Bringing transparency into the field of AI in radiology is our core value and we are always looking for ways to take that one step further. A recent commentary in npj Digital Medicine explores how “Model Cards”, standardized, layered summaries inspired by nutrition labels, could offer a clear and accessible way to communicate essential information about medical AI applications.
Developed by the Coalition for Health AI (CHAI), these cards are designed to present key details such as intended use, performance metrics, risks, fairness, and bias, in a format tailored to clinicians, patients, IT teams, and regulators alike.
However, the authors offer a word of caution: without thoughtful implementation, Model Cards risk becoming marketing tools rather than meaningful sources of transparency. They identify three major pitfalls: duplication of existing regulatory documents, lack of usability testing with real users, and the danger of unverifiable claims. To truly support safe adoption, Model Cards must be linked to auditable data and refined through user feedback.
Read the full study for a deeper dive into this important topic.
Read full study
Could transparent model cards with layered accessible information drive trust and safety in health AI?
NPJ, digital medicine
Abstract
We place ‘Model Cards’ and graphical ‘nutrition labels’ for health AI in context with the information needs of patients, health care providers and deployers. We discuss the applicability of Model Cards for General Purpose AI (GPAI) models. If these approaches are to be useful and safe they need to be integrated with regulatory approaches and linked to deeper layers of open and detailed model information and optimized through user testing.