eduzhai > Applied Sciences > Engineering >

Self domain adapted network

  • king
  • (0) Download
  • 20210505
  • Save

... pages left unread,continue reading

Document pages: 11 pages

Abstract: Domain shift is a major problem for deploying deep networks in clinicalpractice. Network performance drops significantly with (target) images obtaineddifferently than its (source) training data. Due to a lack of target labeldata, most work has focused on unsupervised domain adaptation (UDA). CurrentUDA methods need both source and target data to train models which performimage translation (harmonization) or learn domain-invariant features. However,training a model for each target domain is time consuming and computationallyexpensive, even infeasible when target domain data are scarce or source dataare unavailable due to data privacy. In this paper, we propose a novel selfdomain adapted network (SDA-Net) that can rapidly adapt itself to a single testsubject at the testing stage, without using extra data or training a UDA model.The SDA-Net consists of three parts: adaptors, task model, and auto-encoders.The latter two are pre-trained offline on labeled source images. The task modelperforms tasks like synthesis, segmentation, or classification, which maysuffer from the domain shift problem. At the testing stage, the adaptors aretrained to transform the input test image and features to reduce the domainshift as measured by the auto-encoders, and thus perform domain adaptation. Wevalidated our method on retinal layer segmentation from different OCT scannersand T1 to T2 synthesis with T1 from different MRI scanners and with differentimaging parameters. Results show that our SDA-Net, with a single test subjectand a short amount of time for self adaptation at the testing stage, canachieve significant improvements.

Please select stars to rate!

         

0 comments Sign in to leave a comment.

    Data loading, please wait...
×