Application of Improved U-Net in Retinal Vessel Segmentation

doi:10.3778/j.issn.1673-9418.2010061

Abstract

Abstract:

In order to solve the problems that it is difficult to accurately identify the vascular boundary and the low contrast between the blood vessel and the background in fundus retinal vascular segmentation, an encoder-decoder algorithm is proposed. In order to improve the segmentation ability of the algorithm at the vascular boundary, the global convolutional network (GCN) and boundary refinement (BR) are used to replace the traditional convolution layer in the coding part, and the improved position attention (PA) module and channel attention (CA) module are introduced in the jump connection part. The aim is to increase the contrast between the blood vessels and the background, so that the network can better separate the blood vessels from the background. In addition, in order to improve the performance of the network, the dense convolution network is used in the last layer of the coding part to solve the problem of network overfitting, and in order to solve the problem of gradient explosion and gradient disappearance to a certain extent, in each layer of the decoding part, the convolution long-short memory network is used to improve the ability of the network to obtain feature information. Tested on the common datasets DRIVE and CHASE_DB1, the sensitivity, specificity, accuracy, F1-Score and AUC are used as evaluation indicators, in which the accuracy and AUC reach 96.99%, 98.77% and 97.51%, 99.01%, respectively. This algorithm can effectively improve the accuracy of blood vessel segmentation in fundus image.

Key words: retinal vessels, U-Net, boundary refinement (BR), position attention (PA) module, channel attention (CA) module, global convolutional network (GCN)

摘要：

针对眼底视网膜血管分割中血管边界难以精确识别以及血管与背景对比度低而难以分割的问题,提出一种编码器-解码器结构的算法。为了提高算法在血管边界的分割能力,在编码部分采用全局卷积网络（GCN）和边界细化（BR）替换传统的卷积层;在跳跃连接部分引入改进的位置注意模块（PA）和通道注意模块（CA）,目的是增加血管与背景之间的对比度,使网络更好地将血管与背景分割开;此外,为了提高网络的性能,在编码部分的最后一层使用密集卷积网络解决网络过拟合问题,同时为了在一定程度上解决梯度爆炸、梯度消失的问题,在解码部分的每一层使用卷积长短记忆网络提升网络获取特征信息的能力。在公共的数据集DRIVE和CHASE_DB1中进行测试,以敏感性、特异性、准确性、F1-Score和AUC为评价指标,其中准确性和AUC分别达到了96.99%、98.77%和97.51%、99.01%。该算法能有效提高眼底图像血管分割的准确率。

关键词: 视网膜血管, U-Net, 边界细化（BR）, 位置注意模块（PA）, 通道注意模块（CA）, 全局卷积网络（GCN）

CLC Number:

TP391

GU Penghui, XIAO Zhiyong. Application of Improved U-Net in Retinal Vessel Segmentation[J]. Journal of Frontiers of Computer Science and Technology, 2022, 16(3): 683-691.

谷鹏辉, 肖志勇. 改进的U-Net在视网膜血管分割上的应用[J]. 计算机科学与探索, 2022, 16(3): 683-691.

