMS_MICCAISAT2026,
author = { Wang, Fangchen AND Zhang, Xiao AND Chen, Xiaoan AND Luo, Shengda AND Chen, Jingkun},
title = { { AGSW-LAMS: Adaptive Gaussian Soft Windowing for Left Atrial Multi-Structure Segmentation } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17273},
month = {pending},
page = {pending}
}
