Summary: Researchers trained a deep neural network on magnetic resonance imaging (MRI) scans from nearly 15,000 cognitively healthy individuals aged 19 to 100. Moving beyond traditional single-number brain age metrics, the model generates high-resolution 3D maps displaying local brain age acceleration. Applied to participants with mild cognitive impairment and Alzheimer’s disease, the AI identified localized premature aging concentrated in the hippocampus, amygdala, and frontal-temporal regions, establishing a strong correlation between localized structural degeneration and cognitive test performance.
Key Facts
- Voxel-Level Spatial Resolution: Replaces single-number global “brain age” estimates with high-resolution 3D maps calculating regional aging at the level of individual voxels across the entire brain volume.
- Baseline Asymmetry and Regional Dynamics: In healthy populations, the frontal and temporal lobes consistently appear biologically older than occipital and parietal regions, while the right hemisphere exhibits slightly more advanced structural aging than the left regardless of hand dominance.
- Targeted Neurodegenerative Acceleration: Individuals with mild cognitive impairment and Alzheimer’s disease showed pronounced regional age acceleration concentrated in the hippocampus, amygdala, and deep memory pathways long before global changes manifest.
- Cognitive Assessment Correlation: Accelerated local brain age directly mirrored lower scores on standardized cognitive assessments, with the tightest structure-function coupling occurring in advanced Alzheimer’s disease cases.
- Prognostic Precision Care Potential: Provides a computational framework to track regional drug efficacy in clinical trials and identify early-stage dementia risk prior to overt clinical symptoms.
Source: USC
USC researchers have developed an approach that uses artificial intelligence to generate detailed maps that highlight differences in how distinct parts of the brain age.
The new model also sheds light on how patterns of brain changes correlate with changes in cognitive function across the lifespan, according to a new USC study published in the journal Proceedings of the National Academy of Sciences.
The researchers, led by Associate Professor Andrei Irimia of the USC Leonard Davis School of Gerontology, used magnetic resonance imaging from nearly 15,000 cognitively healthy individuals to train a deep learning AI model.
The data provided a baseline against which the model could measure local brain age, or how old specific regions of the brain appear. When the AI model was then used to analyze MRI images from people with mild cognitive impairment and Alzheimer’s disease, it revealed distinct patterns of accelerated aging in brain regions known to be affected early in neurodegeneration.
While most studies of brain age measure this phenomenon using a single number, the new model provides a much richer picture of typical aging and neurodegeneration. Rather than assigning a single “brain age” to an individual, the approach generates a detailed map showing how old different parts of the brain appear relative to what is typical for someone of the same chronological age.
“Not all brain regions age at the same rate,” Irimia said. “Some areas appear to be more resilient, while others are more vulnerable to aging and disease. By measuring local brain aging, we can identify where the brain is aging faster than expected and how those changes relate to cognitive function.”
Brain age as a biomarker
The research builds on previous efforts to estimate “brain age,” an emerging neuroimaging biomarker that compares a person’s brain structure to patterns seen in healthy people across the lifespan. Traditional methods typically reduce the brain to a single age estimate, which can obscure important regional differences.
The new approach instead measures local brain age at the voxel level — the three-dimensional units that make up an MRI scan — producing a much more detailed picture of structural aging throughout the brain.
“This more nuanced understanding of how the brain ages could pave the way for earlier identification of dementia, a better understanding of what factors affect risk and new ideas for treatment approaches,” Irimia said.
To develop the model, the researchers trained a deep-learning neural network using MRI scans from 14,748 cognitively normal adults ages 19 to 100 drawn from six large public datasets, including the UK Biobank, the Human Connectome Project and the Alzheimer’s Disease Neuroimaging Initiative.
They then tested the model using MRI scans from more than 1,900 additional participants in the Alzheimer’s Disease Neuroimaging Initiative, including cognitively normal adults, people with mild cognitive impairment and people with Alzheimer’s disease.
Across healthy adults, the model consistently found that the frontal and temporal lobes — regions involved in decision-making, memory and other higher cognitive functions — appeared biologically older than the parietal and occipital regions, which are involved in spatial awareness and sensory processing functions. The researchers also found that the brain’s right hemisphere tended to show slightly more advanced aging than the left, a pattern that persisted regardless of whether participants were right- or left-handed.
As cognitive impairment progressed, the differences became even more pronounced. Compared with cognitively normal adults, participants with mild cognitive impairment or Alzheimer’s disease showed significantly older local brain ages in structures that are among the first affected by Alzheimer’s pathology, including the hippocampus, amygdala and several deep brain regions involved in memory and cognitive processing.
