Multi-model adaptive fusion-based graph network for Alzheimer's disease prediction

Area of Science:

• Neuroscience
• Computer Science
• Medical Imaging

Background:

• Alzheimer's disease (AD) is a prevalent cognitive disorder requiring accurate diagnostic tools.
• Deep learning, particularly graph neural networks (GNNs), shows promise for AD prediction by modeling population-level relationships.
• Current GNN methods often rely on single models, limiting algorithm selection and data integration for improved diagnosis.

Purpose of the Study:

• To develop and evaluate a multi-model fusion framework for enhanced Alzheimer's disease prediction.
• To address the limitations of single-model approaches in graph-based disease prediction.
• To integrate diverse data patterns into a unified model for superior diagnostic quality.

Main Methods:

• A spectral graph attention model was employed to aggregate intra- and inter-cluster node embeddings.
• A bilinear aggregation model served as an auxiliary component to enhance abnormality detection.
• An adaptive fusion module dynamically integrated outputs from both models for improved AD prediction.

Main Results:

• The proposed multi-model fusion framework demonstrated superior performance compared to existing methods.
• The spectral graph attention and bilinear aggregation models effectively captured population-level data.
• The adaptive fusion module successfully enhanced the accuracy of Alzheimer's disease prediction.

Conclusions:

• The developed multi-model fusion framework offers a significant advancement in computer-aided diagnosis for Alzheimer's disease.
• This approach overcomes the challenges of single-model limitations and data integration in graph-based prediction.
• The findings suggest a promising direction for improving the early and accurate detection of Alzheimer's disease.