AI and the World's Hard Problems · Medicine & drug discovery · arXiv · 2026-10-19
Covariance Eigenspace Provides Latent Attribution of Longitudinal Effects in Brain Age Gap
Brain age gap is a machine-learning biomarker derived from neuroimaging data. Using a covariance neural network (VNN) framework and Integrated Gradients attribution, the study examined longitudinal brain age gap effects in a mild cognitive impairment (MCI) cohort. The eigenspectrum of the anatomical covariance matrix explained the differences between amyloid-positive and amyloid-negative subcohorts more clearly than individual brain regions, with the leading eigenvector accounting for 83.2% versus at most 7.7% for any single region.