Abstract
Accurate left atrial (LA) scar segmentation in late gadolinium enhancement magnetic resonance imaging (LGE-MRI) is a prerequisite for image-based scar quantification in atrial fibrillation and assessment of ablation outcome. Scar voxels occupy a small proportion of the image and can be confounded by imaging artifacts and adjacent bright tissues. Consequently, full-volume segmentation without anatomical constraints can produce false-positive predictions. We propose a wall-band anatomical prior-guided segmentation framework. A predicted LA cavity mask defines a local region of interest (ROI), from which a wall-band prior is constructed by three-dimensional dilation and erosion. The cropped LGE-MRI and the prior form a two-channel input to a 3D nnU-Net. The five-fold ensemble output is resampled and converted to scar probabilities by softmax; a fixed threshold produces the binary mask, which is restored to the original image space. The framework achieved G-DSC 0.4361, ACC 0.7549, and SEN 0.5099 on the test set. By reusing the predicted LA cavity mask for both ROI localization and wall-band construction, the framework converts sparse target search in the complete volume into local anatomically guided prediction around the atrial wall.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/CARE_024.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=tXCGPUhtnh
BibTex
@InProceedings{QiYif_WallBand_MICCAISAT2026,
author = { Qi, Yifan AND Huang, Haojie AND Xu, Zhezhuang},
title = { { Wall-Band Anatomical Prior-Guided 3D nnU-Net for Left Atrial Scar Segmentation in LGE-MRI } },
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}
}
