Classification of Alzheimer's Disease by Modeling Brain Networks as Signed Networks Under Deep Learning Frameworks.

Alzheimer's disease (AD) is a progressive neurodegenerative disorder that remains a global challenge due to its complex pathology and the lack of definitive diagnostic tools. This paper introduces an innovative approach to predicting and analyzing Alzheimer's disease by constructing signed brain network models and leveraging signed graph neural network technologies. By modeling the brain network as a signed graph that incorporates both positive and negative correlations, we capture the nuanced interactions between brain regions more effectively than traditional methods. We utilize graph convolutional networks (GCNs) and their variants to process these signed brain networks, significantly improving the accuracy of Alzheimer's disease prediction. Comparative analysis reveals that the signed graph model outperforms its unsigned counterparts in diagnostic precision (with an improvement of at least 19%), emphasizing the importance of incorporating negative correlations in neural interactions. Furthermore, precisely because of the additional negative edge information that we can utilize both positive and negative attention matrices, derived from these prediction tasks, to determine important brain region biomarkers. This work is an attempt to systematically validate the role of negative information through comparisons of different signed graph variants, which holds particular promise for enhancing Alzheimer's disease diagnostic accuracy at early stages. We believe that this approach will have significant clinical applications in the future.