Johanna Schwinn from Augsburg University Hospital gave a presentation on the potential of Federated Learning (FL) compared to Centralized Learning (CL) and Local Learning (LL) at the Medical Informatics Europe (MIE) 2024, conference of the EFMI-European Federation for Medical Informatics on August 29, 2024.

In her study, Johanna investigated the effectiveness of FL in predicting oxygen saturation (SpO2) in intensive care patients based on five large databases. Her results show that FL not only achieves comparable, but in some cases even better results than CL and LL – while strictly maintaining the privacy of patient data.

This study is the first of its kind to combine all five publicly available ICU databases.

FL is proving to be a promising alternative, especially for hospitals with smaller datasets that still want to benefit from powerful and privacy-friendly machine learning (ML) models.

“In the future, we will test more complex data to expand our understanding of FL’s capabilities in different medical contexts and further improve prediction accuracy,” says Schwinn.

You can read the full paper by authors Johanna Schwinn, Seyedmostafa Sheikhalishahi, Matthäus Morhart, Mathias Kaspar and Christian Hinske here: https://bit.ly/474RJly

An outstanding contribution to the further development of clinical decision support!