Research shows that only 1 of 46 studies presented at a recent medical imaging conference used pediatric data, highlighting its scarcity in research.

The trend of aggregating datasets for foundation models often results in the loss of age data, worsening bias against children in medical AI.

A 2023 study found that only 22 out of 692 FDA-approved medical AI devices are evaluated for pediatric use, revealing a significant gap in AI benefits for children.

A review of public medical imaging datasets found that less than 1% of data is from children, despite them making up 25% of the global population.

A recent pre-print study suggests that the underrepresentation of children in public datasets hampers the development of pediatric AI.

Many medical imaging datasets lack patient age information, indicating that dataset creators often overlook the importance of demographic details like age.

In fact, some AI applications have almost no pediatric samples, with a review finding only 5 pediatric MRI images out of nearly 19,000 total images.

The study emphasizes the need for the AI healthcare community to collaborate on initiatives to collect and share pediatric data to advance pediatric AI applications.

To address this gap, the American College of Radiology has established a Pediatric AI working group to promote equitable access to safe medical AI technologies for children.

The scarcity of pediatric AI models forces clinicians to rely on 'off-label' adult AI models, which can be risky; for example, a model predicting cardiomegaly had a 50% error rate in young children.