Abstract: U-Net and its variants have showcased exceptional performance in the domain of breast medical image segmentation. By employing a fully convolutional network (FCN) structure for semantic segmentation, the symmetrical structure of U-Net offers remarkable flexibility and adaptability. It can be tailored to diverse image segmentation tasks and challenges by adjusting network depth and incorporating new modules, leaving a significant impact on subsequent network designs. This paper aims to delve into the application of U-shaped convolutional networks in breast medical image segmentation, categorizing and summarizing U-shaped convolutional networks used for this purpose in recent years. It outlines the widely used breast medical image datasets and evaluation metrics, discusses common data augmentation techniques, and provides a detailed introduction to the network structure of the U-Net model along with traditional segmentation methods for breast medical images. Furthermore, it summarizes the improvements made to the U-Net network structure for breast medical image segmentation, including modifications like residual structures, multi-scale features, dilation mechanisms, attention mechanisms, skip connection mechanisms, and integration with Transformers. Finally, it addresses the current challenges and problems encountered in breast medical image segmentation and offers insights into future research directions.

Key words: medical image segmentation, U-shaped convolutional network, deep learning, breast disease, image processing

摘要： U-Net及其变体模型在乳腺医学图像分割领域展现了卓越的性能，U-Net采用全卷积网络（FCN）结构进行语义分割，U-Net对称结构的高度灵活性和适应性可以通过调整网络深度、引入新的模块来适应不同的图像分割任务和挑战，这种创新结构对后续网络设计产生了深远影响。深入探讨了基于U型卷积网络在乳腺医学图像分割中的应用，并对近年来用于乳腺医学图像分割的U型卷积网络进行了分类与归纳。针对U-Net网络结构改进的乳腺医学图像分割技术进行了如下总结。阐述了目前广泛使用的乳腺医学图像数据集及评价指标，陈述了常用的数据增强方法；详细介绍了U-Net模型的网络结构以及用于乳腺医学图像的传统分割方法；对用于乳腺医学图像分割方法的U型网络结构按照残差结构、多尺度特征、膨胀机制、注意力机制、跳跃连接机制、结合Transformer等方面改进进行归纳总结。讨论了当下乳腺医学图像分割所遇到的问题与挑战，对未来的研究走向做出了展望。

关键词: 医学图像分割, U型卷积网络, 深度学习, 乳腺疾病, 图像处理