An Alzheimer's Disease classification network based on MRI utilizing diffusion maps for multi-scale feature fusion in...

Area of Science:

• Neuroimaging
• Machine Learning
• Medical Diagnostics

Background:

• Graph Convolutional Networks (GCNs) are effective for Alzheimer's disease (AD) classification using MRI data.
• GCN performance can be limited by adjacency matrix construction, potentially missing patient correlations.
• Accurate AD classification is crucial for early diagnosis and intervention.

Purpose of the Study:

• To propose a novel GCN-based network for AD classification using MRI data.
• To address limitations in feature representation by integrating diffusion maps for multi-scale feature fusion.
• To improve the accuracy and generalization of AD classification.

Main Methods:

• Utilized diffusion maps to extract features independent of the adjacency matrix.
• Integrated diffusion features from varying neighbor counts with a self-attention mechanism for adaptive weighting.
• Employed metric learning to refine feature similarity and dissimilarity within and between classes.
• Validated the approach on the Alzheimer's Disease Neuroimaging Initiative (ADNI) dataset.

Main Results:

• The proposed method demonstrated competitive performance on AD classification tasks.
• The approach effectively captures patient characteristics and intrinsic relationships.
• Achieved superior generalization capabilities compared to existing methods.

Conclusions:

• The novel GCN approach with diffusion maps enhances AD classification accuracy.
• This method offers a valuable tool for early AD diagnosis and personalized treatment decisions.
• The findings provide significant insights for clinical applications in neurodegenerative disease management.