What local brain age reveals about uneven aging

The measure called local brain age is not the birth date of a particular brain region. It is an estimate of how old that region’s structure appears on an MRI scan. A new deep-learning method suggests that the brain does not age as a single unit: some areas retain more youthful structural features, while others acquire age-associated changes earlier.

Most brain-age models compress an entire scan into one number. A 65-year-old person, for example, might receive an estimated brain age of 70. That summary can be useful, but it hides regional differences: the hippocampus, temporal lobe and occipital cortex in the same person may follow very different structural trajectories.

The new approach replaces the single estimate with a three-dimensional map. A neural network analyzes T1-weighted MRI scans and predicts age at the level of individual voxels, the tiny volume elements that make up a scan. Researchers can then compare each local estimate with the person’s chronological age and identify areas that look older or younger than expected.

A map trained on thousands of MRI scans

The model was trained on MRI scans from 14,748 cognitively normal adults aged 19 to 100, drawn from six large public datasets. From these scans, it learned which structural patterns are typically associated with different stages of adulthood.

The researchers then applied the model to a separate Alzheimer’s Disease Neuroimaging Initiative sample. The analysis included 1,102 cognitively normal participants, 354 people with mild cognitive impairment and 529 people with Alzheimer’s disease.

Even among cognitively healthy adults, the resulting maps were not uniform. Frontal and temporal regions generally appeared older than parietal and occipital regions. The right hemisphere also showed slightly more advanced structural aging than the left, a pattern that remained regardless of whether participants were right- or left-handed.

This does not mean that one brain region has literally lived more years than another. The algorithm recognizes combinations of shape, volume and other structural features that were more common at particular ages in the training data. Local age is therefore a computational measure of structural resemblance, not a direct measurement of neuronal age.

What changes in Alzheimer’s disease

In people with mild cognitive impairment and Alzheimer’s disease, areas with advanced estimated age were more extensive and pronounced. The strongest deviations appeared in frontal and temporal regions, as well as the hippocampus, amygdala and other structures involved in memory and information processing.

That pattern is consistent with regions known to be affected early in neurodegeneration. However, the model does not directly measure amyloid, tau or neuronal death. It detects an MRI-based structural pattern that is statistically associated with older age and cognitive impairment.

Regional age deviations were also associated with cognitive test performance. On average, areas that appeared older relative to the norm were linked to poorer performance in functions supported by those regions. The relationships were strongest among participants with Alzheimer’s disease. This remains an association, however: the study does not show that an older local age itself causes memory decline.

Why one number may not be enough

The main advantage of the method is anatomical specificity. Two people could have similar global brain-age estimates but very different maps. One might show the largest deviations in memory-related structures, while another might show them in regions involved in attention or executive function. In research settings, such maps could help connect structural change with particular cognitive difficulties more precisely.

The method might eventually be tested for monitoring disease progression or assessing whether an experimental treatment slows degeneration in selected regions. It is not yet a clinical diagnostic tool. The model was developed largely with research-quality MRI data and still needs validation across different hospitals, scanners, populations and medical conditions.

Most of the analysis was also cross-sectional: it compared different people at one point in time rather than following each participant for years. The study therefore cannot establish whether local brain age can predict who will progress from healthy aging to mild cognitive impairment or Alzheimer’s disease.

For now, the value of the approach lies not in providing a ready-made prognosis, but in changing the scale of the question. Instead of asking only how old the brain looks overall, researchers can ask which specific regions appear to be aging faster than expected.