Abstract

During brain aging, atrophy of cortical surface area (SA) exhibits region-specific patterns that extend beyond classical anatomical divisions. However, a specialized brain atlas of healthy brain aging has not yet been developed. We propose a two-stage spatially regularized non-negative matrix factorization (NMF) strategy to construct a brain aging atlas. In the first stage, we extract initial morphological covariance features from 1,066 cortical SA scans of 436 (196 males and 240 females) cognitively normal adults (aged 51–96) from the Alzheimer’s Disease Neuroimaging Initiative cohort using standard NMF to ensure sparsity and biological interpretability. In the second stage, we perform refined smoothing on the manifold structure to rectify boundary noise. This strategy enhances the spatial continuity of the atlas while preserving data-driven patterns. Using this atlas, we identified three distinct cortical atrophy patterns: frontal–temporal–occipital dominant (FTOD), parietal slow-changing (PSC), and diffuse patterns. Our findings revealed that: (1) Cortical aging exhibits pronounced spatiotemporal heterogeneity, with the frontal, temporal, and occipital cortices showing sensitivity to age-related changes. Significant interhemispheric differences were observed in atrophy rates, with no sex-related effects. (2) The right hemisphere exhibited faster atrophy than the left, and all atrophy patterns showed limited interhemispheric spatial overlap (Dice = 0.653), indicating pronounced hemispheric asymmetry in cortical SA aging. (3) Different atrophy patterns demonstrate pattern-specific associations with cognitive performance: higher relative preservation of left-hemispheric FTOD pattern is associated with less subjective cognitive decline (SCD), whereas greater relative preservation of diffuse and PSC patterns correlates with higher SCD.