基于RAU-net的视网膜血管图像分割

张丽娟; 梅畅; 李超然; 章润

基于RAU-net的视网膜血管图像分割

1.
长春工业大学计算机科学与工程学院，吉林长春 130012
2.
长春理工大学光电信息学院信息工程学院，吉林长春 130114

基金项目:

吉林省教育厅“十三五”科学研究规划项目 JJKH20210747KJ

吉林省生态环境厅科研项目吉环科字第2021-07

详细信息

作者简介:
张丽娟（1978-），女，吉林梅河口人，博士，教授，主要从事计算机视觉及光学图像处理等方面研究。E-mail: zhanglijuan@ccut.edu.cn

中图分类号: TP394.1; TH691.9
计量
- 文章访问数: 129
- HTML全文浏览量: 133
- PDF下载量: 34
- 被引次数: 0
出版历程
- 收稿日期: 2020-10-08
- 修回日期: 2020-10-14
- 刊出日期: 2021-12-20

Retinal Vessel Image Segmentation Based on RAU-net

1.
College of Computer Science and Technology, Changchun University of Technology, Changchun 130012, China
2.
College of Information Engineering, College of Optical And Electronical Information, Changchun University of Science and Technology, Changchun 130114, China

摘要

摘要: 在眼科疾病的诊断中，对视网膜血管进行分割是非常有效的一种方法。在方法使用中，经常会遇到由于视网膜血管背景对比度低及血管末梢细节复杂导致的血管分割难度较大的问题，通过在设计网络的过程中在基础U-net网络中引入残差学习，注意力机制等模块，并将两者巧妙地结合在一起，提出一种新型的基于U-net的RAU-net视网膜血管图像分割算法。首先，在网络的编码器阶段加入残差模块，解决了模型网络加深导致梯度爆炸以及梯度消失的问题。其次，在网络的解码器阶段引入注意力门（attention gate, AU）模块，用来抑制不必要的特征，从而使模型产生更高的精度。通过在DRIVE数据集上进行验证，该算法的准确率、灵敏度、特异性和F₁-score分别达到了0.7832，0.9815，0.9568和0.8192。分割效果相对于普通监督学习算法较为良好。
- 图像分割 /
- 视网膜血管 /
- 全卷积网络 /
- 残差模块 /
- 注意力机制
Abstract: In the diagnosis of ophthalmic diseases, segmentation of retinal blood vessels is a quite effective method. However, using this method, difficulties in blood vessel segmentation are often encountered due to the low contrast of the retinal blood vessel background and complex details of the blood vessel end. Thus, residual learning is introduced by adding a basic U-net network to the process of network design. Through the introduction of residual learning and attention mechanism modules into the basic U-net network in the process of network design, a new type of U-net-based RAU-net retinal blood vessel image segmentation algorithm is proposed. First, the residual module is added to the encoder stage of the network to address gradient explosion and disappearance caused by the deepening of the model network. Second, the attention gate module is introduced in the decoder stage of the network to suppress unnecessary features to ensure high accuracy of the model. Through verification of DRIVE, the accuracy, sensitivity, specificity, and F₁-score of the algorithm reached 0.7832, 0.9815, 0.9568, and 0.8192, respectively. Thus, the segmentation effect was better than that of ordinary supervised learning algorithms
- image segmentation /
- retinal vessels /
- fully convolutional network /
- residual module /
- attention mechanism

HTML全文

图 1 RAU-net视网膜血管图像分割算法模型

Figure 1. RAU-net retinal vessel image segmentation algorithm model

下载: 全尺寸图片幻灯片

图 2 残差模块的构成

Figure 2. Residual learning: a building block

下载: 全尺寸图片幻灯片

图 3 注意力机制结构图

Figure 3. Structure diagram of attention mechanism

下载: 全尺寸图片幻灯片

图 4 图像扩充后的图像块

Figure 4. Image block after image expansion

下载: 全尺寸图片幻灯片

图 5 DRIVE数据集上的分割结果

Figure 5. Segmentation results on the DRIVE dataset

下载: 全尺寸图片幻灯片

图 6 DRIVE数据集上的ROC曲线

Figure 6. ROC curve on the DRIVE data set

下载: 全尺寸图片幻灯片

表 1 基于U-Net的常见算法和本文算法在DRIVE数据集上的视网膜血管分割性能对比

Table 1. Comparison of retinal blood vessel segmentation performance between common algorithms based on U-Net and this algorithm on the DRIVE dataset

Methods	Sen	Spe	Acc	F₁-score	AUC
U-net^[10]	0.7537	0.9820	0.9531	0.8142	0.9755
Res-net^[10]	0.7726	0.9820	0.9533	0.8149	0.9779
Recurrent-net^[10]	0.7751	0.9816	0.9556	0.8155	0.9782
R2U-net^[10]	0.7792	0.9813	0.9556	0.8171	0.9784
RAU-net	0.7832	0.9815	0.9568	0.8192	0.9792

