Bilkent University
Department of Computer Engineering


Rule Based Segmentation of Colon Glands


Simge Yücel
MS Student
Computer Engineering Department
Bilkent University

Gland segmentation is an important step to automate the analysis of biopsies that contain glandular structures. This work proposes to decompose an image into smaller subregions, defines a set of rules based on the morphological properties of the subregions, and determines whether or not a subregion corresponds to a gland using this rule set. For subregion decomposition, this work first finds the superpixels on an image and characterizes each superpixel based on its color and the spatial distribution of its adjacent superpixels. Then for each superpixel type, it finds the connected components on the superpixels of this type to define the subregions. Eventually, it defines a set of rules based on the morphological properties of these components and proposes to use these rules for gland segmentation. Our preliminary experiments show that the use of superpixels, instead of directly using pixel values, facilitates to define a better image representation, which is expected to give accurate gland segmentation results.


DATE: 10 April, 2017, Monday @ 16:05