Graph Neural Network Maps Local Brain Aging Across the Cortex in Health and Alzheimer’s Disease

Researchers have built a graph neural network that estimates how fast individual regions of the brain’s outer surface are aging, using nothing more than the shape and texture of the cortex measured from standard T1-weighted MRI scans. The model, described in the journal Neuroinformatics, was trained on brain images from 14,423 cognitively normal adults and produces a map of “local brain age” across the entire cortical surface at a spatial resolution of roughly 1.37 millimeters between neighboring points on the mesh.

Where a single whole-brain age estimate summarizes overall brain health, the new approach assigns an age to each location on the folded cortical sheet. To do this, the network ingests five morphometric features extracted from the MRI: cortical thickness, surface area, curvature, the gray-to-white matter intensity ratio, and sulcal depth. Each vertex of the cortical mesh carries its own combination of these measurements, and the graph structure of the mesh lets the network learn how neighboring and distant regions relate to one another when predicting local age.

According to the authors, the model achieves better test-retest performance than the existing state of the art, meaning its local age estimates are more stable when the same person is scanned repeatedly. The team also reports that the aging patterns the network identifies in Alzheimer’s disease are more biologically plausible than those produced by earlier methods.

When applied to data from the Alzheimer’s Disease Neuroimaging Initiative, the model revealed a clear hierarchy of aging across diagnostic groups. In cognitively normal adults, association cortices—the regions that integrate information across sensory and cognitive systems rather than handling a single modality—emerged as the primary sites of morphometric aging. This suggests that even typical, non-pathological aging leaves its strongest imprint on higher-order cortical networks.

In people with Alzheimer’s disease itself, significant aging appeared across the entire cortex, with the strongest effects concentrated in medial temporal regions and the cortical networks connected to them.

The study also probed which of the five morphometric features carry the most disease-specific signal. A feature ablation analysis—in which inputs are systematically removed and the model’s performance is re-examined—highlighted cortical curvature and the gray-to-white matter intensity ratio as the features preferentially sensitive to Alzheimer’s pathology. This is a technically meaningful result: it suggests that measures of cortical folding geometry and tissue contrast, rather than thickness alone, may be the most informative markers for detecting pathological aging patterns.

Crucially, the regional gaps between local brain age and chronological age were not merely anatomical curiosities. The researchers found that these gaps were significantly associated with neuropsychological measures of the cognitive impairment related to Alzheimer’s disease, linking the spatial pattern of cortical aging to clinically measured outcomes. In other words, regions that looked disproportionately old in the model’s estimates corresponded to worse performance on tests of cognition.

The authors frame the work as closing a methodological gap: although previous studies had examined which anatomical features contribute to global brain age estimates, no established framework had existed for estimating local brain age from cortical morphology. By combining high-resolution surface meshes with graph-based learning, the new model delivers both spatial precision and interpretability, since its predictions can be traced back to specific morphometric features at specific cortical locations.

The researchers write that these results demonstrate that graph neural network modeling of cortical morphometry enables biologically interpretable mapping of local brain aging with greater fidelity than prior work.

Subject of Research: Estimating region-specific brain aging from cortical morphology using graph neural networks in normal cognition and Alzheimer's disease

Article Title: Graph Neural Network Reveals the Cortical Morphology of Local Brain Aging in Normal Cognition and Alzheimer’s Disease

Article References: Anderson, S. D., Jomsky, J., Chaudhari, N. N., Chowdhury, N. F., Zheng, X. R., Irimia, A., & for the Alzheimer’s Disease Neuroimaging Initiative (2026). Graph Neural Network Reveals the Cortical Morphology of Local Brain Aging in Normal Cognition and Alzheimer’s Disease. Neuroinformatics, 24(4), Article 70. https://doi.org/10.1007/s12021-026-09808-y

Image Credits: AI Generated

DOI: 10.1007/s12021-026-09808-y

Keywords: graph neural network, brain age, cortical morphology, Alzheimer's disease, MRI, mild cognitive impairment, parahippocampal gyrus, cortical thickness, gray-white matter ratio, neuroinformatics, ADNI, machine learning