Bias and Representation in Public Cardiac Imaging AI Datasets

A review of 38 public cardiac imaging datasets finds nearly 80% from high-income countries, none from Africa or South America. Fewer than two-thirds report demographics, and only a minority are linked to outcomes. This 'echo chamber' may limit AI equity.

• Of 38 publicly available echocardiography, cardiac MRI, and cardiac CT datasets, approximately 80% originate from high-income nations, with zero datasets from Africa or South America.
• Fewer than two-thirds of these datasets report any demographic information, and only a small minority are linked to clinical outcomes, creating a 'echo chamber' that may limit global AI generalizability.
• Harmonization and label quality often improve model performance more than changes in architecture, yet consistent racial and ethnic performance gaps persist.

Automated summary

RadPigeon summaries are original and for information only. They are not clinical advice.