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.
