Abstract
Multi-sequence cardiac magnetic resonance (CMR) imaging provides vital structural and pathological cues for precise cardiac pathology segmentation (scar and edema). Despite its clinical potential, existing multi-modal methods for automated segmentation often suffer from suboptimal feature fusion and severe scarcity of diverse pathological data. In this work, we propose a novel multi-modal learning framework combining self-supervised representation learning with generative model based data augmentation to alleviate the gap. First, we introduce a Foundation model-Augmented Dynamic Mixture-of-Experts (FAD-MoE) architecture where each expert model’s feature space is augmented by the richer, discriminative visual features from the visual foundation model: DINOv3. These experts are then dynamically coordinated via a smart router to optimally weigh sequence-specific contributions for final decision making. Second, we introduce a dual diffusion-based region-aware pathological data augmentation strategy to synthesize diverse, realistic scar and edema patterns alongside structural deformations with two advanced deep generative models, capturing diverse yet realistic variations at both tissue and structural levels. Evaluations on the CARE 2026 MyoPS validation platform demonstrate the proposed methodology achieves Dice scores of 70.21 for scar regions and 72.15 for edema. Compared with traditional early-fusion nnU-Net baselines, which simply take multiple sequences as multi-channel input, our method yields significant performance gains on both in-domain and out-of-domain test sets.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/CARE_041.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=Rk9L8B5v3t
BibTex
@InProceedings{SonZhe_Foundation_MICCAISAT2026,
author = { Song, Zhengxuan AND Li, Sijie AND Ramzan, Farheen AND Kiberu, Yusuf AND Jamil-Copley, Shahnaz AND Clayton, Richard H. AND Chen, Chen},
title = { { Foundation Model Augmented Mixture-of-Experts with Diffusion Augmentation for Robust Cardiac Myocardial Pathology 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}
}