Geometrical Deep Learning: A Feasible Methodology for Brain Age Prediction

Abstract

The dynamic nature of the human brain through aging brings forth structural and functional changes that contribute to cognitive decline and increase the risk of neurodegenerative disorders. Predicting brain age, particularly through deviations from chronological age, can serve as a significant indicator of pathological aging. In this study, we integrate geometrical deep learning with neuroimaging data to predict brain age. We construct Graph Convolutional Networks by using resting-state magnetoencephalography data and structural connectivity measures derived from diffusion-weighted magnetic resonance imaging. These networks incorporate cortical morphometry and spectral features as node annotations to capture the multivariate patterns of age-related brain changes. Our findings highlight the effectiveness of Graph Convolutional Networks in leveraging functional and structural brain connectivity to predict brain age. This approach underscores the potential of graph-based deep learning methods in advancing our understanding of brain aging and its implications for neurodegenerative conditions.