Brain Age Estimation in Young Adults from Resting-State fMRI: Comparison of Functional Connectivity, Spectral Amplitude Features, and Regularized Regression Models

Brain age is an estimate of the brain's biological age, derived from neuroimaging data using machine learning algorithms, that allows quantification of brain ageing. The difference between predicted and chronological age, known as Brain Age Gap (BAG), can serve as a biomarker of deviations from typical ageing. Estimating brain age in young adults, where age-related differences are subtle, poses specific challenges. We developed a reproducible pipeline using the Human Connectome Project (HCP S1200), comprising 1,003 adults aged 22–37 years with resting-state functional magnetic resonance imaging (rs-fMRI). Time series from 50 brain networks were obtained using independent component analysis (ICA). Multiple BOLD signal representations were evaluated, including functional connectivity, ALFF, fALFF, and low-frequency subband variants. Ridge, ElasticNet, and Lasso regression were compared using grouped cross-validation while preserving family structure. Functional connectivity showed the best overall performance, while spectral representations incorporating multiple low-frequency subbands outperformed the original ALFF representation. Ridge consistently outperformed ElasticNet and Lasso, suggesting that age-related information is distributed across multiple functional features. These results indicate that low-frequency rs-fMRI information is relevant to brain ageing in young adults, with functional connectivity providing an informative representation for Brain Age estimation.