[1]宋 杰,刘彩霞,李慧婷.基于 U-Net 网络的医学图像分割研究综述[J].计算机技术与发展,2024,34(01):9-16.[doi:10. 3969 / j. issn. 1673-629X. 2024. 01. 002]
　SONG Jie,LIU Cai-xia,LI Hui-ting.Review of Medical Image Segmentation Based on U-Net Network[J].,2024,34(01):9-16.[doi:10. 3969 / j. issn. 1673-629X. 2024. 01. 002]

点击复制

基于 U-Net 网络的医学图像分割研究综述()

分享到：

《计算机技术与发展》[ISSN:1006-6977/CN:61-1281/TN]

卷:

34

期数:

2024年01期

页码:

9-16

栏目:

综述

出版日期:

2024-01-10

文章信息/Info

Title:

Review of Medical Image Segmentation Based on U-Net Network

文章编号:

1673-629X(2024)01-0009-08

作者:

宋 杰¹ ; 刘彩霞¹; 2 ; 李慧婷¹

1. 江苏师范大学 智慧教育学院,江苏 徐州 221116;

2. 江苏师范大学 江苏省教育信息化工程技术研究中心,江苏 徐州 221116

Author(s):

SONG Jie1 ; LIU Cai-xia1; 2 ; LI Hui-ting1

1. School of Wisdom Education,Jiangsu Normal University,Xuzhou 221116,China;

2. Jiangsu Engineering Research Center of Educational Informatization,Jiangsu Normal University,Xuzhou 221116,China

关键词:

医学图像分割; 深度学习; 人工智能; U-Net; 卷积神经网络

Keywords:

medical image segmentation; deep learning; artificial intelligence; U-Net; convolutional neural network

分类号:

TP391

DOI:

10. 3969 / j. issn. 1673-629X. 2024. 01. 002

摘要:

近年来随着深度学习技术的快速发展,卷积神经网络( CNN) 成为语义分割的重要支撑框架,被广泛运用于多种目标检测与分割的任务当中。 在医学图像分割任务中,U-Net 网络以其优异的分割性能、可拓展性的网络结构等特点成为该领域研究的热点。 如今有众多学者从网络的结构等方面对 U-Net 进行改进以优化网络性能、提升分割准确度。 研究通过对相关文献的分析,首先介绍了基于 U-Net 的经典改进模型;然后阐述了六大 U-Net 改进机制:注意力机制、inception 模块、残差结构、空洞机制、密集连接结构以及集成网络结构;随后介绍了医学图像分割常用评价指标和非结构化改进方案,这些非结构化改进方法包括数据增强、优化器、激活函数和损失函数四个方面;之后列举并分析了在肺结节、视网膜血管、皮肤病和颅内肿瘤新冠肺炎四大医学图像分割领域的改进模型;最后对 U-Net 网络的未来发展进行展望,为相关研究提供思路。

Abstract:

With the rapid development of deep learning technology in recent years,convolutional neural network ( CNN) has become animportant support framework for semantic?

segmentation and is widely used in a variety of target detection and segmentation tasks. Inmedical image segmentation tasks, U - Net network has become a hot research?

topic in this field with its excellent segmentationperformance and expandable network structure. Nowadays,many scholars have improved U-Net in terms of the structure of the networkto optimize the network performance and improve the segmentation accuracy. The study first introduces the classical improved modelbased on U-Net by analyzing the relevant literature. Then,six U - Net improvement mechanisms are described:attention mechanism,inception module,residual structure,dilated mechanism,dense connection?

structure and integrated network structure. Common evaluationmetrics and unstructured improvement schemes for medical image segmentation are then presented. These unstructured improvementmethods include four aspects of data enhancement,optimizers,activation functions,and loss functions. After that,improved models infour major medical image segmentation areas,　namely,pulmonary nodules,retinal vessels,skin diseases and intracranial tumors,are listedand analyzed. Finally,the future development of U-Net network is prospected to provide ideas for related research.

常用功能