Conditional brain charts link cortico-subcortical structural dissociation to Alzheimer's disease progression

Conditioning lifespan normative models of cortical thickness on subcortical anatomy enables site calibration with 20 scans and reveals structural dissociations predictive of conversion to dementia.

Primary source: arXiv:2610.21660: https://arxiv.org/abs/2610.21660

What’s new: Normative neuroimaging models chart how brain anatomy varies across the human lifespan, mapping an individual patient as statistical deviations (z-scores) relative to a reference cohort. Standard regional models evaluate cortical thickness independently across regions, ignoring coordinated dependencies between cortical thickness and underlying subcortical structures. Furthermore, transferring these models across scanner hardware and acquisition protocols usually requires extensive retraining or coarse location-scale recalibration. In Transferable conditional brain charts link cortico-subcortical structure to Alzheimer’s disease progression (arXiv:2610.21660, 15 Oct 2026), a multi-institution consortium led by Nitin Sharma, Yanwu Yang, Richard Dinga, and Thomas Wolfers introduces conditional lifespan brain charts that model regional cortical thickness conditioned directly on subcortical volumes.

Using an aggregation of 21,351 healthy individuals across 216 imaging sites, the authors conditioned generalized additive models for location, scale, and shape (implemented in their open-source NormGAMLSS toolkit) on ten subcortical volumes. Conditioning substantially improved out-of-sample predictions, with subcortical volumes in the hippocampus and amygdala explaining the greatest variance in temporal and limbic cortical thickness. For multi-site deployment, refitting only the intercept terms across all four distributional parameters using as few as 20 healthy reference scans successfully calibrated models to unseen acquisition sites without architectural retraining. When tested on longitudinal patient cohorts from the Alzheimer’s Disease Neuroimaging Initiative (ADNI) and AIBL, both global deviation and structural dissociation (the reduction in mean absolute cortical deviation once conditioned on subcortical volumes) significantly predicted conversion from mild cognitive impairment (MCI) to Alzheimer’s dementia beyond conventional Clinical Dementia Rating (CDR) scores and baseline cognitive metrics.

Why it matters: In neurodegenerative disease research, regional atrophy measures frequently fail to distinguish generalized physiological aging from localized pathology. Conditioning cortical measures on subcortical volume contextualizes whether a patient’s cortical thinning is expected given their subcortical geometry, or whether an anatomically decoupled dissociation is underway. The ability to calibrate these normative models on small target-site calibration batches (around 20 scans) removes a key barrier to deploying normative clinical scoring tools across heterogeneous hospital imaging networks.

Caveats: While the method supports site recalibration with modest sample sizes, performance remains sensitive to severe site-level segmentation artifacts or outliers in the calibration cohort. The current formulation conditions cortical thickness on volumetric subcortical summaries rather than dense voxelwise geometry or white matter tractography. Additionally, while structural dissociation demonstrated strong predictive power for conversion to dementia over multi-year follow-ups, clinical utility in individual acute treatment selection warrants prospective diagnostic trial validation.