In recent years, the accuracy of recognition methods using deep learning is much improved. Many people pay attention to the segmentation that class labels are assinged to all pixels in an image. So, we are doing research on the segmentation of cell membrane and nucleus in images. In previous research, we can not get high accuracy by deep learning in comparison with human because cell images include much noise and the shape of cell membrane is unstable. We would like to improve the robustness of noise and shape changes. This is the purpose of this research grant.