Scholarlink

Alzheimer’s disease (AD) is one of the most common neurodegenerative disorders, mainly affecting people aged 60 and above. Detecting AD at an early stage is crucial for slowing cognitive decline and mitigating irreversible brain damage. This paper presents a novel hybrid Deep Learning (DL) architecture that combines the strengths of Convolutional Neural Networks (CNNs), Graph Convolutional Networks (GCNs), and Multi-Head Self-Attention (MHSA) to automatically classify brain Magnetic Resonance Imaging (MRI) scans into four Alzheimer’s stages. Within this framework, the CNN module extracts detailed local features, the GCN models the structural connectivity among brain regions using an adaptive cosine-similarity graph, and the MHSA mechanism enhances global contextual reasoning by directing attention to the informative regions. The proposed CNN–GCN–MHSA model attained a classification accuracy of 99.68%, highlighting its effectiveness and generalization capability.

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

CNN–GCN–MHSA: A Multi-level Deep Learning Framework for Alzheimer’s Disease Classification

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Maysam Chaari,
  • Yassine Ben Ayed

摘要

Alzheimer’s disease (AD) is one of the most common neurodegenerative disorders, mainly affecting people aged 60 and above. Detecting AD at an early stage is crucial for slowing cognitive decline and mitigating irreversible brain damage. This paper presents a novel hybrid Deep Learning (DL) architecture that combines the strengths of Convolutional Neural Networks (CNNs), Graph Convolutional Networks (GCNs), and Multi-Head Self-Attention (MHSA) to automatically classify brain Magnetic Resonance Imaging (MRI) scans into four Alzheimer’s stages. Within this framework, the CNN module extracts detailed local features, the GCN models the structural connectivity among brain regions using an adaptive cosine-similarity graph, and the MHSA mechanism enhances global contextual reasoning by directing attention to the informative regions. The proposed CNN–GCN–MHSA model attained a classification accuracy of 99.68%, highlighting its effectiveness and generalization capability.