改进的U-Net网络用于视网膜血管分割

doi:10.3778/j.issn.1673-9418.2107060

摘要/Abstract

摘要： 视网膜血管的形态结构信息可以为糖尿病、高血压等疾病提供诊断依据。现有的大部分算法通过融合血管不同阶段的特征信息来提升分割精度，却忽略了融合后浅层信息在采样过程中容易被稀释，且上采样次数越多，稀释越严重的问题。为此提出了一种融合概率分布注意力的提升算法用以解决上述问题。该算法基于U-Net模型，首先利用多个串行提升块学习血管浅层的特征信息，然后在提升块中引入概率分布注意力机制用来改变浅层信息的概率分布，提升浅层信息在最终分割任务中的贡献度，最后利用深度监督非对称聚合补充各提升块的梯度流，以避免模型在训练过程中产生梯度消失的问题。在DRIVE、STARE数据集上与近年来不同算法进行了对比，其中SP值分别为0.985 7、0.991 7，相较于第二名提升了0.2%、0.25%，均优于目前最新的算法结果。实验证明了提出的提升算法能较好地改善血管前期信息，一定程度上解决了信息稀释问题，提高了对视网膜血管的分割效果。

关键词: 视网膜血管, 提升算法, 概率分布注意力, 深监督非对称聚合

Abstract: The morphological and structural information of retinal blood vessels can provide a basis for diagnosis of diabetes and hypertension. Most of the existing algorithms improve the segmentation accuracy by fusing the feature information of different stages of blood vessels, but they ignore the problem that the shallow information is diluted during the upsampling process after fusing, and the more times of upsampling, the more serious the dilution is. A boosting algorithm that incorporates probabilistic distributions of attention is proposed to solve the above problem. Based on the U-Net model, this paper firstly uses multiple serial boosting blocks to learn the shallow feature information of the vessels. Then, it introduces a probability distribution attention mechanism in the boosting blocks to change the probability distribution and improve the contribution of the shallow information in the final segmen-tation task. Finally, it uses deeply supervised asymmetric aggregation to supplement the gradient flow of each boosting block to avoid gradient disappearance. It is compared with different algorithms in recent years on the DRIVE and STARE datasets. The SP values of proposed algorithm are 0.9857 and 0.9917 respectively, which are 0.2% and 0.25% higher than the second one, both better than the latest algorithm results. The experiments demonstrate that the boosting algorithm can better improve the pre-vascular information, solve the information dilution problem, and improve the segmentation of retinal vessels.

Key words: retinal vessels, boosting algorithm, probability distribution attention, deeply supervised asymmetric aggregation

吕佳, 马超, 程超. 改进的U-Net网络用于视网膜血管分割[J]. 计算机科学与探索, 2023, 17(3): 657-666.

LYU Jia, MA Chao, CHENG Chao. Improved U-Net Network for Retinal Vascular Segmentation[J]. Journal of Frontiers of Computer Science and Technology, 2023, 17(3): 657-666.

