Brain age predicted using graph convolutional neural network explains neurodevelopmental trajectory in preterm...

Area of Science:

• Neuroscience
• Medical Imaging
• Artificial Intelligence

Background:

• The third trimester of gestation involves significant brain development.
• Preterm neonates face risks of postnatal abnormalities and altered neurodevelopmental trajectories.

Purpose of the Study:

• To investigate if predicted brain age (PBA), derived from a graph convolutional network (GCN) analyzing third-trimester cortical morphometrics, correlates with postnatal issues and neurodevelopmental outcomes.
• To assess the utility of PBA in understanding the neurodevelopmental path of preterm infants.

Main Methods:

• Analyzed 577 T1 MRI scans from preterm neonates across two datasets.
• Utilized the NEOCIVET pipeline for cortical surface and morphology extraction.
• Employed a GCN to predict brain age, calculating the brain age index (BAI = PBA - chronological age).
• Examined relationships between BAI and factors like preterm birth, perinatal brain injuries, postnatal events, and 30-month neurodevelopmental scores using structural equation models (SEM).

Main Results:

• GCN-based PBA prediction for preterm neonates without brain lesions achieved a mean absolute error of 0.96 weeks, surpassing conventional machine learning methods.
• SEM revealed that BAI mediated the impact of preterm birth and postnatal clinical factors on neurodevelopmental outcomes at 30 months.
• BAI did not mediate the effects of perinatal brain injuries on neurodevelopmental outcomes.

Conclusions:

• Brain morphology, as measured by GCN-based PBA, is clinically relevant, linking to postnatal factors and predicting neurodevelopmental outcomes.
• The developed brain age index offers enhanced accuracy and clinical interpretability for assessing neonatal neurodevelopment.
• This approach may enable earlier detection of developmental abnormalities and guide interventions for preterm infants, improving long-term prognosis.