Abstract
Early identification of cognitive impairment requires integrating complementary functional, structural, and clinical evidence. However, existing multimodal brain network methods often distort hierarchical topology, align imaging modalities without anatomical or subject-specific context, and use fixed regional correspondence. We propose LLM-HypSFCN, an LLM-guided hyperbolic structure-function coupling network. Dual hyperbolic kernel graph attention network (HKGAT) encoders first preserve modality-specific connectome hierarchy. In parallel, a frozen LLM-based encoder represents medical descriptions from the 90-region Automated Anatomical Labeling atlas and label-free subject metadata, while individualized semantic gates inject these priors into functional and structural ROI embeddings. Tri-modal HyperGRAM directly aligns the resulting functional, structural, and textual representations through higher-order Euclidean-Lorentz Gramian volumes. Adaptive coupling, second-stage HKGAT fusion, and Lorentz hyperbolic prediction further model subject-specific interactions while maintaining geometric consistency. Experiments on Alzheimer’s Disease Neuroimaging Initiative connectomes evaluate the framework against conventional, graph-based, and multimodal baselines.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/ELAMI_021.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to Open Review
Open Review Page: Not Available
BibTex
@InProceedings{ZheJun_LLMHypSFCN_MICCAISAT2026,
author = { Zheng, Junhua AND Li, Jiaqiang AND Cheng, Nina AND Wang, Tianfu AND Lei, Baiying AND Lei, Haijun AND Yang, Peng},
title = { { LLM-HypSFCN: LLM-Guided Hyperbolic Structure-Function Coupling Network for Early Cognitive Assessment } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17262},
month = {pending},
page = {pending}
}