We finally put together a scientific paper presenting and validating Dafne’s functionality and performance. We found that the lifelong collaborative learning concept of Dafne is indeed able to produce models that evolve in the right direction, with generally higher performance over time.

The paper is at the moment uploaded to arXiv here and I hope you will enjoy reading it!

I think that the results are interesting, and all the evaluation is open and public. You can find a jupyter notebook that generates both the images and the statistical analysis live here on Github.

If you are using Dafne for your research, I would appreciate if you could cite the paper in your future publications. The citation is the following:

Santini F, Wasserthal J, Agosti A, et al. Deep Anatomical Federated Network (Dafne): an open client/server framework for the continuous collaborative improvement of deep-learning-based medical image segmentation. 2023 doi: 10.48550/arXiv.2302.06352.