Neonatal brain-age models in full- and preterm infants

Prematurity affects brain development and increases risk for neurodevelopmental impairments. Yet reliable biomarkers for at-risk infants remain limited. We developed brain-age prediction models using diffusion magnetic resonance imaging-derived white matter features from two datasets: (1) the developing Human Connectome Project (dHCP; 368 healthy infants) and (2) a clinical sample collected at the Lucile Packard Children's Hospital (LPCH; 162 high-risk preterm infants). The goals of this study were (1a) to construct a white matter neonatal brain-age model including full-term and preterm neonates from a large research data set, (1b) to evaluate the accuracy of a separate brain-age model using the same architecture but trained and evaluated via cross-validation on a restricted subset of very preterm infants (<32 weeks gestational age) from the same data set to improve comparability with our clinical cohort; (2) to determine if a similar model can predict brain-age based on clinical MRI scans from high-risk neonates born preterm; and (3) to evaluate whether this preterm white matter-based brain-age model provides information about the infant's health beyond conventional clinical and demographic measures. White matter features demonstrated strong predictive performance in the dHCP dataset (within 3.9 days) and the LPCH clinical dataset (within 6.6 days). However, brain-age metrics (i.e., brain-age gap) showed no significant associations with health complications measured by a composite score of common prematurity complications. While tractometry-derived brain-age models accurately characterize brain maturation in the neonatal brain, their sensitivity to clinical complications in preterm infants appears limited. Global white matter maturation measures derived from clinical grade data may be insufficiently sensitive to capture the cumulative burden of prematurity-related morbidities, suggesting need for multimodal or longitudinal biomarkers.