Source-Free Unsupervised Domain Adaptation with Norm and Shape Constraints for Medical Image Segmentation

Unsupervised domain adaptation (UDA) is one of the key technologies to solve a problem where it is hard to obtain ground truth labels needed for supervised learning. In general, UDA assumes that all samples from source and target domains are available during the training process. However, this is not a realistic assumption under applications where data privacy issues are concerned. To overcome this limitation, UDA without source data, referred to source-free unsupervised domain adaptation (SFUDA) has been recently proposed. Here, we propose a SFUDA method for medical image segmentation. In addition to the entropy minimization method, which is commonly used in UDA, we introduce a loss function for avoiding feature norms in the target domain small and a prior to preserve shape constraints of the target organ. We conduct experiments using datasets including multiple types of source-target domain combinations in order to show the versatility and robustness of our method. We confirm that our method outperforms the state-of-the-art in all datasets.