U-Net及其变体在医学图像分割中的应用研究综述

医学图像分割可以为临床诊疗和病理学研究提供可靠的依据，并能辅助医生对病人的病情做出准确的判断。基于深度学习的分割网络的出现解决了传统自动分割方法鲁棒性不强、准确率低等问题。U-Net凭借其出色的性能在众多的分割网络中脱颖而出，研究者以U-Net为基础相继提出了多种改进变体。以U-Net网络及其变体为主线，首先详细介绍U-Net的网络结构及常用改进方法；然后根据分割对象的不同，将U-Net变体网络进一步划分为泛用型分割网络及特定型分割网络，并就其在医学图像分割中的研究进展进行论述；最后，分析了目前研究中工作尚存在的难点与问题，并对今后的发展方向进行展望。

Review on Applications of U-Net and its Variants in Medical Image Segmentation

Medical image segmentation can provide a reliable basis for clinical diagnosis and pathology research and assist doctors to make accurate diagnosis. The emergence of medical image segmentation based on deep learning has solved the problems of low robustness and low accuracy in traditional automatic segmentation methods, among which, U-Net stands out among many segmentation networks with its excellent performance. Researchers have successively proposed a variety of improved variants based on U-Net. Taking U-Net and its network variants as the main content, this article first introduced the network structure and common improvement methods of U-Net in detail. Then, divided the U-Net variants into general-purpose networks and specific network according to the different segmentation objects, and discussed the research progress of these networks in the medical image segmentation. At the end, the difficulties and problems existing in this research field were analyzed, and the development directions were prospected.