Sidra Medicine contributes to international AI study advancing neonatal care

Doha: Sidra Medicine, part of Qatar Foundation, participated in an international study showcasing how AI models initially created for adult patients can be tailored to aid neonatal care. Published in the prestigious journal Nature Biomedical Engineering, the research offers a promising avenue to enhance the precision of AI-powered medical imaging for infants.

The study, titled "Repurposing Adult AI Radiograph Models for Neonatal Care Through Continual Learning: An International Multicenter Study," involved 30 neonatal intensive care units from 20 different countries. Led by Harvard Medical School's Medical AI Data for All (MAIDA) initiative, the investigation explored ways to modify existing AI models to suit newborn patients while safeguarding their privacy.

Dr. Khalid Al-Kharaz, Senior Attending Physician, Director of the Pediatric Neurosurgery Fellowship program at Sidra Medicine and co-author of the study, remarked, "AI holds immense potential to aid clinical decision-making, but its efficacy hinges on its performance across diverse patient groups. This research demonstrates how existing AI models can be adapted to cater better to the unique needs of neonatal patients, potentially enabling clinicians to assess medical images more accurately and detect issues requiring immediate attention."

Dr. Khalid Al Kharaz, Manager of Imaging Information Systems at Sidra Medicine and co-author of the study, added, "Sidra Medicine's involvement in this global research underscores our dedication to advancing medical innovation through collaboration with top institutions worldwide. By sharing our expertise and clinical imaging data, we are addressing a critical gap in pediatric healthcare.

The findings also underscore how international partnerships can expedite the creation of AI technologies more inclusive of children and their specific healthcare requirements."

Dr. Abdusamea Shabani, Chair of Radiology at Sidra Medicine and the one who selected the radiographs for the study, stated, "This study marks a significant step towards addressing the shortcomings of AI technologies primarily designed for adult populations. Its insights suggest that continual learning could offer a practical method for adapting existing medical imaging models to pediatric settings while preserving patient data privacy."

The research concentrated on evaluating the placement of endotracheal tubes, which aid in breathing for critically ill newborns. Correct tube positioning is vital in neonatal intensive care, as improper placement can result in severe complications. Researchers explored if AI models originally trained to assess tube placement in adults could be adapted to recognize anatomical landmarks and identify potential incorrect tube placement in newborns.

Through a technique called continual learning, the research team fine-tuned an existing AI model using sequential training with neonatal chest X-rays from participating hospitals. Each institution contributed between 100-300 radiographs, facilitating the model's ability to learn from varied patient populations without necessitating the central aggregation of medical imaging data. Sidra Medicine contributed 100 anonymized neonatal chest X-ray images to contribute to the research.