Exploring Healthy Neurocognitive Ageing with Deep Learning Interpretability Methods

Abstract

Healthy ageing reorganises large-scale functional brain networks. Using resting-state fMRI from 615 adults aged 18 to 88 years in Cam-CAN. We combined raw functional-connectivity analyses, graph-theoretical topology, Ridge and MLP age prediction and their attributions in order to understand how functional brain organisation varies with age and relates to cognitive performance. Both models predicted age accurately, with a modest advantage for the MLP. Within-network FC decreased with age, while between-network FC and participation increased. Topological analyses showed increased participation and a redistribution of nodes across connector, provincial, satellite, and peripheral roles. The MLP SmoothGrad attributions emphasised high-participation and weakly specialised nodes, whereas counterfactual tests showed that replacing CON-SMN, CON-DMN and DMN-FPN connections with a young-adult template produced the largest reductions in predicted age. Brain-cognition PLS linked these mechanisms to a broad cognitive-performance axis and a secondary semantic/connectomic axis. Together, these findings suggest that healthy ageing involves a redistribution of functional organisation that extends beyond a single DMN-FPN axis towards a broader DMN-CON-SMN-FPN configuration.