Decoding brain age predictions from sleep electroencephalography across infancy to adolescence

Childhood sleep electroencephalography (EEG) reveals brain maturation patterns aligned with age, offering a window into development at the bedside. Leveraging this, we used supervised neural networks to predict age from overnight EEG and derive a Functional Brain Age (FBA) across wake, NREM (N1-N3), and REM sleep in 814 children with clinically normal sleep studies. We evaluated how FBA varies across sleep architecture and assessed the accuracy of neural networks, EEG channels, sleep segments, data quality, quantitative EEG features, and explainability methods on prediction performance. Prediction accuracy varied developmentally, with a mean absolute error (MAE) of 0.78 years in infancy (0-2 years), 0.87 years in childhood (2-12 years), and 1.55 years in adolescence (12-18 years), yielding an overall MAE of 0.96 years (95CI 0.90-1.01). FBA fell within ± 25% of chronological age in over 95% of children, with highest accuracy (< 1 year MAE) during N2, N3, and REM stages that reflect well-defined developmental EEG changes such as delta power and spindles. FBA reliability was shaped by signal quality and stable, age-specific patterns across sleep stages. Explainability analyses showed that network activations aligned with quantitative EEG features, supporting the biological validity of FBA. These findings support scalable, non-invasive tools that use sleep EEG to track brain maturation as an objective marker of neurodevelopmental health.