EEG-based deep learning estimates preterm infants' cerebral age: Contribution of slow-wave and cross-frequency coupling

Reliably estimating cerebral maturation in preterm newborns is crucial for early identification of atypical neurodevelopment. We developed a deep-learning framework using a convolutional neural network (CNN) to predict postmenstrual age (PMA) from spontaneous neural activity in EEG recordings of 165 healthy preterm infants (28-38 weeks PMA). The model achieved a mean absolute error of 0.75 ± 0.60 weeks, with 73.9% of prediction errors within one week. To identify key neurophysiological drivers, we systematically manipulated signal components. Results showed that removing low-frequency activity (<4 Hz) significantly degraded performance and disrupting phase-amplitude coupling selectively impaired prediction accuracy before 34 weeks PMA, while high-frequency activity (>12 Hz) had minimal impact. Our results demonstrate that maturation signatures are primarily driven by slow-wave dynamics and their hierarchical coupling with faster oscillations, which are fundamental hallmarks of early neurodevelopment.