You never know when you’re the exception – Jakob Ambsdorf is working towards fair AI for every pregnancy scan

A research initiative called FairFM aims to make artificial intelligence used in pregnancy scans more accurate and fair for every patient. Led in part by postdoc Jakob Ambsdorf through the DIREC project, researchers and industry partners are developing advanced models to help doctors answer critical questions about fetal health.

Traditional medical AI often performs poorly when applied to patients, scanners, or hospitals that differ from the original training data. To solve this, the project utilizes foundation models that first learn from massive amounts of unlabelled data. This approach helps the technology account for rare scanner types, unusual patient characteristics, and underrepresented subgroups that are typically missed by standard datasets.

The project focuses on two of the hardest tasks in fetal medicine which are predicting spontaneous preterm births and estimating fetal weight accurately. Partner hospitals around the world are currently testing these models.

Collaboration between computer scientists, clinicians, and industry partners is driving the work forward to ensure the technology moves out of the lab and into actual patient care. If successful, the new models will ensure that patients who are statistical exceptions still receive accurate diagnoses and proper medical treatment during their pregnancy scans.