The researchers also found that older local brain age was associated with poorer performance on cognitive assessments, strengthening the link between structural brain changes and real-world function. The strongest relationships appeared in people with Alzheimer’s disease, suggesting that regional brain aging may become increasingly informative as neurodegeneration advances.
What’s ahead
Because the model produces anatomically detailed maps, it could eventually help scientists better understand why some people experience faster decline in specific cognitive abilities than others. The approach may also prove useful for tracking disease progression or evaluating whether experimental therapies are slowing degeneration in targeted brain regions.
Although the findings are promising, Irimia emphasized that the method remains a research tool. The model was trained primarily on research-quality MRI data and will require additional validation using more diverse clinical datasets before it can be adopted in routine patient care.
The study also relied largely on cross-sectional data, meaning that future longitudinal studies will be needed to determine whether local brain aging can reliably predict who will progress from healthy aging to mild cognitive impairment or Alzheimer’s disease.
Still, the researchers believe that moving beyond a single measure of brain age represents an important advance for neuroscience.
“Brain aging isn’t uniform,” Irimia said. “By understanding how individual regions age, as well as how those patterns differ from person to person, we’re moving toward a much more precise understanding of healthy aging and neurodegenerative disease. Ultimately, that could help us identify people at risk earlier and develop more personalized approaches to preserving brain health.”
About the study
Irimia’s co-authors include first author Nikhil N. Chaudhari, Owen M. Vega Huerta, Samayan Bhattacharya and Nahian F. Chowdhury, all of USC.
Funding: The study received support from the National Institutes of Health (R01 AG 079957 to Irimia), the Hanson-Thorell Family Research Scholarship Fund, the Center for Undergraduate Research in Viterbi Engineering (CURVE) at USC and from anonymous donors.
Key Questions Answered:
A: Traditional neuroimaging algorithms collapse an entire brain scan into a single overall age figure, hiding regional differences. The USC deep learning model evaluates brain structure at the voxel level, 3D pixel-like units of an MRI scan, generating an anatomically detailed heatmap that shows exactly which regions are aging faster or slower relative to chronological norms.
A: Even in healthy adults, brain aging is not uniform. The AI revealed that frontal and temporal lobes naturally appear biologically older than sensory-focused occipital and parietal lobes. Additionally, the right cerebral hemisphere consistently displayed slightly more advanced structural aging than the left, independent of whether an individual was right-handed or left-handed.
A: Not yet. Senior author Andrei Irimia emphasized that the model remains a research tool. It was trained on research-grade MRI datasets and requires further validation across diverse clinical settings and longitudinal patient studies to prove it can reliably predict individual progression from healthy aging to dementia.
Editorial Notes:
- This article was edited by a Neuroscience News editor.
- Journal paper reviewed in full.
- Additional context added by our staff.
About this AI and brain aging research news
Author: Elizabeth Newcomb
Source: USC
Contact: Elizabeth Newcomb – USC
Image: The image is credited to Neuroscience News
Original Research: Open access.
“Deep learning maps local brain aging in relation to cognition across human adulthood” by Nikhil N. Chaudhari, Owen M. Vega Huerta, Samayan Bhattacharya, Nahian F. Chowdhury, Andrei Irimia, Alzheimer’s Disease Neuroimaging Initiative. PNAS
DOI:10.1073/pnas.2532233123
Abstract
Deep learning maps local brain aging in relation to cognition across human adulthood
Brain aging, the strongest risk factor for Alzheimer’s disease (AD), varies across cortical regions. Global brain age (GBA), an imaging-derived measure of neuroanatomic decline, reduces structural aging to a single summary value. This can potentially obscure regional patterns of cognitive vulnerability preceding AD.
This study introduces a deep-learning architecture trained on the T1-weighted MRIs of 14,748 cognitively normal (CN) participants from multiple sites to estimate local brain age (LBA) at voxel level. By mapping spatial variations in brain aging, the model reveals relatively advanced aging in frontal and temporal lobes compared to parietal and occipital regions.
Beyond aging in CN aging adults (N = 1.102), findings reveal a pattern of progressively advanced frontotemporal aging as a function of neurodegeneration stage, ranging from mild cognitive impairment (MCI, N = 354) to AD (N = 529).
Compared to CN adults, key cortical and subcortical structures known to manifest early AD pathology exhibit significantly older LBAs in both early MCI and AD (P <0.05). Deviations from normative regional aging are significantly associated with cognitive performance supported by neural processes linked to those regions (P <0.05), thereby relating anatomic aging to functional outcomes.
By quantifying regional variations in brain aging, this framework extends GBA models to provide anatomically interpretable measures that can improve characterization of typical and pathological aging.
