Sidra Medicine advances neonatal AI imaging research through global study

QATAR—Sidra Medicine has contributed to an international study published in Nature Biomedical Engineering in November 2026, demonstrating how artificial intelligence (AI) models developed for adults can be adapted to improve medical imaging in newborns while protecting patient data privacy.

Titled “Repurposing Adult AI Radiograph Models for Neonatal Care Through Continual Learning: An International Multicenter Study,” the research involved 30 neonatal intensive care units (NICUs) across 20 countries and territories.

Led through the Medical AI Data for All (MAIDA) initiative at Harvard Medical School, the study explored ways to adapt existing AI technologies to neonatal care without requiring hospitals to share patient imaging data centrally.

Addressing the paediatric AI gap

Most commercially available medical imaging AI models rely on adult patient data, limiting their suitability for children and newborns.

Fewer than one in five have regulatory clearance for paediatric use, highlighting a gap in technologies designed for younger patients.

The research investigated whether adult-trained models could be adapted to recognise neonatal anatomy and support clinical decisions.

At Sidra Medicine, the project was led by Dr. Khalid Al-Kharazi, Senior Attending Physician and Director of the Pediatric Neurosurgery Fellowship Programme, and Zafar Iqbal, Manager of Imaging Information Systems, with support from Dr. Abdusamea Shabani, Chair of Radiology.

Prof. Khaled Al-Surimi, Head of the Department of Healthcare Management at the University of Doha for Science and Technology, facilitated the partnership with the Harvard Medical School team.

Dr. Al-Kharazi, a co-author, noted that AI’s clinical value depends on its ability to perform reliably across different patient populations.

He explained that newborns have distinct anatomical and clinical characteristics that require specialised medical imaging approaches.

Continual learning improves imaging accuracy

The study focused on endotracheal tube placement, a critical procedure in neonatal intensive care.

Incorrect positioning can cause serious complications, making accurate assessment essential for clinicians managing critically ill newborns.

Researchers used a technique called continual learning to adapt an adult-trained AI model using neonatal chest X-rays from participating hospitals.

Each institution contributed between 100 and 300 images, allowing the model to learn from diverse patient populations without centralising medical records.

Sidra Medicine contributed 100 anonymised neonatal chest X-rays.

The adapted model reduced the average anatomical landmark localisation error to 0.67 centimetres, compared with 1.29 centimetres for hospital-specific fine-tuning and 4.30 centimetres for the original adult-trained model.

It also improved localisation accuracy at 29 of the 30 participating sites and detection of deeply positioned endotracheal tubes at 27 sites.

Iqbal said Sidra Medicine’s participation reflects its commitment to international research collaboration and developing AI technologies that address gaps in paediatric healthcare.

He highlighted the importance of combining clinical expertise and imaging data to better represent children in medical AI research.

Dr. Shabani, who selected the radiographs for the study, noted that continual learning offers a potential approach to adapting adult-focused imaging technologies for paediatric settings while maintaining patient privacy.

Dr. Al-Kharazi also acknowledged Prof. Al-Surimi’s contribution to establishing the partnership, expressing Sidra Medicine’s interest in continued collaboration on healthcare innovation and AI research in Qatar and internationally.

Sign up HERE to receive our email newsletters with the latest news and insights from Africa and around the world, and follow us on LinkedIn for updates.