Progressive Multiscale Attention Network with Dynamic Loss for Retinal Vessel Segmentation

doi:10.3778/j.issn.1002-8331.2308-0035

Abstract

Abstract: There are still many challenges in retinal vessel segmentation, such as complex biological backgrounds, small and blurry capillaries, and difficulty in fully utilizing features in fundus images. In response to these issues, this paper proposes a network that integrates dynamic loss and progressive multi-scale attention. The progressive feature extraction strategy can gradually capture features that are beneficial for segmentation in the network, while retaining more details. The designed multi-scale channel attention module can obtain complex channel dependencies and suppress background noise during skip connections, achieving the goal of highlighting important features. Finally, it is proposed that the dynamic loss uses the weights of each loss function in the deep supervision through adaptive adjustment to optimize the training strategy.The proposed method has been fully validated in two publicly available datasets, DRIVE and CHASEDB1, with sensitivity reaching 0.838?9 and 0.846?8, and accuracy reaching 0.971?5 and 0.974?5, respectively, demonstrating good segmentation performance.

Key words: retinal vessel segmentation, progressive feature extraction, multi-scale channel attention, dynamic loss

摘要： 视网膜血管分割任务仍然存在许多挑战，例如眼底图像中生物学背景复杂、毛细血管细小且模糊以及特征难以得到充分利用。针对这些问题，提出了一种融合动态损失与渐进多尺度注意力的网络。其中渐进式特征提取策略可以在网络中逐步捕获对分割有益的特征，并且保留更多细节。设计的多尺度通道注意力模块能够获得复杂的通道依赖关系并且抑制跳跃连接过程中的背景噪音，达到突出重要特征的目的。最后提出动态损失用来自适应调整深监督中各损失函数的权重，优化训练策略。所提出的方法在两个公开的数据集DRIVE和CHASE_DB1中进行了充分的验证，其中灵敏度分别达到0.838?9和0.846?8，准确率分别达到0.971?5和0.974?5，展现出了较好的分割性能。

关键词: 视网膜血管分割, 渐进式特征提取, 多尺度通道注意力, 动态损失

LI Zongmin, CHU Tianzhi, YANG Chaozhi, LIU Yujie. Progressive Multiscale Attention Network with Dynamic Loss for Retinal Vessel Segmentation[J]. Computer Engineering and Applications, 2024, 60(23): 209-218.

李宗民, 初天志, 杨超智, 刘玉杰. 融合动态损失与渐进多尺度注意力的视网膜血管分割网络[J]. 计算机工程与应用, 2024, 60(23): 209-218.

