Researchers have pooled diffusion MRI data from more than 54,000 people across 19 international datasets to produce the most comprehensive normative model of the brain’s white matter ever created. Published in Nature Communications on 27 May 2026, the work establishes lifespan centile curves for key measures of brain microstructure – offering clinicians and trialists a sensitive new tool for detecting disease-related changes at the individual level.
When a paediatrician plots a child’s height on a growth chart, they are asking a deceptively simple question: is this person developing as expected for their age and sex? A landmark study from the USC Mark and Mary Stevens Neuroimaging and Informatics Institute (Stevens INI) at the Keck School of Medicine of USC has now applied the same logic to the brain itself – specifically to the vast network of white matter fibres that wire its regions together. The result is a lifespan normative model of brain microstructure with direct implications for the diagnosis and monitoring of Alzheimer’s disease, schizophrenia, and a wide range of other neurological and psychiatric conditions.
The clinical significance is hard to overstate. Traditional case-control neuroimaging studies compare group averages, which can obscure the substantial variation that exists between individuals carrying the same diagnosis. This new framework sidesteps that limitation. As the authors write in their paper, normative modelling “yields individual profiles of anomalies, which are agnostic to the clinical labels and accommodate the well-supported premise that not all individuals with the same disease deviate in the same brain regions and in the same direction.” In practice, that means clinicians and researchers can now ask not merely whether a patient’s group differs from controls, but precisely where – and by how much – that individual’s brain wiring departs from what would be expected for someone of their age and sex.
Building the atlas
White matter is the brain’s cabling system – dense bundles of myelinated axons that carry signals between cortical and subcortical regions. It is exquisitely sensitive to ageing and disease, yet until now there has been no large-scale normative reference for its microstructure comparable to those already available for brain volume and cortical thickness.
The Stevens INI team addressed that gap by aggregating diffusion MRI (dMRI) data from 54,583 individuals aged 4 to 91 years, drawn from 19 publicly available international datasets spanning cohorts from the United States, the United Kingdom, the Netherlands, Australia, China, Cuba and beyond. Using diffusion tensor imaging (DTI), they extracted four widely used metrics – fractional anisotropy (FA), mean diffusivity (MD), axial diffusivity (AD), and radial diffusivity (RD) – across 21 major deep white matter regions plus a global white matter skeleton measure, yielding 22 regions of interest in total.
The statistical backbone of the model is hierarchical Bayesian regression (HBR), a method specifically designed to handle the thorny problem of multi-site neuroimaging data, where scanner type, field strength, voxel size, and diffusion-weighting protocols all introduce systematic differences between cohorts. By treating acquisition protocol as a batch effect within the HBR framework, the team could pool data across 37 distinct dMRI protocols without inadvertently stripping out genuine biological variance – a persistent pitfall of simpler harmonisation approaches such as ComBat.
“Just as paediatric growth charts help clinicians determine whether a child’s height or weight is developing as expected, these brain charts provide a reference for how the brain’s neural pathways typically change over the lifespan,” said Julio E. Vil-lalón-Reina, MD, PhD, a postdoctoral researcher at the Stevens INI and the study’s first author. “That gives us a powerful new way to identify when an individual’s brain wiring falls outside the expected range.”
Trajectories across the lifespan
The resulting centile curves – running from the 2nd to the 98th percentile – paint a detailed picture of how white matter evolves from early childhood through to the ninth decade of life. FA, a marker of fibre organisation and myelination, follows an inverted U-shaped trajectory, rising during maturation and declining after a peak. The three diffusivity measures follow the reverse: a U-shaped pattern, falling as the brain matures and rising again in older age as microstructural integrity wanes.
Critically, the timing differs substantially between metrics and between regions. On average, FA peaked earliest (around age 29 for the global white matter skeleton), followed by RD (minimum at 37 years), MD (minimum at 43 years), and AD (minimum at 53 years). Among individual tracts, FA peaked as early as 16 years in the genu of the corpus callosum and as late as 39 years in the fornix. “Brain development and brain ageing are not uniform processes,” Villalón-Reina noted. “The brain’s neural pathways mature on distinct timelines, and some are more vulnerable to decline than others. Our model reveals this structure by merging data on a truly global scale.”
