BRB U-Net for Instance Segmentation of Cells in Microscopy

Abstract: Biological cell analysis like cell segmentation, cell classification, cell tracking etc. aid in quantitative analysis of cells which is useful for cellular level knowledge of biological activity. Microscopy imaging allows for the generation of cell images and are used for cell studies in computational biology research and clinical disease diagnosis. In this article we explore instance segmentation of DIC-C2DH-HeLa cells in microscopy images. Specifically, a deep learning model Bottleneck residual blocks U-Net (BRB U-Net) is utilized for the task of separating individual cell instances from the background. This method achieved a Dice Index of 88% for DIC-C2DH-HeLa cells.

Keywords: Instance segmentation, Deep Learning, Microscopy, Marker controlled Watershed algorithm.