A graph neural network trained on cortical surface morphometry from 14,423 cognitively normal adults estimates local brain age (LBA) at high spatial resolution, shows better test-retest performance than the existing state-of-the-art, and yields biologically interpretable aging patterns in normal cognition, mild cognitive impairment, and Alzheimer's disease that relate to cognitive outcomes.
Key Findings
Methods
The authors introduce a graph neural network framework that estimates local brain age (LBA) across the cortical surface from T1-weighted MRI morphometry, which they report is the first such framework based on cortical morphology.
Input features were cortical thickness, surface area, curvature, gray/white matter intensity ratio (GWR), and sulcal depth.
LBA was estimated at high spatial resolution, with a mean inter-vertex distance of 1.37 mm.
The model was trained on cortical surface meshes extracted from MRIs of cognitively normal adults (N = 14,423).
Results
The GNN model achieved better test-retest performance than the existing state-of-the-art LBA approach.
The abstract states the model "achieves better test-retest performance than the existing state-of-the-art."
Specific test-retest metrics are not given in the abstract.
Results
The model identified more biologically plausible patterns of aging in Alzheimer's disease on the ADNI dataset than prior work.
Evaluation of AD-related patterns used the ADNI dataset.
The authors conclude that GNN-based cortical morphometry mapping enables interpretable local aging maps "with greater fidelity than prior work."
Results
In cognitively normal adults, association cortices emerged as the primary sites of morphometric aging.
This pattern was identified from LBA maps in CN subjects.
Results
Mild cognitive impairment was characterized by widespread cortical aging that was most pronounced in the parahippocampal gyrus.
The aging was described as widespread across the cortex.
The parahippocampal gyrus showed the strongest effect.
Results
Alzheimer's disease subjects showed significant aging across the entire cortex, particularly in medial temporal regions and associated cortical networks.
Significant LBA elevation was reported across the whole cortex.
The strongest effects were in medial temporal regions and associated cortical networks.
Results
Feature ablation showed that curvature and GWR were preferentially sensitive to AD pathology.
Ablation analysis assessed the contribution of each morphometric feature.
Curvature and gray/white matter intensity ratio were highlighted over cortical thickness, surface area, and sulcal depth.
Results
Regional LBA gaps were significantly associated with neuropsychological measures of AD-related cognitive impairment.
This links local cortical aging patterns to clinical outcomes.
Specific measures and effect sizes are not given in the abstract.
What This Means
This study built a computer model, a type of artificial intelligence called a graph neural network, that estimates how old different parts of the brain's outer layer appear, based on standard MRI scans. Earlier methods mostly gave a single "brain age" for the whole brain. This model instead gives a detailed map of apparent aging across the cortex, using measures like thickness, surface area, folding, and the contrast between gray and white matter. It was trained on scans from over 14,000 cognitively healthy adults.
The model gave more consistent results across repeat scans than a leading existing method. In healthy people, the regions that appeared to age most were association areas of the cortex. People with mild cognitive impairment showed widespread apparent aging, especially in the parahippocampal gyrus. People with Alzheimer's disease showed apparent aging across the whole cortex, most strongly in medial temporal regions, which are known to be affected early in the disease. Measures of cortical folding (curvature) and gray/white matter contrast were especially sensitive to Alzheimer's-related changes.
This research suggests that mapping local brain aging from MRI could give a more detailed and interpretable picture of how aging and Alzheimer's disease affect different brain regions. The regional aging measures were also linked to scores on cognitive tests, suggesting they relate to real clinical symptoms. The abstract does not report results for individual patient diagnosis or care, so these findings are about research and understanding disease patterns.
Anderson S, Jomsky J, Chaudhari N, Chowdhury N, Zheng X, Irimia A. (2026). Graph Neural Network Reveals the Cortical Morphology of Local Brain Aging in Normal Cognition and Alzheimer's Disease.. Neuroinformatics. https://doi.org/10.1007/s12021-026-09808-y