eduzhai > Applied Sciences > Engineering >

Pairwise Relation Learning for Semi-supervised Gland Segmentation

  • king
  • (0) Download
  • 20210507
  • Save

... pages left unread,continue reading

Document pages: 11 pages

Abstract: Accurate and automated gland segmentation on histology tissue images is anessential but challenging task in the computer-aided diagnosis ofadenocarcinoma. Despite their prevalence, deep learning models always require amyriad number of densely annotated training images, which are difficult toobtain due to extensive labor and associated expert costs related to histologyimage annotations. In this paper, we propose the pairwise relation-basedsemi-supervised (PRS^2) model for gland segmentation on histology images. Thismodel consists of a segmentation network (S-Net) and a pairwise relationnetwork (PR-Net). The S-Net is trained on labeled data for segmentation, andPR-Net is trained on both labeled and unlabeled data in an unsupervised way toenhance its image representation ability via exploiting the semanticconsistency between each pair of images in the feature space. Since bothnetworks share their encoders, the image representation ability learned byPR-Net can be transferred to S-Net to improve its segmentation performance. Wealso design the object-level Dice loss to address the issues caused by touchingglands and combine it with other two loss functions for S-Net. We evaluated ourmodel against five recent methods on the GlaS dataset and three recent methodson the CRAG dataset. Our results not only demonstrate the effectiveness of theproposed PR-Net and object-level Dice loss, but also indicate that our PRS^2model achieves the state-of-the-art gland segmentation performance on bothbenchmarks.

Please select stars to rate!

         

0 comments Sign in to leave a comment.

    Data loading, please wait...
×