Brain Age from Neuroimaging

Key Takeaways

Brain-age models learn age-related patterns from brain scans and estimate the chronological age most compatible with a new scan. [1] [2]
The brain age gap is usually predicted age minus chronological age; a positive gap describes an older-appearing brain relative to the model's reference data. [2]
Brain age gaps are associated with neurological disorders, physical and cognitive measures, and mortality in cohorts, but these associations do not make the measure a diagnosis or a direct reading of ageing rate. [3] [4] [7]
Scanner differences, training samples, image processing, prediction error, and age-related statistical bias can materially change the result. [5] [6] [8]

Brain age is a model-derived neuroimaging biomarker. Researchers train an algorithm to predict age from scans in a reference sample, apply it to scans not used for training, and compare each prediction with the person's chronological age. The resulting difference is often called the brain age gap, predicted age difference, or BrainAGE score. [1] [2]

Who This Is Useful For

This page is useful for readers evaluating claims that an MRI scan reveals how old a brain “really” is. It explains why brain age can summarize age-related imaging patterns in research while remaining dependent on the model, reference population, imaging protocol, and statistical analysis. [5] [6]

How Brain Age Is Estimated

Most established models use T1-weighted structural MRI. Preprocessing can align scans to a common space and derive features such as regional grey-matter volume, cortical thickness, surface area, or voxel-level tissue patterns. A regression model then learns how those features vary with chronological age in its training data. [1] [3]

Other approaches use diffusion MRI, functional MRI, or positron-emission tomography, and some combine modalities. These inputs capture different aspects of brain structure, connectivity, microstructure, or metabolism, so estimates from different models are not necessarily interchangeable. [2] [5]

From Scan to Brain Age Gap

Stage	What Happens	Interpretive Constraint
Reference sampling	Scans and ages are assembled from a population treated as the normative reference [1]	The learned norm reflects that sample's age range, health criteria, and demographics [4]
Image processing	Images are aligned, segmented, or supplied to a model as voxel data [1]	Scanner and preprocessing choices can alter prediction error and transfer to new sites [8]
Age prediction	A regression model estimates age from imaging features [2]	Accuracy in held-out data does not by itself establish biological or clinical validity [5]
Gap calculation	Chronological age is subtracted from predicted age [2]	The raw gap commonly retains age-related bias and may require carefully designed correction [6]

What an Older-Appearing Brain Means

A positive brain age gap means that a scan resembles scans from older people in the model's reference data. It does not identify one mechanism. The signal can reflect distributed differences in tissue volume and morphology, disease-related change, early-life influences, vascular or other exposures, and technical features of image acquisition and processing. [3] [4] [7] [8]

Large studies have reported older-appearing brains across several disorders, including mild cognitive impairment, dementia, multiple sclerosis, schizophrenia, bipolar disorder, and major depression. Effects vary across disorders and brain regions, and may reflect disease, treatment, comorbidity, or other correlated factors rather than a shared ageing mechanism. [4]

Associations with Health Outcomes

In the Lothian Birth Cohort 1936, an older brain-predicted age was associated with weaker grip strength, poorer lung function, slower walking speed, lower fluid intelligence, higher allostatic load, and mortality risk. This supports cohort-level prognostic relevance, although the observational design cannot establish that the brain age gap causes these outcomes. [3]

Studies of neurological and psychiatric conditions likewise show group differences in apparent brain age. Their distributions overlap, however, so a gap is not specific to one disorder and does not replace clinical history, cognitive assessment, or diagnostic imaging interpretation. [4] [5]

Cross-Sectional Age Is Not Ageing Rate

A single scan compares one person's anatomy with a cross-sectional age pattern. It does not directly observe how that person's brain changes over time. In analyses of UK Biobank and Lifebrain data, cross-sectional brain age differences were not associated with longitudinal rates of brain change and were related to early-life and genetic factors. An older-looking scan therefore should not automatically be described as evidence that the brain is currently ageing faster. [7]

