A SURVEY ON THE ROADWAY OF CELL SEGMENTATION TECHNIQUES, IMPLEMENTATION OF UNET FOR BIOMEDICAL IMAGE SEGMENTATION

Abstract

Cell segmentation has become an important technique in obtaining accurate and image based analysis of cells morphology that help diagnosing and determining certain conditions, especially in human bodies. The aim of this thesis is to elaborate and describe several techniques that handle the issue of cell segmentation from the early stages of the cell analysis up to modern deep learning algorithms. The paperwork conducts a detailed review of existing literature, and there is attempted to implement UNet architecture for cell segmentation. A simple custom dataset of original samples of cell images is used to feed the model and the results seem to be unsatisfactory, thus making UNet of critical use in case of customized applied datasets. The limitations of the dataset such as lack of sample images and inaccuracy of labelling tools are presented as key factors to obtained results. These features are proposed to be improved in further implementations as future work.