Deep Learning-Based Approaches for Nucleus Segmentation

Abstract

The accurate identification of cell nuclei is a critical aspect of various analyses, given that human cells, numbering around 30 trillion, contain DNA as their genetic code. In this research paper, we provide a comprehensive overview of deep learning-based techniques for nucleus segmentation. We have replicated and assessed the state-of-the-art methods using datasets like FCN, SegNet, U-net, and DoubleU-net, with a focus on the Data Science Bowl 2018 dataset comprising 670 training data folders and 65 testing data folders. Our experimental findings reveal that DoubleU-Net surpasses U-Net and other baseline models, yielding more precise segmentation masks. This promising outcome suggests that DoubleU-Net could serve as a robust model for addressing various challenges in medical image segmentation.