The ‘last in, first out’ principle
One of the more striking ancillary findings concerns an old theoretical question in neuroscience: does the brain age in the reverse order of how it developed? This retrogenesis hypothesis – sometimes called ‘last in, first out’ – predicts that white matter tracts that mature late in development should be among the first to degenerate in old age, because they are less robustly myelinated and therefore more vulnerable.
Using the centile curves to measure percentage change in DTI metrics at different life stages, the researchers found significant support for the hypothesis for FA, MD, and AD – but not for RD. Tracts with a later age of peak maturation showed faster percentage change in the oldest age bands (75-91 years), consistent with a developmental origin for individual vulnerability to age-related white matter decline. The team did not find support for a related but distinct idea, the ‘gain-predicts-loss’ hypothesis, which predicts that the rate of maturation and the rate of degeneration should be directly correlated.
Detecting disease in individuals
The model’s practical value was tested in three clinical datasets. In participants with mild cognitive impairment (MCI) and Alzheimer’s dementia from the ADNI3 and OASIS3 cohorts, the normative framework identified atypical white matter patterns concentrated in the cingulum of the hippocampus (CGH) and the splenium of the corpus callosum (SCC) – regions involved in memory and long-range cortical communication. For dementia, the strongest discriminative performance was seen for radial diffusivity in the CGH, with 37% of dementia patients showing extreme positive deviations in that tract. MCI showed more modest but still significant deviations – expected, given that microstructural changes are subtler in prodromal stages.
In a separate cohort of individuals with 22q11.2 deletion syndrome – a neurogenetic condition that markedly raises the risk for schizophrenia – the model identified widespread negative deviations in mean and axial diffusivity, particularly across the entire corona radiata. Crucially, a subset of individuals showed deviations in the opposite direction, a pattern that would have been invisible in a standard group-comparison analysis.
“This monumental study took seven years to complete,” said Paul M. Thompson, PhD, associate director of the Stevens INI and senior author of the study. “The vast scale of the data and the fine scale of the brain features assessed means we can now evaluate your neural pathways relative to other people of the same age, sex, and demographics.”
Implications for clinical trials
Beyond diagnostics, the charts open a new avenue for treatment evaluation. If a therapy targets white matter integrity – whether in Alzheimer’s disease, multiple sclerosis, or a psychiatric condition – researchers could track whether a patient’s DTI metrics shift towards age-expected centiles, or whether disease progression is slowed relative to the normative trajectory.
“This study demonstrates the power of large-scale, international data sharing to create tools the entire research community can use,” said Arthur W. Toga, PhD, director of the Stevens INI and Provost Professor at USC. “By establishing a lifespan framework for the brain’s communication pathways, this work opens new opportunities to detect subtle disease-related changes, compare conditions more rigorously, and move toward a more individualised understanding of brain health.”
The normative models are publicly available via the PCNPortal <https://pcnportal.dccn.nl >, and the analysis code is accessible on GitHub.
The team notes several limitations – the ENIGMA-DTI protocol is not optimised for children under five, some anatomically small or CSF-adjacent tracts show noisier model fits, and a fuller accounting of ethnoracial ancestry would strengthen generalisability – but the model’s built-in adaptation framework means it can be re-calibrated to any new clinical site with as few as 25-30 local control participants.
The charts will now be applied across more than 30 brain conditions, creating a common reference standard that spans neurodevelopmental, psychiatric, and neurodegenerative disease.
Journal reference:
Villalón-Reina, J. E., Zhu, A. H., Nabulsi, L., et al. (2026). Lifespan normative modeling of brain microstructure. Nature Communications, 17, 4693. https://doi.org/10.1038/s41467-026-72875-x
