The article presents a novel method for automatic semantic segmentation of MRI images using unsupervised federated domain adaptation. This approach is designed to overcome the challenges posed by the variability of MRI images and the need for extensive annotated data for training deep neural networks. The authors propose a method that utilizes multiple annotated source domains to adapt a model for effective use in an unannotated target domain. This process involves ensuring that the target domain data shares similar representations with each source domain in a latent embedding space. The effectiveness of the method is demonstrated through theoretical analysis and experiments on the MICCAI 2016 multi-site dataset.
Publication date: 8 Jan 2024
Project Page: https://openreview.net/forum?id=
Paper: https://arxiv.org/pdf/2401.02941
