Scholarlink

<p>Resting-state functional magnetic resonance imaging (rs-fMRI) is a key tool for characterizing functional connectivity networks in Alzheimer’s disease (AD). However, most existing classification methods rely on static, single-view connectivity, neglecting temporal dynamics and high-order relationships between subjects. To overcome this limitation, we propose a multimodal dynamic hypergraph learning (MVDHL) framework for automatic AD identification. This framework integrates spatiotemporal patterns from rs-fMRI, structural connectivity from diffusion tensor imaging, and fine-grained features from the raw signal in a hierarchical architecture. It consists of three main components: a multimodal fusion module, a raw signal feature extraction module, and a two-stream hypergraph construction module. A class balancing strategy enhances representational capabilities under data imbalance, while a cross-modal attention mechanism captures nonlinear interactions between functional connectivity, white matter metrics, and cognitive scores. A temporal convolutional network further extracts transient dynamics from the raw signal. A dual hypergraph structure is then constructed to model group-level neuropathological relationships and integrated via a probabilistic association matrix. During hypergraph convolution, a dynamic attention mechanism adaptively regulates cross-modal and cross-subject feature aggregation. Experiments on the ADNI dataset demonstrate that MVDHL achieves an accuracy of 85.71%, outperforming competing methods. By jointly modeling temporal dynamics, multimodal features, and population-level high-order patterns, this framework provides a new paradigm for analyzing complex brain connectivity in neurodegenerative diseases and has the potential to improve AD diagnosis in clinical practice.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Multi-view dynamic hypergraph learning for Alzheimer’s disease detection via fMRI analysis

  • 期刊论文

  • 发表日期：

    2026年2月12日

查看全文
  • Xinyue Yan,
  • Sun Hao,
  • Yiyu Feng,
  • Xianfu Zhang,
  • Shengxiang Xia

摘要

Resting-state functional magnetic resonance imaging (rs-fMRI) is a key tool for characterizing functional connectivity networks in Alzheimer’s disease (AD). However, most existing classification methods rely on static, single-view connectivity, neglecting temporal dynamics and high-order relationships between subjects. To overcome this limitation, we propose a multimodal dynamic hypergraph learning (MVDHL) framework for automatic AD identification. This framework integrates spatiotemporal patterns from rs-fMRI, structural connectivity from diffusion tensor imaging, and fine-grained features from the raw signal in a hierarchical architecture. It consists of three main components: a multimodal fusion module, a raw signal feature extraction module, and a two-stream hypergraph construction module. A class balancing strategy enhances representational capabilities under data imbalance, while a cross-modal attention mechanism captures nonlinear interactions between functional connectivity, white matter metrics, and cognitive scores. A temporal convolutional network further extracts transient dynamics from the raw signal. A dual hypergraph structure is then constructed to model group-level neuropathological relationships and integrated via a probabilistic association matrix. During hypergraph convolution, a dynamic attention mechanism adaptively regulates cross-modal and cross-subject feature aggregation. Experiments on the ADNI dataset demonstrate that MVDHL achieves an accuracy of 85.71%, outperforming competing methods. By jointly modeling temporal dynamics, multimodal features, and population-level high-order patterns, this framework provides a new paradigm for analyzing complex brain connectivity in neurodegenerative diseases and has the potential to improve AD diagnosis in clinical practice.