Fully Hyperbolic Graph Convolutional Neural Networks for Age Prediction with Multi-Modal Brain Data

Abstract

Characterizing age-related alterations in MEG brain networks holds great promise in understanding aging trajectories and revealing aberrant patterns of neurodegenerative disorders, such as Alzheimer’s disease. In this study, we utilize a Fully Hyperbolic Neural Network (FHNN) to embed functional brain connectivity graphs, derived from magnetoencephalography (MEG) data, into low dimensions on a Lorentz Hyperboloid model for hyperbolic space. Using these embeddings, we aim to detect changes in the intrinsic hierarchy of functional subnetworks across time as well as predict age for patients across multiple decades. We use the hyperbolic embedding pipeline in tandem with multimodal MEG and MRI data to create embeddings from the Cam-CAN (Cambridge Centre for Ageing and Neuroscience) dataset for the downstream task of brain age prediction in healthy patients to better understand how brain connectivity structure impacts brain aging trends. Our hyperbolic MEG brain network embedding framework effectively transforms high-dimensional complex MEG brain networks into lower-dimensional hyperbolic representations, facilitating structural brain hierarchy analysis across age, as well as age prediction. Our versatile embedding pipeline allows for the ready implementation of other downstream tasks like clustering and classification. This constitutes a novel way of studying connectivity alterations in brain networks.