Computational framework to assess brain maturation in small vulnerable newborns, from womb to cot

Over 25% of infants worldwide are born too soon and/or too small due to adverse exposures during pregnancy, such as maternal malnutrition and infections. Such exposures can also disrupt the maturation of the fetal brain, which is rapidly developing during pregnancy and the first year of life. Hence, these small vulnerable newborns (SVNs), if they survive, are at high risk of poor neurodevelopmental outcomes in later life.Monitoring anatomical growth is key to optimising and determining SVN care pathways, especially in the first days after birth. Ultrasound (US) enables non-invasive imaging of fetal and neonatal brain anatomy, but growth assessments are based on coarse metrics (e.g., head circumference), with pre- and post-birth measurements rarely integrated. WOMB2COT will tackle these hurdles with a technically ambitious programme that combines US imaging with deep learning (DL) and computational neuroanatomy. We will develop advanced DL-based US image analysis techniques, including neural radiance fields, image registration, and domain adaptation, to quantify brain development in the most vulnerable preterm infants compared to their expected in utero growth trajectory. This approach will utilise longitudinal cUS scans collected in neonatal intensive care units (NICUs), offering a more detailed, objective evaluation of neonatal brain growth and bridging the care gap as these infants transition from intra- to extra-uterine life. We aim to reduce US imaging artefacts (e.g., acoustic shadows) and enhance structural specificity to sub-millimetre resolution to extract brain morphometrics (BMs), surpassing coarse and qualitative assessments. Building on our population atlas of healthy brain maturation (Nature 2023) and advanced computational tools, we will compare SVN growth with age-matched healthy fetuses. To maximise its diagnostic utility, WOMB2COT will be developed for low-cost, point-of-care US equipment, benefiting regions with the highest SVN burden.