A study recently published by Cincinnati Children’s Hospital Medical Center/Zamora Hua et al. (2025), “Lack of children in public medical imaging data points to growing age bias in biomedical AI” highlights the age disparity in public medical imaging datasets, possibly stemming from a fundamental data gap. Findings drawn from a systematic review of 181 public medical imaging datasets demonstrated that children represent less than 1% of available data, resulting in “biased predictions across medical image foundation models, with the youngest patients facing the highest risk of misdiagnosis” (Zamora Hua et al, 2025, abstract). Additionally, researchers used cardiomegaly classifiers to train adult chest radiograph datasets, finding that the “models exhibit [ed] significant age bias when applied to pediatric populations, with higher false positive rates in younger children,” (Zamora Hua et al, 2025, abstract).
This study further highlights one of the biggest safety concerns in AI model development: lack of diversity (e.g., race, ethnicity, age, geographic location) in training data. It also presents a call to action for researchers, policymakers, data curators and AI developers, along with hospitals and healthcare organizations who are hesitant to share secondary data, emphasizing the critical need for pediatric representation in publicly accessible medical datasets. To account for this, researchers, hospital systems and healthcare organizations are encouraged to participate in “greater collaboration and initiatives to collect, prepare and release AI-ready pediatric data to the public [and] to support the development of pediatric AI applications,” (Zamora Hua et al, 2025, 12). As demonstrated in this study, pediatric data is necessary for the development of AI models and applications for children to ensure that the benefits of these models and applications reach all patients equally.
Read the full study here: Lack of children in public medical imaging data points to growing age bias in biomedical AI
For more information about how the world-leading imaging data provider, Vega Imaging Informatics, under core division Vega Data, can help your healthcare organization effectively and safely share its pediatric data to contribute to important AI innovations for your pediatric patient population and generate additional revenue for your practice, schedule a meeting with Al Kappel here.
Related posts:
What is your Strategy for Encounter-based Imaging?
The Value of the CIIP: Part 5 – Artificial Intelligence
The Value of the CIIP: Part 4 – Interoperability
European Congress of Radiology 2025 (#ECR25) Round-Up
Reframing Legacy Mindsets on Secondary Use of Imaging Data
Vega Pro Managed Services
Considerations on the Secondary Use of Imaging Data
Why Imaging Practices Are Slow to Implement AI
Acquiring Imaging Data for AI Developers
Is AI Revolutionizing Image Management Strategies?
How Can AI Automate Radiology Reporting Workflows?
FHIR for Imaging AI: What HL7 v2 Couldn’t Deliver
Do Imaging AI Solutions Have an ROI?
Workflow-Driven Strategies for Imaging Success
Transforming Teleradiology with AI
Get Involved to Get Ahead in Imaging Informatics
Collaborating for Success in AI Development
Vega Celebrates 8 Years of Innovation
Navigating Your Career in Imaging Informatics
Stanford’s Journey in Digital Pathology