eduzhai > Applied Sciences > Engineering >

Domain-Adversarial Learning for Multi-Centre Multi-Vendor and Multi-Disease Cardiac MR Image Segmentation

  • king
  • (0) Download
  • 20210507
  • Save

... pages left unread,continue reading

Document pages: 10 pages

Abstract: Cine cardiac magnetic resonance (CMR) has become the gold standard for thenon-invasive evaluation of cardiac function. In particular, it allows theaccurate quantification of functional parameters including the chamber volumesand ejection fraction. Deep learning has shown the potential to automate therequisite cardiac structure segmentation. However, the lack of robustness ofdeep learning models has hindered their widespread clinical adoption. Due todifferences in the data characteristics, neural networks trained on data from aspecific scanner are not guaranteed to generalise well to data acquired at adifferent centre or with a different scanner. In this work, we propose aprincipled solution to the problem of this domain shift. Domain-adversariallearning is used to train a domain-invariant 2D U-Net using labelled andunlabelled data. This approach is evaluated on both seen and unseen domainsfrom the M &Ms challenge dataset and the domain-adversarial approach showsimproved performance as compared to standard training. Additionally, we showthat the domain information cannot be recovered from the learned features.

Please select stars to rate!

         

0 comments Sign in to leave a comment.

    Data loading, please wait...
×