Cell Segmentation of Histopathological Images of Glioma Using Voronoi Tessellation and Quadtree Representation

Abstract

Automatic cell segmentation is a challenging task in histopathological image analysis which is responsible for examining tissues and cells for diagnosing the severity of cancer in patients. Cell segmentation is the technique of breaking down a microscopic image area into sections that reflect individual cell occurrences. Finding the high density of cells and finer edge detection of cells is complicated due to the overlapping regions. To overcome this difficulty, a novel method of segmenting Whole Slide Images (WSI) of Low Grade Glioma (LGG) and High Grade Glioma (HGG) using voronoi tessellation of polygon approximation is proposed. The suggested approach consistently segments images of distinct cell types growing in dense cultures that were captured using various morphological techniques. The edge detection of the cells obtained from voronoi is finer compared to other existing edge detection methods. Quadtree image representation is implemented using proposed approach which is utilized to calculate the cell count and density estimate of tumor from voronoi tessellation in O(log4 n). The factors gleaned from the histopathological analysis helps the pathologist in finding the densely populated brain tumors cells in WSI and treating the patient accordingly.