下载: 导出CSV

参考文献(19)

[1]	Abràmoff M D, Folk J C, Han D P, et al. Automated analysis of retinal images for detection of referable diabetic retinopathy[J]. JAMA Ophthalmol, 2013, 131(3): 351-357. doi: 10.1001/jamaophthalmol.2013.1743
[2]	ZENG X, CHEN H, LUO Y, et al. Automated diabetic retinopathy detection based on binocular siamese-like convolutional neural network[J]. IEEE Access, 2019, 7: 30744-30753 (doi: 10.1109/ACCESS.2019.2903171).
[3]	Chaudhuri S, Chatterjee S, Katz N, et al. Detection of blood vessels in retinal images using two-dimensional matched filters[J]. IEEE Transactions on Medical Imaging, 1989, 8(3): 263-269. doi: 10.1109/42.34715
[4]	Vlachos M, Dermatas E. Multi-scale retinal vessel segmentation using line tracking[J]. Computerized Medical Imaging & Graphics the Official Journal of the Computerized Medical Imaging Society, 2010, 34(3): 213-227. http://www.onacademic.com/detail/journal_1000034065429410_d1d5.html
[5]	Zana F, Klein J C. Segmentation of vessel-like patterns using mathematical morphology and curvature evaluation[J]. IEEE Transactions on Image Processing, 2001, 10(7): 1010-1019. doi: 10.1109/83.931095
[6]	WU C, YI B, ZHANG Y, et al. Retinal vessel image segmentation based on improved convolutional neural network[J]. Acta Optica Sinica, 2018, 38(11): 1111004-1-1111004-7. doi: 10.3788/AOS201838.1111004
[7]	Dharmawan D A, LI D, Ng B P, et al. A new hybrid algorithm for retinal vessels segmentation on fundus images[J]. IEEE Access, 2019, 7: 41885-41896 (doi: 10.1109/ACCESS.2019.2906344).
[8]	FU H, XU Y, WONG D, et al. Retinal vessel segmentation via deep learning network and fully-connected conditional random fields[C]//13th International Symposium on Biomedical Imaging of IEEE (ISBI 2016), 2016: 698-701.
[9]	JIANG Y, TAN N, PENG T, et al. Retinal vessels segmentation based on dilated multi-scale convolutional neural network[J]. IEEE Access, 2019, 7: 76342-76352. doi: 10.1109/ACCESS.2019.2922365
[10]	Alom M Z, Yakopcic C, Taha T M, et al. Nuclei segmentation with recurrent residual convolutional neural networks based U-Net (R2U-Net)[C]//IEEE National Aerospace and Electronics Conference (NAECON 2018), 2018: 228-233(doi: 10.1109/NAECON.2018.8556686).
[11]	ZHUANG J. Multi-path networks based on U-Net for medical image segmentation[J]. arXiv preprint, 2018, arXiv: 1810.07810.
[12]	HE K, ZHANG X, REN S, et al. Deep residual learning for image recognition[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2016: 770-778.
[13]	Oktay O, Schlemper J, Folgoc L L, et al. Attention U-net: Learning where to look for the pancreas[J]. arXiv preprint, 2018, arXiv: 1804.03999.
[14]	Gkantsidis C, Mihail M, Saberi A. Random walks in peer-to-peer networks: algorithms and evaluation[J]. Performance Evaluation, 2006, 63(3): 241-263. doi: 10.1016/j.peva.2005.01.002
[15]	Staal J, Abramoff M D, Niemeijer M, et al. Ridge-based vessel segmentation in color images of the retina[J]. IEEE Transactions on Medical Imaging, 2004, 23(4): 501-509. doi: 10.1109/TMI.2004.825627
[16]	Satapathy S C, Raja N S M, et al. Multi-level image thresholding using Otsu and chaotic bat algorithm[J]. Neural Computing and Applications, 2018, 29(12): 1285-1307. doi: 10.1007/s00521-016-2645-5
[17]	PAN X, ZHANG Q, ZHANG H, et al. A fundus retinal vessels segmentation scheme based on the improved deep learning U-Net model[J]. IEEE Access, 2019, 7(99): 122634-122643. http://ieeexplore.ieee.org/document/8796382
[18]	雷军明, 王成钢, 刘雪燕. 基于改进U-Net的视网膜血管图像分割[J]. 信息技术与信息化, 2019(10): 168-169. doi: 10.3969/j.issn.1672-9528.2019.10.052 LEI J M, WANG C G, LIU X Y. Retinal vessel image segmentation based on improved U-Net[J]. Information Technology and Information Technology, 2019(10): 168-169. doi: 10.3969/j.issn.1672-9528.2019.10.052
[19]	Kingma D, Ba J. Adam: A method for stochastic optimization[J]. arXiv preprint, 2014, arXiv: 1412.6980.