AI X-ray model successfully adapted for neonatal care

An international research team has successfully adapted an artificial intelligence X-ray model, originally designed for adults, to serve newborns. Because neonatal anatomy and clinical requirements differ significantly from adults, standard models often perform poorly on pediatric patients.

In a study involving 30 neonatal intensive care units across 20 countries and regions, researchers utilized a method known as ‘continual learning’. This approach allowed the model to be trained sequentially across different locations using local datasets without the need to centralize sensitive patient data. Each participating clinic provided approximately 100 annotated X-ray images to facilitate this process.

The primary application tested was the monitoring of endotracheal tube placement in chest X-rays. Precise positioning is critical for ventilated newborns to ensure proper lung inflation and to avoid the risks associated with unnecessary re-intubation or incorrect placement.