eduzhai > Applied Sciences > Engineering >

Stain Style Transfer of Histopathology Images Via Structure-Preserved Generative Learning

  • king
  • (0) Download
  • 20210505
  • Save

... pages left unread,continue reading

Document pages: 11 pages

Abstract: Computational histopathology image diagnosis becomes increasingly popular andimportant, where images are segmented or classified for disease diagnosis bycomputers. While pathologists do not struggle with color variations in slides,computational solutions usually suffer from this critical issue. To address theissue of color variations in histopathology images, this study proposes twostain style transfer models, SSIM-GAN and DSCSI-GAN, based on the generativeadversarial networks. By cooperating structural preservation metrics andfeedback of an auxiliary diagnosis net in learning, medical-relevantinformation presented by image texture, structure, and chroma-contrast featuresis preserved in color-normalized images. Particularly, the smart treat ofchromatic image content in our DSCSI-GAN model helps to achieve noticeablenormalization improvement in image regions where stains mix due to histologicalsubstances co-localization. Extensive experimentation on public histopathologyimage sets indicates that our methods outperform prior arts in terms ofgenerating more stain-consistent images, better preserving histologicalinformation in images, and obtaining significantly higher learning efficiency.Our python implementation is published onthis https URL.

Please select stars to rate!

         

0 comments Sign in to leave a comment.

    Data loading, please wait...
×