Mapping the brain-web morphology to characterise ageing and cognitive impairment

Abstract

Ageing reshapes human brain anatomy across spatial scales, yet its geometrical organisation remains difficult to quantify. Here we introduce SHARP (Spectral Hierarchy Analysis, Reconstruction and Prediction), which transfers hierarchical cosmic-web classification to structural MRI, identifying knots, filaments, sheets and voids at four spectral levels and combining their morphology with higher-order spatial statistics and anatomical measurements. In CamCAN, selected filament- and sheet-associated features show opposing correlations with age and distinct associations with subcortical and ventricular volumes. To quantify their contribution, we compare SHARP with a neural-network baseline combining 36 regional anatomical measurements, total brain volume and sex. Adding cosmologically motivated descriptors reduces the mean absolute error of MRI-only age prediction from 6.04 to 4.71 years (22%) in CamCAN and from 4.57 to 4.07 years (11%) in OASIS-3 under participant-independent validation. Brain-web morphology with only brain volume and sex already achieves 5.12 years in CamCAN. For cognitive-status classification in OASIS-3, SHARP identifies 85% of Clinical Dementia Rating (CDR)-positive observations at 86% specificity, compared with 78% sensitivity for the anatomical baseline at matched empirical specificity. These results demonstrate that cosmological methods provide complementary geometrical information and measurable gains in characterising brain ageing and cognitive impairment.