Segmentation of Nuclei in Histopathology images using Fully Convolutional Deep Neural Architecture

Nuclei segmentation is an initial step in the automated analysis of digitized microscopic images. This paper focuses on utilizing the LinkNET-34 architecture for semantic segmentation of nuclei from the H&E stained breast cancer histopathology images. The segmentation process is implemented in two stages where in the first stage the H&E stained images are pre-processed to reduce the variance caused because of staining the microscopic images and scanning the slides. During the second stage the preprocessed images are given as input to the LinkNET network which consists of both down-sampling and up-sampling layers. The network is trained using a set of WSI patches released during the Data Science bowl 2018 competition. The performance of the deep learning model is evaluated based on the segmentation accuracy measured using the Dice Coefficient.