Abstract
Breast multiparametric MRI (mpMRI) provides complementary diagnostic information, yet conventional fusion methods often rely on fixed mappings that cannot accommodate patient-specific variations in multimodal characteristics. To address this limitation, we propose a sample-adaptive framework for benign–malignant breast lesion classification. First, independent three-dimensional encoders extract modality-specific features from T2WI, DCE2, DWI, and ADC images. These features are projected into a shared token space, where a Transformer models cross-modal dependencies and produces a joint patient representation. Subsequently, a mixture-of-experts (MoE) module employs an input-dependent gate to generate patient-specific routing weights. Multiple dense residual experts process the joint representation in parallel, and their outputs are adaptively aggregated according to the routing weights, enabling different expert compositions for different patients. A Kullback–Leibler load-balancing objective further regulates aggregate expert utilization while preserving nonuniform routing for individual patients. Experiments on an in-house breast mpMRI cohort demonstrate that the proposed method outperforms the evaluated fusion methods, while ablation studies confirm the complementary benefits of cross-modal modeling and sample-adaptive expert fusion.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/CLIP_007.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=MrylvAxItC
BibTex
@InProceedings{WuYuy_SampleAdaptive_MICCAISAT2026,
author = { Wu, Yuyang AND Yue, Xiaodong AND He, Yongsheng AND Chen, Yufei AND Qi, Xuan AND Wei, Zhipeng AND Huang, Yushuo},
title = { { Sample-Adaptive Cross-Modal Expert Fusion for Robust Breast mpMRI Classification } },
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}
}
