Abstract:
Medical image processing has been a field of large interest recently. Our field of
interest in this work is Medical Image Segmentation. We perform Medical Image
Segmentation to extract or visualise information regrating tissues cells or organs. In this
work we are trying to extract cell body from the background. This segmentation is vital as
it can help extract information about cells morphology which can later be used to train
another neural network for prediction if a cell is healthy or non healthy. It can also be used
to predict the type of cancer. We will use a FCN to perform the segmentation. From all
FCN networks we chose U-Net to perform image segmentation. U-Net has proved to be
very efficient for small datasets as our dataset. In this work we trained U-Net in different
ways and changed its parameters to obtain the best model. We also designed two model
evaluation functions for our field of interest. In the end we present our best model of U-Net
to perform Cell Image Segmentation based on our experiment set.
