Fully Hyperbolic Neural Networks: A Novel Approach to Studying Aging Trajectories.

Area of Science:

• Neuroscience
• Computational Neuroscience
• Network Science

Background:

• Understanding brain network alterations during aging is vital for identifying neurodegenerative disorders like Alzheimer's disease.
• Magnetoencephalography (MEG) provides functional brain connectivity data crucial for such analyses.
• Existing methods may not fully capture the complex hierarchical changes in aging brain networks.

Purpose of the Study:

• To develop and apply a Fully Hyperbolic Neural Network (FHNN) for analyzing age-related changes in brain networks.
• To embed functional brain connectivity graphs from MEG data into hyperbolic space using a Lorentz model.
• To investigate hierarchical organization changes in brain subnetworks due to aging.

Main Methods:

• Utilized a Fully Hyperbolic Neural Network (FHNN) for dimensionality reduction of brain connectivity graphs.
• Employed hyperbolic embeddings on magnetoencephalography (MEG) data from 587 individuals in the Cam-CAN dataset.
• Leveraged the radius of node embeddings as a metric for hierarchical brain organization.

Main Results:

• Identified significant age-related reductions in the hierarchical organization of numerous brain subnetworks.
• Observed both gradual and rapid hierarchical changes in brain networks during aging, particularly in the elderly.
• Demonstrated that hyperbolic features are superior to traditional graph-theoretic measures for capturing age-related brain network information.

Conclusions:

• This study is the first to evaluate hyperbolic embeddings in MEG brain networks for aging research.
• Hyperbolic embeddings effectively capture age-related hierarchical alterations in brain networks.
• The findings highlight critical brain regions undergoing significant age-related changes in a large cohort.