Figures/Tables 13

References 30

[1]	KANSKI J J, BOWLING B. Clinical ophthalmology: a systematic approach[M]. New York: Elsevier Science Inc., 2011.
[2]	STRISCIUGLIO N, AZZOPARDI G, VENTO M, et al. Supervised vessel delineation in retinal fundus images with the automatic selection of B-COSFIRE filters[J]. Machine Vision and Applications, 2016, 27(8): 1137-1149. DOI URL
[3]	LUPASCU C A, TEGOLO D, TRUCCO E. FABC: retinal vessel segmentation using AdaBoost[J]. IEEE Transactions on Information Technology in Biomedicine, 2010, 14(5): 1267-1274. DOI URL
[4]	RONNEBERGER O, FISCHER P, BROX T. U-Net: convolu- tional networks for biomedical image segmentation[C]// LNCS 9351: Proceedings of the 18th International Conference on Medical Image Computing and Computer-Assisted Intervention, Munich, Oct 5-9, 2015. Cham: Springer, 2015: 234-241.
[5]	代洋洋, 王宽全. UU-Net: 基于U-Net的U形多路径网络的视网膜血管分割[J]. 哈尔滨工程大学学报, 2020, 41(5): 718-723.
	DAI Y Y, WANG K Q. UU-Net: a U-shaped network with a multi-path structure based on U-Net for vessel segmentation in retinal images[J]. Journal of Harbin Engineering Univer-sity, 2020, 41(5): 718-723.
[6]	ZHANG S H, FU H Z, YAN Y G, et al. Attention guided network for retinal image segmentation[C]// LNCS 11764: Proceedings of the 22nd International Conference on Medical Image Computing and Computer-Assisted Intervention, Shenzhen, Oct 13-17, 2019. Cham: Springer, 2019: 797-805.
[7]	LIU B, GU L, LU F. Unsupervised ensemble strategy for retinal vessel segmentation[C]// LNCS 11764: Proceedings of the 22nd International Conference on Medical Image Computing and Computer-Assisted Intervention, Shenzhen, Oct 13-17, 2019. Cham: Springer, 2019: 111-119.
[8]	LI L Z, VERMA M, NAKASHIMA Y, et al. IterNet: retinal image segmentation utilizing structural redundancy in vessel networks[C]// Proceedings of the 2020 IEEE Winter Conference on Applications of Computer Vision, Snowmass Village, Mar 1-5, 2020. Piscataway: IEEE, 2020: 3656-3665.
[9]	PENG C, ZHANG X Y, YU G, et al. Large kernel matters— improve semantic segmentation by global convolutional network[C]// Proceedings of the 2017 IEEE Conference on Computer Vision and Pattern Recognition, Honolulu, Jul 21-26, 2017. Washington: IEEE Computer Society, 2017: 1743-1751.
[10]	FU J, LIU J, TIAN H J, et al. Dual attention network for scene segmentation[C]// Proceedings of the 2019 IEEE Con-ference on Computer Vision and Pattern Recognition, Long Beach, Jun 16-20, 2019. Piscataway: IEEE, 2019: 3146-3154.
[11]	HUANG G, LIU Z, VAN DER MAATEN L, et al. Densely connected convolutional networks[C]// Proceedings of the 2017 IEEE Conference on Computer Vision and Pattern Recognition, Honolulu, Jul 8-12 21-26, 2017. Washington: IEEE Computer Society, 2017: 2261-2269.
[12]	刘辰, 肖志勇, 杜年茂. 改进的卷积神经网络在医学图像分割上的应用[J]. 计算机科学与探索, 2019, 13(9): 1593-1603.
	LIU C, XIAO Z Y, DU N M. Application of improved convolu-tional neural network in medical image segmentation[J]. Journal of Frontiers of Computer Science and Technology, 2019, 13(9): 1593-1603.
[13]	HOCHREITER S, SCHMIDHUBER J. Long short-term memory[J]. Neural Computation, 1997, 9(8): 1735-1780. DOI URL
[14]	DENG W W, LING X L, QI Y, et al. Ad click prediction in sequence with long short-term memory networks: an externality-aware model[C]// Proceedings of the 41st International ACM SIGIR Conference on Research & Development in Information Retrieval, Ann Arbor, Jul, 2018. New York: ACM, 2018: 1065-1068.
[15]	王毅, 冯小年, 钱铁云, 等. 基于CNN和LSTM深度网络的伪装用户入侵检测[J]. 计算机科学与探索, 2018, 12(4): 575-585.
	WANG Y, FENG X N, QIAN T Y, et al. CNN and LSTM deep network based intrusion detection for malicious users[J]. Journal of Frontiers of Computer Science and Technology, 2018, 12(4): 575-585.
[16]	黄偲琪, 张冬梅, 闫博. 基于LSTM与多头注意力机制的恶意域名检测算法[J]. 软件, 2019, 40(2): 83-90.
	HUANG S Q, ZHANG D M, YAN B. A detection algorithm of malicious domain based on LSTM and multi-head attention mechanism[J]. Computer Engineering & Software, 2019, 40(2): 83-90.
[17]	SHI X J, CHEN Z R, WANG H, et al. Convolutional LSTM network: a machine learning approach for precipitation nowcasting[C]// Proceedings of the Annual Conference on Neural Information Processing Systems 2015, Montreal, Dec 7-12, 2015. Red Hook: Curran Associates, 2015: 802-810.
[18]	HE K M, ZHANG X Y, REN S Q, et al. Deep residual learning for image recognition[C]// Proceedings of the 2016 IEEE Conference on Computer Vision and Pattern Recognition, Las Vegas, Jun 27-30, 2016. Washington: IEEE Computer Society, 2016: 770-778.
[19]	IOFFE S, SZEGEDY C. Batch normalization: accelerating deep network training by reducing internal covariate shift[J]. arXiv:1502.03167, 2015.
[20]	STAAL J, ABRÀMOFF 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 URL
[21]	OWEN C G, RUDNICKA A R, MULLEN R, et al. Measuring retinal vessel tortuosity in 10-year-old children: validation of the computer-assisted image analysis of the retina (CAIAR) program[J]. Investigative Ophthalmology & Visual Science, 2009, 50(5): 2004-2010.
[22]	韩铖惠, 王慧琴, 胡燕. 低对比度火焰图像增强和分割算法研究[J]. 计算机科学与探索, 2018, 12(1): 163-170.
	HAN C H, WANG H Q, HU Y. Enhancement and segmentation algorithm study for low contrast fire image[J]. Journal of Frontiers of Computer Science and Technology, 2018, 12(1): 163-170.
[23]	储清翠, 王华彬, 陶亮. 图像的局部自适应Gamma校正[J]. 计算机工程与应用, 2015, 51(7): 189-193.
	CHU Q C, WANG H B, TAO L. Local adaptive Gamma correction method[J]. Computer Engineering and Applications, 2015, 51(7): 189-193.
[24]	ALOM M Z, HASAN M, YAKOPCIC C, et al. Recurrent residual convolutional neural network based on U-Net (R2U-Net) for medical image segmentation[J]. arXiv:1802.06955, 2018.
[25]	ZHUANG J T. LadderNet: multi-path networks based on U-Net for medical image segmentation[J]. arXiv:1810.07810, 2018.
[26]	WANG B, QIU S, HE H G. Dual encoding U-Net for retinal vessel segmentation[C]// LNCS 11764: Proceedings of the 22nd International Conference on Medical Image Computing and Computer-Assisted Intervention, Shenzhen, Oct 13-17, 2019. Cham: Springer, 2019: 84-92.
[27]	吴鑫鑫, 肖志勇, 刘辰. 低尺度血管检测在视网膜血管分割中的应用[J]. 计算机科学与探索, 2020, 14(1): 171-180.
	WU X X, XIAO Z Y, LIU C. Application of low-scales vessel detection in retinal vessel segmentation[J]. Journal of Frontiers of Computer Science and Technology, 2020, 14(1): 171-180.
[28]	吕晓文, 邵枫, 熊义明, 等. 基于双流网络的视网膜血管分割方法[J]. 光学学报, 2020, 40(4): 47-55.
	LV X W, SHAO F, XIONG Y M, et al. Retinal vessel segmen-tation method based on two-stream networks[J]. Acta Optica Sinica, 2020, 40(4): 47-55.
[29]	ZHANG S H, FU H Z, XU Y W, et al. Retinal image segmentation with a structure-texture demixing network[C]// LNCS 12265: Proceedings of the 23rd International Conference on Medical Image Computing and Computer-Assisted Intervention, Lima, Oct 4-8, 2020. Cham: Springer, 2020: 765-774.
[30]	WANG W, ZHONG J F, WU H S, et al. RVSeg-Net: an efficient feature pyramid cascade network for retinal vessel segmentation[C]// LNCS 12265: Proceedings of the 23rd Inter-national Conference on Medical Image Computing and Computer-Assisted Intervention, Lima, Oct 4-8, 2020. Cham: Springer, 2020: 796-805.

