Repurposing adult AI radiograph models for neonatal care through continual learning: an international multicentre study.

Continual learning adapted an adult deep learning model for neonatal endotracheal tube assessment, cutting anatomical landmark localization error to 0.67 cm versus 1.29 cm (fine-tuning) and 4.30 cm (adult-only) across 30 NICUs.

• Why it matters: Most commercial radiology AI models are not cleared for paediatric use, and adult-trained systems generalize poorly to neonates; privacy-preserving repurposing strategies are needed without data sharing.
• What they did: In an international retrospective study, 30 neonatal intensive care units across 20 countries contributed 50 neonatal chest radiographs each for sequential retraining, without centralized pooling, using continual learning.
• The result: Continual learning reduced localization error at 29 of 30 sites and improved detection of deep intubations at 27 of 30 sites, supporting privacy-preserving collaborative learning for paediatric deployment.

AI summary of the abstract · may contain errors · not medical advice · read the paper.