eduzhai > Applied Sciences > Engineering >

LORCK Learnable Object-Resembling Convolution Kernels

  • king
  • (0) Download
  • 20210505
  • Save

... pages left unread,continue reading

Document pages: 18 pages

Abstract: Segmentation of certain hollow organs, such as the bladder, is especiallyhard to automate due to their complex geometry, vague intensity gradients inthe soft tissues, and a tedious manual process of the data annotation routine.Yet, accurate localization of the walls and the cancer regions in theradiologic images of such organs is an essential step in oncology. To addressthis issue, we propose a new class of hollow kernels that learn to mimic thecontours of the segmented organ, effectively replicating its shape andstructural complexity. We train a series of the U-Net-like neural networksusing the proposed kernels and demonstrate the superiority of the idea invarious spatio-temporal convolution scenarios. Specifically, the dilatedhollow-kernel architecture outperforms state-of-the-art spatial segmentationmodels, whereas the addition of temporal blocks with, e.g., Bi-LSTM,establishes a new multi-class baseline for the bladder segmentation challenge.Our spatio-temporal model based on the hollow kernels reaches the mean dicescores of 0.936, 0.736, and 0.712 for the bladder s inner wall, the outer wall,and the tumor regions, respectively. The results pave the way towards otherdomain-specific deep learning applications where the shape of the segmentedobject could be used to form a proper convolution kernel for boosting thesegmentation outcome.

Please select stars to rate!

         

0 comments Sign in to leave a comment.

    Data loading, please wait...
×