Federated Learning in Opthalmology

Medical data is especially sensitive and its centralization is often not possible. This can make the development of machine learning models challenging, as traditional approaches rely on large data sets for model training and evaluation.

In a joint effort with researchers from the ZEISS Group and Carl Zeiss Meditec AG, we applied our decentralized approach to carry out anomaly detection in OCT image data from the eye.

ZEISS OCT Scanner, Type CIRRUS HD-OCT 5000, Image Credits: Carl Zeiss Meditec Inc.

Share models, not data! 

In the project, we apply a paradigm shift in machine learning. Instead of transferring large data sets, we transfer only the individual models. These models are trained locally on-the-edge and then combined centrally into a more powerful model. More information on our Decentralized Machine Learning approach can be found here: mlx.

Models Exchange While OCT Data Stays Local - Using prenode's Software mlx

ARVO Conference on Eye Research

The Association for Research in Vision and Ophthalmology (ARVO) hosts their annual meeting in May 2021. We are proud to actively take part and present an evaluation of the described Federated Learning approach on eye data from OCT scans to professionals in the field.

Authors: Robin HirtChristian KungelCaroline DieterichGary LeeDominik FischerNiranchana ManivannanAditya NairHugang RenSophia Yu and Alexander Ulrich.

Link to the presentation workshop: ARVO Conference 2021.

Interested in learning more? Contact us!