Statistical Bias and Generalisability

Brain-age regression commonly overestimates younger ages and underestimates older ages. This dependence between age and prediction error can produce misleading group comparisons when age distributions differ. Correction methods exist, but the correction must be estimated without leaking information from the test data and its implications for downstream statistics must be reported clearly. [6] [9]

Performance can also fall when a model moves to another scanner, site, population, or preprocessing pipeline. A recent multi-pipeline experiment found that preprocessing choices changed prediction error and that offset correction was important when transferring models to unseen datasets. [8]

Evidence Quality and Interpretation

Confidence is strong that machine-learning models can predict chronological age from structural MRI in held-out research data and that the brain age gap can capture reproducible group-level associations. [1] [2] [3]

Confidence is moderate that brain age provides information relevant to brain health beyond chronological age. Associations have been observed across large samples and multiple outcomes, but effect estimates depend on model construction, correction procedures, and the populations studied. [3] [4] [6]

Confidence is weaker for interpreting one estimate as an individual's biological age or ongoing rate of brain ageing. Longitudinal evidence challenges that interpretation, and there is no universal brain-age model or clinically established threshold shared across methods. [5] [7]

What This Does Not Mean

It does not mean the model has measured a literal or uniquely defined biological age of the brain. [5]
It does not mean a positive gap proves active accelerated ageing within that person. [7]
It does not mean a brain age gap identifies a particular disease or mechanism. [4]
It does not mean estimates from different scanners, modalities, models, and correction methods can be compared directly. [6] [8]

Practical Interpretation Examples

If predicted brain age is five years above chronological age: the scan appears older by that model's definition; the number includes prediction uncertainty and is not a five-year forecast. [2] [5]
If a patient group has a higher average gap: this is evidence of a group-level imaging difference after appropriate adjustment, not a diagnostic rule for each participant. [4] [6]
If repeated scans yield different estimates: biological change is one possibility, but scanner, acquisition, preprocessing, and model variability must also be considered. [7] [8]

Summary

Neuroimaging brain age compresses a complex scan into an age-like estimate by comparing it with patterns learned from a reference dataset. It is useful for studying population differences and associations with health outcomes, but the gap is shaped by the imaging modality, sample, model, bias correction, and technical pipeline. Its clearest interpretation is an older- or younger-appearing imaging pattern within a specified model—not a diagnosis, a universal biological age, or a direct measurement of within-person ageing speed. [5] [6] [7]

References

Franke, K., et al. (2010). Estimating the age of healthy subjects from T1-weighted MRI scans using kernel methods: exploring the influence of various parameters. NeuroImage. https://doi.org/10.1016/j.neuroimage.2010.01.005
Cole, J. H., et al. (2017). Predicting brain age with deep learning from raw imaging data results in a reliable and heritable biomarker. NeuroImage. https://doi.org/10.1016/j.neuroimage.2017.07.059
Cole, J. H., et al. (2018). Brain age predicts mortality. Molecular Psychiatry. https://www.nature.com/articles/mp201762
Kaufmann, T., et al. (2019). Common brain disorders are associated with heritable patterns of apparent aging of the brain. Nature Neuroscience. https://www.nature.com/articles/s41593-019-0471-7
Cole, J. H., & Franke, K. (2017). Predicting age using neuroimaging: innovative brain ageing biomarkers. Trends in Neurosciences. https://doi.org/10.1016/j.tins.2017.10.001
Smith, S. M., et al. (2019). Estimation of brain age delta from brain imaging. NeuroImage. https://doi.org/10.1016/j.neuroimage.2019.06.017
Vidal-Piñeiro, D., et al. (2021). Individual variations in ‘brain age’ relate to early-life factors more than to longitudinal brain change. eLife. https://elifesciences.org/articles/69995
Dular, L., et al. (2024). Extensive T1-weighted MRI preprocessing improves generalizability of deep brain age prediction models. Computers in Biology and Medicine. https://doi.org/10.1016/j.compbiomed.2024.108320
Butler, E. R., et al. (2021). Pitfalls in brain age analyses. Human Brain Mapping. https://doi.org/10.1002/hbm.25733

Educational Disclaimer

This content is provided for educational purposes only and does not constitute medical advice.