References

[1] CHALLOOB M, GAO Y, BUSCH A, et al. Separable paravector orientation tensors for enhancing retinal vessels[J]. IEEE Transactions on Medical Imaging, 2022, 42(3): 880-893.
[2] 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.
[3] LONG J, SHELHAMER E, DARRELL T. Fully convolutional networks for semantic segmentation[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2017, 39(4): 640-651.
[4] RONNEBERGER O, FISCHER P, BROX T. U-Net: convolutional networks for biomedical image segmentation[C]//Proceedings of the International Conference on Medical Image Computing and Computer-Assisted Intervention, 2015: 234-241.
[5] AZAD R, AGHDAM E K, RAULAND A, et al. Medical image segmentation review: the success of U-Net[J]. arXiv:2211.14830, 2022.
[6] WANG B, QIU S, HE H. Dual encoding U-Net for retinal vessel segmentation[C]//Proceedings of the International Conference on Medical Image Computing and Computer-Assisted Intervention, 2019: 84-92.
[7] LIAN S, LI L, LIAN G, et al. A global and local enhanced residual U-Net for accurate retinal vessel segmentation[J]. IEEE/ACM Transactions on Computational Biology and Bioinformatics, 2019, 18(3): 852-862.
[8] 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.
[9] LI K, QI X, LUO Y, et al. Accurate retinal vessel segmentation in color fundus images via fully attention-based networks[J]. IEEE Journal of Biomedical and Health Informatics, 2021, 25(6): 2071-2081.
[10] LI X, WANG W, HU X, et al. Selective kernel networks[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2019: 510-519.
[11] LI L, VERMA M, NAKASHIMA Y, et al. IterNet: retinal image segmentation utilizing structural redundancy in vessel networks[C]//Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision, 2020: 3656-3665.
[12] WANG K, ZHANG X, HUANG S, et al. CTF-Net: retinal vessel segmentation via deep coarse-to-fine supervision network[C]//Proceedings of the 2020 IEEE 17th International Symposium on Biomedical Imaging, 2020: 1237-1241.
[13] WU Y, XIA Y, SONG Y, et al. NFN+: a novel network followed network for retinal vessel segmentation[J]. Neural Networks, 2020, 126: 153-162.
[14] GU P, ZHENG H, ZHANG Y, et al. kCBAC-Net: deeply supervised complete bipartite networks with asymmetric convolutions for medical image segmentation[C]//Proceedings of the International Conference on Medical Image Computing and Computer-Assisted Intervention, 2021: 337-347.
[15] DING X, GUO Y, DING G, et al. ACNet: strengthening the kernel skeletons for powerful CNN via asymmetric convolution blocks[C]//Proceedings of the IEEE/CVF International Conference on Computer Vision, 2019: 1911-1920.
[16] WU H, WANG W, ZHONG J, et al. SCS-Net: a scale and context sensitive network for retinal vessel segmentation[J]. Medical Image Analysis, 2021, 70: 102025.
[17] FU H, CHENG J, XU Y, et al. Joint optic disc and cup segmentation based on multi-label deep network and polar transformation[J]. IEEE Transactions on Medical Imaging, 2018, 37(7): 1597-1605.
[18] YUAN Y, ZHANG L, WANG L, et al. Multi-level attention network for retinal vessel segmentation[J]. IEEE Journal of Biomedical and Health Informatics, 2022, 26(1): 312-323.
[19] TAN Y, YANG K F, ZHAO S X, et al. Retinal vessel segmentation with skeletal prior and contrastive loss[J]. IEEE Transactions on Medical Imaging, 2022, 41(9): 2238-2251.
[20] 王银宇，孟凡云，王金鹤，等. 改进ASPP及多层次特征语义融合分割方法[J]. 计算机工程与应用, 2023, 59(13):220-228.
WANG Y Y, MENG F Y, WANG J H, et al. Improved ASPP and multilevel feature semantic fusion segmentation method[J]. Computer Engineering and Applications, 2023, 59(13):220-228.
[21] WANG W, ZHONG J, WU H, et al. RVSeg-Net: an efficient feature pyramid cascade network for retinal vessel segmentation[C]//Proceedings of the International Conference on Medical Image Computing and Computer-Assisted Intervention, 2020: 796-805
[22] FENG S, ZHAO H, SHI F, et al. CPFNet: context pyramid fusion network for medical image segmentation[J]. IEEE Transactions on Medical Imaging, 2020, 39(10): 3008-3018.
[23] 梁礼明, 陈鑫, 余洁, 等. 多尺度注意力细化视网膜分割算法[J]. 计算机工程与应用, 2023, 59(6): 212-220.
LIANG L M, CHEN X, YU J, et al. Multi-scale attention refinement retinal segmentation algorithm[J]. Computer Engineering and Applications, 2023, 59(6):212-220.
[24] YE Y, PAN C, WU Y, et al. MFI-Net: multiscale feature interaction network for retinal vessel segmentation[J]. IEEE Journal of Biomedical and Health Informatics, 2022, 26(9): 4551-4562.
[25] WANG C, XU R, XU S, et al. DA-Net: dual branch transformer and adaptive strip upsampling for retinal vessels segmentation[C]//Proceedings of the International Conference on Medical Image Computing and Computer-Assisted Intervention, 2022: 528-538.
[26] QU Z, ZHUO L, CAO J, et al. TP-Net: two-path network for retinal vessel segmentation[J]. IEEE Journal of Biomedical and Health Informatics, 2023, 27(4): 1979-1990.
[27] HUANG G, LIU Z, VAN D M L, et al. Densely connected convolutional networks[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2017: 4700-4708.
[28] HU J, SHEN L, SUN G. Squeeze-and-excitation networks[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2018: 7132-7141.
[29] WANG Q L, WU B G, ZHU P F, et al. ECA-Net: efficient channel attention for deep convolutional neural networks [C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2020: 11534-11542.
[30] QIN X, DAI H, HU X, et al. Highly accurate dichotomous image segmentation[C]//Proceedings of the European Conference on Computer Vision, 2022: 38-56.
[31] LIU Y, SHEN J, YANG L, et al. ResDO-UNet: a deep residual network for accurate retinal vessel segmentation from fundus images[J]. Biomedical Signal Processing and Control, 2023, 79: 104087.
[32] KANJEE Z, CROWE B, RODMAN A. Accuracy of a generative artificial intelligence model in a complex diagnostic challenge[J]. JAMA:the Journal of the American Medical Association, 2023, 330(1): 78-80.