Explainable Graph Attention Networks for Identifying Structural Brain Connectivity Changes Across Normal Aging

Abstract

Normal brain aging refers to the gradual and natural changes in brain structure and function that occur as individuals grow older, in the absence of neurodegenerative diseases such as Alzheimer’s disease. Identifying biomarkers of normal aging is essential for distinguishing healthy brain aging from pathological conditions and for understanding mechanisms of cognitive resilience, especially as the global aging population continues to increase. Although numerous studies have investigated biological age estimation and age-related functional and structural brain changes using neuroimaging, the application of modern techniques such as explainable artificial intelligence to comprehensively analyze brain aging remains relatively underexplored. In this study, we propose a deep learning-based framework utilizing graph neural networks to identify and interpret biomarkers of normal brain aging. Diffusion-weighted and T1-weighted MRI data from healthy individuals were obtained from the publicly available IXI dataset and categorized into three lifespan stages (young: 20–39y; middle-aged: 40–59y; older: 60–79y). Structural brain networks were constructed, and graph attention networks (GATs) were employed for pairwise age-group classification (young vs. middle-aged; middle-aged vs. older), achieving high classification performance. Furthermore, GNN explainability methods revealed key discriminative brain regions and connections associated with age-related connectivity changes, highlighting trajectories within the default mode, attention, and salience networks as plausible biomarkers of normal brain aging.