Skip to Main Content
 

Global Search Box

 
 
 

ETD Abstract Container

Abstract Header

Laplacian Pyramid FCN for Robust Follicle Segmentation

Wang, Zhewei
http://rave.ohiolink.edu/etdc/view?acc_num=ohiou1565620740447982

Abstract Details

2019, Master of Science (MS), Ohio University, Biomedical Engineering (Engineering and Technology).
The thyroid gland secretes hormones that regulate metabolic rate and protein synthesis. At the microscopic level, the thyroid consists of three major components: follicles, follicular cells and parafollicular cells. As the functional status of thyroid glands can often be monitored through certain morphological characteristics (e.g., size and shape) of their follicles, segmenting follicles and/or follicular cells from histology images is of great importance. Follicle extraction is a challenging task as well-defined boundaries are often absent between neighboring follicles. Traditional edge-based and region-based segmentation solutions, such as watershed, region growing and active contours, would be insufficient to separate individual follicles. Fully convolutional networks (FCNs), a family of deep learning models with object recognition and boundary delineation capacity, can potentially provide a remedy. In this thesis, we propose a pyramid network structure to improve FCN-based segmentation solutions and apply it to label thyroid follicles and follicular cells in histology images. Our design is based on the notion that a hierarchical updating scheme, if properly implemented, can help FCNs capture the major objects, as well as their structural details in an image. To this end, we devise a Laplacian-residual module to be mounted on consecutive network layers, through which pixel labels would be propagated from the coarsest layer towards the finest layer in a bottom-up fashion. We add five Laplacian units along the decoding path of a modified U-Net to construct our segmentation network, Lap-Seg-Net. Four different combination strategies are explored to best utilize the available colloid ground-truth delineations, as well as to ensure the proper topological hierarchy between colloid and follicles. Experiments demonstrate that the multi-resolution set-up in our model is effective in producing segmentations with improved accuracy and robustness.
Jundong Liu (Advisor)
Douglas Goetz (Committee Chair)
Li Xu (Committee Member)
Thomas Rosol (Committee Member)
61 p.
Biomedical Engineering
Deep Learning; Image Segmentation; Follicle Segmentation

Recommended Citations

Citations

  • Wang, Z. (2019). Laplacian Pyramid FCN for Robust Follicle Segmentation [Master's thesis, Ohio University]. OhioLINK Electronic Theses and Dissertations Center. http://rave.ohiolink.edu/etdc/view?acc_num=ohiou1565620740447982

    APA Style (7th edition)

  • Wang, Zhewei. Laplacian Pyramid FCN for Robust Follicle Segmentation. 2019. Ohio University, Master's thesis. OhioLINK Electronic Theses and Dissertations Center, http://rave.ohiolink.edu/etdc/view?acc_num=ohiou1565620740447982.

    MLA Style (8th edition)

  • Wang, Zhewei. "Laplacian Pyramid FCN for Robust Follicle Segmentation." Master's thesis, Ohio University, 2019. http://rave.ohiolink.edu/etdc/view?acc_num=ohiou1565620740447982

    Chicago Manual of Style (17th edition)

ohiou1565620740447982
202
© 2019, all rights reserved.
This open access ETD is published by Ohio University and OhioLINK.