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

Despite growing clinical adoption of artificial intelligence in radiology, fewer than one in five commercially available models are cleared for paediatric use. Existing systems are trained predominantly on adult data and generalize poorly to neonates, yet efficient strategies for repurposing them without data sharing or large-scale local data curation are lacking. Here we present an international retrospective study of neonatal chest radiographs from 30 neonatal intensive care units across 20 countries and territories, evaluating continual learning as a privacy-preserving approach to adapt a deep learning model for adult endotracheal tube assessment to neonatal care. Each hospital contributed 50 radiographs for sequential model retraining without centralized data pooling. Continual learning achieved lower anatomical landmark localization error (0.67 cm versus 1.29 cm for hospital-specific fine-tuning and 4.30 cm for adult-only models) at 29 of 30 sites, with improved detection of deep intubations at 27 of 30 sites. These findings demonstrate that privacy-preserving collaborative learning can address critical gaps in paediatric AI deployment.

Authors

Institutions

Publication Details

Journal
Nature Biomedical Engineering
Published
2026-10-13
DOI
https://doi.org/10.1038/s41551-026-01812-6
Primary Topic
Artificial Intelligence in Healthcare and Education
Type
article
Field-Weighted Citation Impact
0.00

Funders

Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
article

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

Henrik Marklund, Pranav Rajpurkar, Tijana Prodanović, Atchara Mahayosnond et al.
Nature Biomedical Engineering
Artificial Intelligence in Healthcare and Education
article

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

Henrik Marklund, Pranav Rajpurkar, Tijana Prodanović, Atchara Mahayosnond, Emma Chen, Reza Shahriarirad, Sira Sriswasdi, Amirhossein Erfani, Alain S. Abi‐Ghanem, Moria Be’er, Daniel Truhn, Jens Kleesiek, Ümit Ayşe Tandırcıoğlu, Phumza Nongena, Kelsey L. Pomykala, Georg Marcus Schmölzer, Susan Cheng Shelmerdine, Almudena Alonso‐Ojembarrena, Desmond Shi Wei Lim, Israel Amirav, Kent R. Kleinschmidt, Sathvik Suryadevara, Thomas Küstner, Yasuhito Mitsuyama, Didem Arman, Wen‐Hui Fang, David Andrew Mong, Khalid Al-Kharazi, Montserrat Bret Zurita, Evan James Zucker, Oishi Banerjee, Saif Afat, María José Veloza, Mossad Abdelhak Shaban Mohamed, Mario Scala, Nikola S. Prodanovic, Serdar Alan, Julian F. Niedermair, Serdar Cömert, Juan Sebastián Osorio Valencia, Nikolaus Stranger, Soroosh Tayebi Arasteh, Melissa L. Cooke, Piyapong Khumrin, Sebastian Tschauner, Daiju Ueda, James Thomas Patrick Decourcy Hallinan, Morgan P. McBee, Wei‐Shiang Lin, Richard H. Jones, Naila Nizami, Naela Himayati Afifah, Sung Eun Kim, Jiaxi Lu, J Ramírez Palacio, Abdullah Al Shahrani, Rithvik Akula, Sri Sai Dinesh Jaliparthi, Lucas Bijnens, Suphakris Tanmuk, Amna Kashgari, Alexandra Nemes, Oscar Jaramillo Robledo, Dr. Rafeah Khan, Agustina Saenz, Michael Sadler, Zafar Iqbal, Wei-Ting Lin, Safaa Al Zakleet, Yung-Chieh Lin, Vishwanatha M. Rao, Oscar Julian López Uribe, Wan Irfan Bin W Mustapha, Rozan M. Irfan, Esther Chung
article en

Abstract

Despite growing clinical adoption of artificial intelligence in radiology, fewer than one in five commercially available models are cleared for paediatric use. Existing systems are trained predominantly on adult data and generalize poorly to neonates, yet efficient strategies for repurposing them without data sharing or large-scale local data curation are lacking. Here we present an international retrospective study of neonatal chest radiographs from 30 neonatal intensive care units across 20 countries and territories, evaluating continual learning as a privacy-preserving approach to adapt a deep learning model for adult endotracheal tube assessment to neonatal care. Each hospital contributed 50 radiographs for sequential model retraining without centralized data pooling. Continual learning achieved lower anatomical landmark localization error (0.67 cm versus 1.29 cm for hospital-specific fine-tuning and 4.30 cm for adult-only models) at 29 of 30 sites, with improved detection of deep intubations at 27 of 30 sites. These findings demonstrate that privacy-preserving collaborative learning can address critical gaps in paediatric AI deployment.

Nature Biomedical Engineering
Titu Maiorescu University (RO), Johannes Kepler University of Linz (AT), Newark Beth Israel Medical Center (US), Brigham and Women's Hospital (US), Tri-Service General Hospital (TW), Qatar Airways (Qatar) (QA), Harvard University (US), University of Kragujevac (RS), University of Caldas (CO), Medical University of South Carolina (US), University of Alberta (CA), Chulalongkorn University (TH), Tel Aviv University (IL), Universitas Gadjah Mada (ID), Palo Alto University (US), University of Luxembourg (LU), Shiraz University of Medical Sciences (IR), Medical University of Graz (AT), Tel Aviv Sourasky Medical Center (IL), Great Ormond Street Hospital (GB), Hospital Universitario La Paz (ES), Somerset Hospital (ZA), Indus Hospital (PK), Namazi Hospital (IR), International Islamic University Malaysia (MY), Children's Hospital Colorado (US), Maharaj Nakorn Chiang Mai Hospital (TH), King Abdul Aziz University Hospital (SA), Kepler Universitätsklinikum (AT), Essen University Hospital (DE), Universitätsklinikum Aachen (DE), Biomedical Research and Innovation Institute of Cadiz (ES), University Children's Hospital Tübingen (DE), Sidra Medical and Research Center (QA), Stanford Medicine (US), Istanbul Eye Hospital (TR), National University Hospital (SG), King Abdullah Medical City (SA), Rumah Sakit Umum Pusat Dr. Sardjito (ID), National Cheng Kung University Hospital (TW), Children's Specialized Hospital (US), Osaka Metropolitan University (JP), Luxembourg Centre for Systems Biomedicine (LU), Asociación Colombiana de Radiología (CO), Stiftung Universitätsmedizin Essen (DE), National Guard Health Affairs (SA), Kırıkkale University (TR), Saint Louis University (US), Chiang Mai University (TH), University of Colorado Anschutz Medical Campus (US), National Defense Medical Center (TW), University of Pennsylvania (US), University of Tübingen (DE), RWTH Aachen University (DE), National Cheng Kung University (TW), Stanford University (US), American University of Beirut (LB), KU Leuven (BE)
Children's Hospital Colorado, American University of Beirut, University of South Carolina, Kepler Universitätsklinikum, Sidra Medicine, University of Alberta, Chulalongkorn University, Indus Hospital, Shiraz University of Medical Sciences, National Cheng Kung University, Medizinische Universität Graz, Tri-Service General Hospital
Openalex Percentile: Top 20%
Artificial Intelligence in Healthcare and Education
DOIOpen Access PDFOpenAlex
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.