Early Alzheimer's disease classification via structure and feature-based graph attention network from multi-center...

Area of Science:

• Neuroscience
• Medical Imaging
• Artificial Intelligence

Background:

• Alzheimer's disease (AD) poses a significant global health challenge, necessitating advancements in early diagnostic tools.
• Multi-modal data analysis offers potential for enhanced feature expression and improved diagnostic accuracy in AD.
• Current diagnostic methods often lack the sensitivity required for early and precise AD detection.

Purpose of the Study:

• To develop a novel graph attention network for integrating multi-modal data (imaging and non-imaging) for improved Alzheimer's disease diagnosis.
• To leverage structure-based and feature-based approaches for robust feature extraction and classification.
• To enhance the diagnostic performance by fusing information from various sources and utilizing attention mechanisms.

Main Methods:

• A group sparse representation approach was used to constrain correlations between blood-oxygen-level-dependent (BOLD) signals in brain functional networks.
• Structural brain networks were utilized to supervise the construction of biologically significant functional networks.
• A sparse graph was constructed incorporating subject demographics (gender, age) and multi-center neuroimaging features.
• Individual and common embeddings were extracted for classification using a graph attention network with parameter sharing and attention mechanisms.

Main Results:

• The proposed graph attention network effectively integrates multi-modal data, including BOLD signals and demographic information.
• Fusion of topological structure and feature information via attention mechanisms significantly improved diagnostic performance.
• Experimental results on the Alzheimer's Disease Neuroimaging Initiative (ADNI) and an in-house dataset demonstrated superior performance compared to conventional methods.

Conclusions:

• The novel graph attention network provides a powerful framework for early Alzheimer's disease diagnosis by effectively integrating multi-modal data.
• The method achieves superior diagnostic accuracy, highlighting the potential of advanced machine learning techniques in neurodegenerative disease research.
• This approach offers a promising avenue for improving patient outcomes through earlier and more accurate detection of Alzheimer's disease.