Dataset	Method	Se	Ac	AUC	F1-Score
DRIVE	未预处理	81.78	96.99	98.74	82.67
	预处理	83.24	96.99	98.77	82.91
	未后处理	79.74	97.06	98.77	82.64
	后处理	83.24	96.99	98.77	82.91
CHASE_DB1	未预处理	80.12	97.48	99.00	83.48
	预处理	81.49	97.51	99.01	83.55
	未后处理	79.56	97.54	99.01	83.19
	后处理	81.49	97.51	99.01	83.55

Dataset	Method	Se	Ac	AUC	F1-Score
DRIVE	未预处理	81.78	96.99	98.74	82.67
	预处理	83.24	96.99	98.77	82.91
	未后处理	79.74	97.06	98.77	82.64
	后处理	83.24	96.99	98.77	82.91
CHASE_DB1	未预处理	80.12	97.48	99.00	83.48
	预处理	81.49	97.51	99.01	83.55
	未后处理	79.56	97.54	99.01	83.19
	后处理	81.49	97.51	99.01	83.55

Method	DRIVE					CHASE_DB1
Method	Se	Ac	AUC	F1-Score		Se	Ac	AUC	F1-Score
U-Net^[24]	75.37	95.31	97.55	81.42	82.88		95.78	97.72	77.83
GCN+BR_U-Net	82.81	96.98	98.73	82.77	81.25		97.49	98.99	83.50
GCN+BR_ConvLSTM_U-Net	83.10	96.98	98.75	82.84	81.37		97.50	98.97	83.52
GCN+BR_ConvLSTM_CA+PA_U-Net	83.24	96.99	98.77	82.91	81.49		97.51	99.01	83.55

Method	DRIVE					CHASE_DB1
Method	Se	Ac	AUC	F1-Score		Se	Ac	AUC	F1-Score
U-Net^[24]	75.37	95.31	97.55	81.42	82.88		95.78	97.72	77.83
GCN+BR_U-Net	82.81	96.98	98.73	82.77	81.25		97.49	98.99	83.50
GCN+BR_ConvLSTM_U-Net	83.10	96.98	98.75	82.84	81.37		97.50	98.97	83.52
GCN+BR_ConvLSTM_CA+PA_U-Net	83.24	96.99	98.77	82.91	81.49		97.51	99.01	83.55

Method	Year	Se/%	Ac/%	AUC/%	F1-Score/%
R2U-Net^[24]	2018	77.92	95.56	97.84	81.71
U-Net^[24]	2018	75.37	95.31	97.55	81.42
LadderNet^[25]	2018	78.56	95.61	97.93	82.02
DEU-Net^[26]	2019	79.40	95.67	97.72	82.70
AG-Net^[6]	2019	81.00	96.92	98.56	N.A
吴鑫鑫等人^[27]	2019	81.92	96.95	97.82	N.A
吕晓文等人^[28]	2020	80.62	95.47	97.39	N.A
AttR2U-Net^[5]	2020	80.28	96.89	98.41	N.A
Zhang等人^[29]	2020	81.51	96.95	98.63	N.A
RVSeg-Net^[30]	2020	81.07	96.81	98.17	N.A
Proposed	2020	83.24	96.99	98.77	82.91