Abstract
Atrial fibrillation (AF) research increasingly relies on virtual clinical trials and digital twins, yet the development of large populations of atrial models with anatomical variability is constrained by the limited availability of late gadolinium-enhanced magnetic resonance imaging (LGE-MRI). Recent mask-conditioned diffusion models have, therefore, emerged as a practical approach for generating realistic LGE-MRI appearance from available anatomical labels, thereby producing paired image–label data for image-based cardiac analysis. However, for digital twin applications, anatomical realism alone is insufficient, as accurate threedimensional reconstruction also requires consistent structural transitions across adjacent slices within a volumetric image stack. Most existing diffusion models generate slices independently and consequently fail to preserve anatomical continuity along the z-axis, which may compromise volumetricreconstructionquality.WeproposeInter-SliceAdaptivelyDistilled ControlNet (IS-ADC), a continuity-aware diffusion framework that incorporates neighbouring slice information through a teacher-only Inter- Slice Attention module. The learned continuity priors are subsequently transferred to a mask-conditioned student branch via adaptive distillation, improving volumetric consistency while requiring only a modest increase in training cost and no additional inference-time computation. Experiments on the Left Atrial Segmentation Challenge 2018 dataset demonstrate that IS-ADC achieves better image fidelity and downstream segmentation performance than both the original ADC framework and a three-slice-input baseline. Furthermore, IS-ADC improves inter-slice anatomical continuity and structural integrity, producing more coherent three-dimensional left atrial reconstructions.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/DT4H_016.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=7aoPTkOdaj
BibTex
@InProceedings{ZhaMia_Improving_MICCAISAT2026,
author = { Zhao, Miao AND Li, Shibo AND Chen, Xu AND Al-Aidarous, Sayed AND Honarbakhsh, Shohreh AND Osmani, Venet AND Slabaugh, Gregory AND Roney, Caroline H.},
title = { { Improving Anatomical Continuity in Synthetic Left Atrial LGE-MRI Using Inter-Slice Attention } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17275},
month = {pending},
page = {pending}
}