Adaptive Spectral Graph Attention Filtering Network for Alzheimer's Disease Classification Using Multimodal Data.

Area of Science:

• Neuroimaging
• Artificial Intelligence
• Computational Neuroscience

Background:

• Early detection of Alzheimer's Disease (AD) is crucial for effective intervention.
• Current graph-based methods for brain network analysis often overlook spectral information.
• Integrating functional and structural brain network data can improve diagnostic accuracy.

Purpose of the Study:

• To develop an advanced network analysis model for enhanced Alzheimer's Disease (AD) classification.
• To leverage spectral-domain information from brain networks for improved diagnostic performance.
• To introduce the Adaptive Spectral Graph Attention Filtering Network (ASGAFN) for AD detection.

Main Methods:

• Constructed structural and functional brain network graphs using diffusion tensor imaging (DTI) and resting-state functional magnetic resonance imaging (rs-fMRI).
• Employed a frequency-encoding-guided attention mechanism for adaptive spectral filter learning.
• Incorporated a spectral energy sensing module for graph-specific adaptation and a Multimodal Fusion and Enhancement Layer (MFEL) for signal integration.

Main Results:

• ASGAFN achieved high classification accuracies: 96.64% (AD vs. NC), 90.48% (MCI vs. NC), and 91.75% (AD vs. MCI).
• The model demonstrated strong performance in a three-class classification task (AD vs. MCI vs. NC) with 87.12% accuracy.
• ASGAFN significantly outperformed existing baseline methods in AD-related classification tasks.

Conclusions:

• The proposed ASGAFN effectively models spectral structures in brain networks for improved AD classification.
• Spectral-domain modeling and multimodal fusion are effective strategies for enhancing diagnostic accuracy in neurodegenerative diseases.
• The ASGAFN framework offers a promising tool for early and accurate detection of Alzheimer's Disease stages.