Journal
2019 IEEE 4TH INTERNATIONAL CONFERENCE ON SIGNAL AND IMAGE PROCESSING (ICSIP 2019)
Volume -, Issue -, Pages 685-689Publisher
IEEE
DOI: 10.1109/siprocess.2019.8868690
Keywords
CT; liver segmentation; post processing-; deep learning
Categories
Funding
- Beijing Natural Science Foundation [4172058]
- Beijing Haidian Original Innovation Joint Foundation [L182054]
Ask authors/readers for more resources
This paper proposes a novel liver segmentation framework using ResUNet with 3D probabilistic and geometric post process. Our semantic segmentation model ResUNet adds residual unit and hatch normalization layer to up sampling and down sampling part of U -Net to construct a deeper network. To quick converge, we propose a new loss function DCE, which is linearly combined by Dice loss and cross entropy loss. We use continuous several CT images as input for training and testing to explore more context information. Based on initial segmentation of ResUNet, fully connected 3D conditional random field is used to refine segmentation results by exploring 2D neighbor regions and 3D volume information. Finally, 3D connected components analyzing is used to remain some large components and reduce segmentation noise. The experimental results on public dataset LiTS show our proposed framework achieve the state of the art performance for liver segmentation.
Authors
I am an author on this paper
Click your name to claim this paper and add it to your profile.
Reviews
Recommended
No Data Available