eduzhai > Applied Sciences > Engineering >

Collaborative Unsupervised Domain Adaptation for Medical Image Diagnosis

  • king
  • (0) Download
  • 20210506
  • Save

... pages left unread,continue reading

Document pages: 16 pages

Abstract: Deep learning based medical image diagnosis has shown great potential inclinical medicine. However, it often suffers two major difficulties inreal-world applications: 1) only limited labels are available for modeltraining, due to expensive annotation costs over medical images; 2) labeledimages may contain considerable label noise (e.g., mislabeling labels) due todiagnostic difficulties of diseases. To address these, we seek to exploit richlabeled data from relevant domains to help the learning in the target task via{Unsupervised Domain Adaptation} (UDA). Unlike most UDA methods that rely onclean labeled data or assume samples are equally transferable, we innovativelypropose a Collaborative Unsupervised Domain Adaptation algorithm, whichconducts transferability-aware adaptation and conquers label noise in acollaborative way. We theoretically analyze the generalization performance ofthe proposed method, and also empirically evaluate it on both medical andgeneral images. Promising experimental results demonstrate the superiority andgeneralization of the proposed method.

Please select stars to rate!

         

0 comments Sign in to leave a comment.

    Data loading, please wait...
×