Normative Brain Atrophy Modeling Incorporating Intracranial Volume Improves Pre-clinical Detection of Dementia Risk
Jonadab Dos Santos Silva1, Francesco La Rosa2, Emma Dereskewicz2, Julia Galasso2, Robin Graney2, Nadia Garcia2, Sarah Levy2, James Sumowski2, Erin Beck2
1Yale School of Medicine, 2Icahn School of Medicine at Mount Sinai
Objective:
To develop and validate a lifespan model of brain volume that integrates estimated total intracranial volume (eTIV) to improve sensitivity for early, preclinical brain atrophy associated with cognitive decline and dementia risk.
Background:
Age-related brain atrophy reflects both developmental and degenerative influences, with considerable inter-individual variability driven by head size and sex. Because eTIV reflects an individual’s maximum brain size, modeling it alongside age may enhance detection of deviations from expected atrophy trajectories. We hypothesized that accounting for eTIV would yield more precise normative metrics and improve prediction of cognitive decline and dementia conversion.
Design/Methods:
MRI data from 80 public datasets (n=21,977; median age 55 years [range 4–100]; 58% female) were modeled using P-spline GAMLSS, stratified by sex and adjusted for age and eTIV. Normative Quantile Residuals (QR; z-score equivalent) were derived for brain volume. In the ADNI cohort (n=2,407; 52.3% female; 7,759 MRI sessions; mean 3.2 per subject), Cox regression and bootstrapped logistic models assessed whether QR predicted conversion from cognitively normal (CN) to MCI or AD.
Results:
Head size moderated age-related atrophy rate (χ² > 38, p<10-7), faster in larger-eTIV individuals. Among 535 ADNI participants, 365 converted to MCI/AD (267 MCI, 98 AD) and 170 remained CN. Median time-to-conversion was 0.51 years (IQR 0.47–0.96); non-converters had 4.05 years follow-up. Each 1-SD decrease in QR corresponded to a 10% higher conversion hazard (HR=0.90, 95% CI 0.87–0.93, p<10-8). QR model outperformed conventional normalization (C-index 0.688 vs. 0.603; ΔC=0.084, 95% CI 0.068–0.099, p<10-5) and achieved AUC=0.731 (95% CI 0.706–0.754), sensitivity=0.68, specificity=0.79. QR most strongly correlated with language (r=0.42, P<0.001) and visuospatial abilities (r=0.30, P=0.016).
Conclusions:
Integrating intracranial volume into normative modeling refines individualized brain aging estimates, enhances cognitive and prognostic sensitivity, and identifies subtle, pre-clinical deviations predictive of dementia within two to three years, supporting its utility for early pathological detection and clinical trial applications.
Disclaimer: Abstracts were not reviewed by Neurology® and do not reflect the views of Neurology® editors or staff.