参考文献

[1] 王娟, 赵建勇, 童龙. 老年2型糖尿病患者并发周围神经病变的影响因素分析[J]. 中国慢性病预防与控制, 2019, 27(1): 52-54.
WANG J, ZHAO J Y, TONG L. Analysis of influencing factors of peripheral neuropathy in elderly patients with type 2 diabetes[J]. Chinese Prevention and Control of Chronic Diseases, 2019, 27(1): 52-54.
[2] 赖小波, 许茂盛, 徐小媚. 基于多模型融合和区域迭代生长的视网膜血管自动分割[J]. 电子学报, 2019, 47(12): 2611-2621.
LAI X B, XU M S, XU X M. Automatic segmentation of retinal blood vessels based on multi-model fusion and regio-nal iterative growth[J]. Chinese Journal of Electronics, 2019, 47(12): 2611-2621.
[3] SAROJ S K, KUMAR R, SINGH N P. Fréchet PDF based matched filter approach for retinal blood vessels segmenta-tion[J]. Computer Methods and Programs in Biomedicine, 2020, 194: 105490.
[4] KARN P K, BISWAL B, SAMANTARAY S R. Robust retinal blood vessel segmentation using hybrid active contour model[J]. IET Image Processing, 2019, 13(3): 440-450.
[5] 朱承璋, 崔锦恺, 邹北骥, 等. 基于多特征融合和随机森林的视网膜血管分割[J]. 计算机辅助设计与图形学学报, 2017, 29(4): 584-592.
ZHU C Z, CUI J K, ZOU B J, et al. Retinal vessel segmen-tation based on multi-feature fusion and random forest[J]. Journal of Computer Aided Design and Graphics, 2017, 29(4): 584-592.
[6] WU H S, WANG W, ZHONG J F, et al. SCS-Net: a scale and context sensitive network for retinal vessel segmenta-tion[J]. Medical Image Analysis, 2021, 70: 102025.
[7] LONG J, SHELHAMER E, DARRELL T. Fully convolutional networks for semantic segmentation[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2015, 39(4): 640-651.
[8] RONNEBERGE O, FISCHER P, BROX T, et al. U-Net: convolutional networks for biomedical image segmentation[C]//LNCS 9351: Proceedings of the 18th International Con-ference on Medical Image Computing and Computer Assis-ted Interventions, Munich, Oct 5-9, 2015. Cham: Springer, 2015: 234-241.
[9] HUANG G, LIU Z, LAURENS V D M, et al. Densely con-nected convolutional networks[C]//Proceedings of the 2017 IEEE Conference on Computer Vision and Pattern Recogni-tion, Honolulu, Jul 21-26, 2017. Washington: IEEE Com-puter Society, 2017: 2261-2269.
[10] JéGOU S, DROZDZAL M, VAZQUEZ D, et al. The one hundred layers tiramisu: fully convolutional densenets for semantic segmentation[C]//Proceedings of the 2017 IEEE Conference on Computer Vision and Pattern Recognition, Honolulu, Jul 21-26, 2017. Washington: IEEE Computer Society, 2017: 11-19.
[11] LIANG L, LAN Z, SHENG X, et al. Segmentation of retinal vessels by fusing contour information and conditional gene-rative adversarial[J]. Journal of Biomedical Engineering, 2021, 38(2): 276-285.
[12] HE J, JIANG D. Fundus image segmentation based on im-proved generative adversarial network for retinal vessel an-alysis[C]//Proceedings of the 3rd International Conference on Artificial Intelligence and Big Data, Chengdu, May 28-31, 2020. Washington: IEEE Computer Society, 2017: 28-31.
[13] KAMRAN S A, HOSSAIN K F, TAVAKKOLI A, et al. RV-GAN: retinal vessel segmentation from fundus images using multi-scale generative adversarial networks[J]. arXiv:2101.00535, 2021.
[14] ZHOU Y K, CHEN Z L, SHEN H L, et al. A refined equili-brium generative adversarial network for retinal vessel seg-mentation[J]. Neurocomputing, 2021, 437: 118-130.
[15] MOU L, CHEN L, CHENG J, et al. Dense dilated network with probability regularized walk for vessel detection[J]. IEEE Transactions on Medical Imaging, 2019, 39(5): 1392-1403.
[16] SAMUEL P M, VEERAMALAI T. VSSC Net: vessel specific skip chain convolutional network for blood vessel segmen-tation[J]. Computer Methods and Programs in Biomedicine, 2021, 198: 105769.
[17] KHAN T M, NAQVI S S, ARSALAN M，et al. Exploiting residual edge information in deep fully convolutional neural networks for retinal vessel segmentation[C]//Proceedings of the 2020 IEEE Conference on International Joint Conference on Neural Networks, Glasgow, Jul 19-24, 2020. Piscataway:IEEE, 2020: 1-8.
