Doha:
Sidra Medicine, a member of Qatar Foundation, has contributed to a major
international study demonstrating how artificial intelligence models originally
developed for adult patients can be adapted to support neonatal care. Published
in Nature Biomedical Engineering, the research highlights a promising approach
to improving the accuracy of AI-assisted medical imaging for newborns.
Titled
"Repurposing Adult AI Radiograph Models for Neonatal Care Through
Continual Learning: An International Multicenter Study," the research
featured input from 30 neonatal intensive care units across 20 countries and
territories, conducted through the Medical AI Data for All initiative at
Harvard Medical School. The study examined how existing AI models could be adapted
to meet the needs of newborn patients while protecting patient data privacy.
The
research focused on assessing the positioning of endotracheal tubes used to
support breathing in critically ill newborns, with researchers investigating
whether AI models trained to assess tube placement in adults could be adapted
to detect potentially incorrect positioning in newborns. Using continual
learning, the team adapted an existing model through sequential training using
neonatal chest X-rays from participating hospitals, with each institution
contributing 100-300 radiographs.
Dr Khalid Al-Kharaz, Senior Attending Physician and co-author, said the study demonstrates how existing AI models can be adapted to better support neonatal care, potentially helping clinicians assess medical images more accurately. Dr Abdusamea Shabani, Chair of Radiology at Sidra Medicine who selected the radiographs, said the findings suggest continual learning could provide a practical approach for adapting existing medical imaging models to pediatric settings while maintaining patient data privacy. Sidra Medicine contributed 100 anonymised neonatal chest X-ray images to support the research.
