Researchers develop way of adapting adult radiology AI models to pediatric population
Researchers have developed a way of adapting adult radiology artificial intelligence models for the pediatric population.
Currently, only about 17% of U.S. Food and Drug Administration-authorized AI medical devices are labeled for use with children. That’s because most models are trained on adult data and do not translate well for infants and kids, experts write Tuesday in Nature Biomedical Engineering.
One key impediment is that medical records in pediatrics are subject to strict privacy regulations, with data tending to remain at the institution where treatment occurred. Currently, kids represent less than 1% of public medical imaging datasets.
“Because it’s so much harder to get data for children, this transformative technology is disproportionately being developed to help adults,” lead author Vishwanatha Rao, a MD-PhD student with the University of Pennsylvania and research fellow at Harvard, said in a statement Oct. 13. “Children are being left behind.”
To address this, Rao and colleagues championed a way of adapting one model, trained on adult X-rays, for the pediatric population. Their approach used data from multiple hospitals without sensitive information ever leaving each institution. “The model travels, but the data stays in-house,” experts noted.
Their study incorporated nearly 3,000 chest radiographs from 30 neonatal intensive care units across 20 countries and territories. To “simulate a scenario in which the data never leaves the hospital,” researchers trained the model sequentially on each institution’s info. Each hospital contributed 50 radiographs for model training, without any centralized data pooling.
The new model—used to aid with assessing breathing tube placement—performed admirably. Despite differences in image quality and characteristics, it outshined adult models trained on pediatric data from a single hospital at 29 of 30 sites. This even when the model was tested on that hospital’s own images, which the new model had never seen.
The fully trained AI model identified the location of the tip of the endotracheal tube with an average error of 0.67. Meanwhile, existing models adapted to kids with data from a single institution averaged errors of 1.29 cm, while adult models were off by an average of 4.3 cm.
Experts believe their approach could be generalized to adapt other adult models for pediatric use. They also see utility outside of pediatrics, with the same framework being applied to any patient population that is underrepresented in AI training data.
“This is a general framework that shows a cheap, privacy-preserving approach to take advantage of existing adult information and get it to work on children,” Rao said in the announcement. “It could usher in a whole new generation of AI models that work for kids, adapted through this broad methodology.”
Read more in Nature.
Marty Stempniak has covered healthcare since 2012, with his byline appearing in the American Hospital Association's member magazine, Modern Healthcare and McKnight's. Prior to that, he wrote about village government and local business for his hometown newspaper in Oak Park, Illinois. He won a Peter Lisagor and Gold EXCEL awards in 2017 for his coverage of the opioid epidemic.