The fairness of medical imaging AI models across sociodemographic groups is indispensable for the successful implementation of trustworthy medical AI. The independent FAIMI consortium (Fairness of AI in Medical Imaging) has held vibrant annual workshops at MICCAI, the leading international conference on medical image analysis, since 2023. At this year’s MICCAI in Daejeon, South Korea, ODELIA’s Eike Petersen (Fraunhofer MEVIS) co-organized both the annual workshop as well as – held for the first time – a separate tutorial on the fairness of medical imaging AI.
In both events, recurring themes included
- moving beyond often-claimed ‘fairness-accuracy trade-offs’ and developing models that are simultaneously more fair and more accurate,
- investigations into root causes of observed demographic unfairness, and
- rigorous methodology for the reliable detection and quantification of biases.
Dishantkumar Sutariya and Eike Petersen (both Fraunhofer MEVIS) also presented a publication at the workshop, meval: A Statistical Toolbox for Fine-Grained Model Performance Analysis, winning a best poster award.
The toolbox is available open-source on GitHub and is being used and further developed within ODELIA for analyzing the robustness of breast MRI cancer classification AI models.
