Bias and Fairness in Medicine DFF, Co-PI · 2020–2023
In this DFF-funded project, I was Co-PI alongside Aasa Feragen and Sune Holm, examining the nature of bias in the diagnosis and treatment of disease, using major depressive disorder as an example. At the interaction of ethics/philosophy and mathematical modelling, we redefined fairness for the medical domain and developed predictive AI algorithms that are fair in this new sense.
How it started
It all started back in 2019, when Aasa Feragen applied for a DFF thematic call with myself and Sune Holm as Co-PIs, to look at "Bias and Fairness in Medicine". The project was funded, started in 2020, and produced a steady stream of output – both public outreach and publications, as well as seminars and workshops.
A special thanks goes to Eike Petersen, who was instrumental in pushing the project forward, and to Merete Osler, who we were introduced to through this project and who advised on the epidemiological side of the work.
What came out of it
The project helped kick-start FAIMI – Fairness of AI in Medical Imaging, which has since become an official MICCAI Special Interest Group. Through a series of interdisciplinary events, FAIMI aims to advance the discourse around fairness issues in the medical image analysis community. It now runs a standing series of free webinars, has organised two MICCAI workshops, maintains a YouTube channel, and has collected papers for a special issue in MELBA.
The project also led directly to the Lundbeck Experiment pilot project ("Measuring the reliability of trustworthy healthcare AI for mental health risk prediction"), which in turn overlapped with and grew into the ongoing TRUSTMIND project.