Identifying the Cells' Nuclei Using Deep Learning

Abstract

The development time of new drugs is a long and complex process with different stages of analysis and screening. In most of the analysis stage, the first step is the detection of cells' nuclei. This allows researchers to identify the individual cells in a sample because most of the cells contain a nucleus filled with DNA (Deoxyribonucleic acid). Identification of cell nuclei help measure the reactions of cells when exposed to various treatments and lead to understanding the biological process underlying the work. This process is laborious and slow because it requires the identification and analysis of thousands of images at a time. Thus, automating this step would speed up the analytical process. Therefore, the time to market for a new drug can be significantly reduced. This work proposes three deep learning techniques to segment the images and to identify the cells' nuclei. Modified architectures based on semantic segmentation networks such as UNet, SegNet, and FCN were developed. The obtained results are very interesting with F1-Scores ranging from 94% for FCN to 96% for UNet. SegNet follows closely UNet with an F1-Score of 95%.