Medical Image Segmentation Based on U-Net

Abstract

Traditional statistical methods, although having a solid theoretical foundation, have been challenged in terms of their efficiency as well as their generalization ability in the face of the ever-increasing amount of massive data. With the rise of deep learning in recent years, the use of new tools such as convolutional neural networks to get information from data has become a new option. In particular, in the field of imaging, segmentation of medical images is important for tasks such as determining the type of disease and the location of lesions, which are excellent application areas for deep learning. U-Net is a particularly important deep model structure with good results for segmentation of medical images. However, there is a lack of discussions on the application of U-Net in the clinical field. In this paper, we introduce traditional image segmentation methods and U-Net, analyze the advantages of deep learning techniques in the field of image segmentation. In addition, we applied U-Net to the problem of cell segmentation and segmentation of covid-19 CT images, showing the potential of U-Net for clinical applications.