[18] 宋姝洁, 崔振超, 陈丽萍, 等. 多特征融合神经网络的眼底血管分割算法[J]. 计算机科学与探索, 2021, 15(12): 2401-2412.
SONG S J, CUI Z C, CHEN L P, et al. Fundus vessel seg-mentation algorithm based on multi-feature fusion neural network[J]. Journal of Frontiers of Computer Science and Technology, 2021, 15(12): 2401-2412.
[19] ZHUANG J. LadderNet: multi-path networks based on U-Net for medical image segmentation[J]. arXiv:1810.07810, 2018.
[20] WANG B, QIU S, HE H G. Dual encoding U-net for retinal vessel segmentation[C]//LNCS 11764: Proceedings of the 2019 International Conference on Medical Image Computing and Computer Assisted Intervention, Shenzhen, Oct 13-17， 2019. Cham: Springer, 2019: 84-92.
[21] FANG L L, ZHANG L R, YAO Y B. Retina blood vessels segmentation based on the combination of the supervised and unsupervised methods[J]. Multidimensional Systems and Signal Processing, 2021, 32(4): 1123-1139.
[22] DING X, PENG Y X, SHEN C M, et al. CAB U-Net: an end-to-end category attention boosting algorithm for seg-mentation[J]. Computerized Medical Imaging and Graphics, 2020, 84(5): 101764.
[23] JIN Q, MENG Z, PHAM T D, et al. DUNet: a deformable network for retinal vessel segmentation[J]. Knowledge-Based Systems, 2019, 178: 149-162.
[24] YANG D, LIU G R, REN M C, et al. A multi-scale feature fusion method based on U-net for retinal vessel segmenta-tion[J]. Entropy, 2020, 22(8): 811.
[25] ATLI ?, GEDIK O S. Sine-Net: a fully convolutional deep learning architecture for retinal blood vessel segmentation[J]. Engineering Science and Technology, 2021, 24(2): 271-283.
[26] SU X, BAI M. Stochastic gradient boosting frequency-severity model of insurance claims[J]. PLoS One, 2020, 15(8): e0238000.
[27] DAI X, LEI Y, ZHANG Y, et al. Automatic multi-catheter detection using deeply supervised convolutional neural net-work in MRI-guided HDR prostate brachytherapy[J]. Medi-cal Physics, 2020, 47(9): 4115-4124.
[28] QU Z, CAO C, LIU L, et al. A deeply supervised convolu-tional neural network for pavement crack detection with multiscale feature fusion[J]. IEEE Transactions on Neural Networks and Learning Systems, 2022, 33(9): 4890-4899.
[29] LUO Y, LU J H, JIANG X L, et al. Learning from archi-tectural redundancy: enhanced deep supervision in deep multi-path encoder-decoder networks[J]. IEEE Transactions on Neural Networks and Learning Systems, 2022, 33(9): 4271-4284.
[30] LIU Y, CHENG M M, ZHANG X Y, et al. DNA: deeply supervised nonlinear aggregation for salient object detec-tion[J]. IEEE Transactions on Cybernetics, 2022, 52(7): 6131-6142.
[31] ZHANG Z L, ZHANG X Y, PENG C, et al. ExFuse: en-hancing feature fusion for semantic segmentation[C]//LNCS 11214: Proceedings of the 15th IEEE European Conference on Computer Vision, Munich, Sep 8-14, 2018. Cham: Springer, 2018: 273-288.
[32] 顾佳, 方志军, 田方正. 全局特征及多层次特征聚合的冠脉分割算法[J]. 计算机科学与探索, 2021, 15(5): 958-970.
GU J, FANG Z J, TIAN F Z. Global feature and multi-level feature aggregation segmentation algorithm for coronary[J]. Journal of Frontiers of Computer Science and Technology, 2021, 15(5): 958-970.
[33] 赵小强, 徐慧萍. 分级特征融合的图像语义分割[J]. 计算机科学与探索, 2021, 15(5): 949-957.
ZHAO X Q, XU H P. Image semantic segmentation method with hierarchical feature fusion[J]. Journal of Frontiers of Computer Science and Technology, 2021, 15(5): 949-957.
[34] YANG J Z, LOU C H, FU J, et al. Vessel segmentation using multiscale vessel enhancement and a region based level set model[J]. Computerized Medical Imaging and Graphics, 2020, 85: 101783.
[35] MILLETARI F, NAVAB N, AHMADI S A. V-Net: fully con-volutional neural networks for volumetric medical image segmentation[C]//Proceedings of the 4th International Con-ference on 3D Vision, Stanford, Oct 25-28, 2016. Washing-ton: IEEE Computer Society, 2016: 565-571.