动态图谱融合Transformer视网膜血管分割算法

doi:10.19665/j.issn1001-2400.20241204

摘要/Abstract

摘要：

针对现有算法编码端血管特征损失过多、病灶区域血管分割能力较差和全局上下文信息提取不足等问题,提出一种动态特征加权与图谱融合Transformer视网膜血管分割算法。首先设计自适应加权编码端,缓解连续卷积及下采样造成的血管损失问题,增强血管纹理特征;其次构建图谱融合Transformer模块,旨在同时提取血管像素级特征和节点之间的关系,有效捕获图像数据中的全局和局部信息;最后构建动态特征增强模块于解码端和底部层,有效提升病灶区域血管分割能力。在DRIVE、CHASE-DB1和STARE数据集上的实验结果表明,所提出算法在仅有0.91M的模型参数下展现出优越分割性能及泛化能力,准确率分别为97.01%、97.37%和97.42%;灵敏度分别为82.51%、84.47%和81.21%;AUC-ROC分别为98.74%、98.83%和98.94%,对临床眼科疾病的诊断具有一定应用价值。

关键词: 视网膜血管分割, 图卷积, Transformer, 自适应加权下采样, 动态特征增强

Abstract:

To address the issues of excessive loss of vascular features at the encoding end,poor segmentation ability of vascular regions in lesion areas,and insufficient extraction of global contextual information in existing algorithms,this paper proposes a dynamic feature weighting and graph reasoning Transformer retinal vessel segmentation algorithm.First,an adaptive weighting encoding end is designed to alleviate the problem of vascular loss caused by continuous convolution and downsampling,thus enhancing vascular texture features.Second,a graph reasoning Transformer module is constructed to simultaneously extract pixel-level vascular features and relationships between nodes,thereby effectively capturing global and local information in image data.The final construction of the dynamic feature enhancement module at the decoder side and the encoder-decoder base effectively improves the ability to segment vascular lesions.Experimental results on DRIVE,CHASE-DB1 and STARE datasets show that the proposed algorithm exhibits a superior segmentation performance and generalization ability with only 0.91M model parameters,with accuracies of 97.01%,97.37%,and 97.42%,sensitivities of 82.51%,84.47%,and 81.21%,and AUC-ROC of 98.74%,98.83%,and 98.94%,respectively,showing a certain clinical application value in the diagnosis of ophthalmic diseases.

Key words: retinal vessel segmentation, graph convolution, Transformer, adaptive weighted downsampling, dynamic feature enhancement

中图分类号:

TP391

梁礼明, 卢宝贺, 龙鹏威, 金家新, 吴健. 动态图谱融合Transformer视网膜血管分割算法[J]. 西安电子科技大学学报, 2025, 52(2): 85-100.

LIANG Liming, LU Baohe, LONG Pengwei, JIN Jiaxin, WU Jian. Dynamic graph reasoning transformer retinal vessel segmentation algorithm[J]. Journal of Xidian University, 2025, 52(2): 85-100.

图/表 26

图1

图2

图3

图4

图5

图6

图7

表1

表2

表3

表4

表5

表6

图8

图9

表7

表8

图10

表9

表10

表11

表12

表13

表14

表15

图11

参考文献 33

[1]	马思珂, 赵萌, 石凡, 等. 注意力驱动的细胞团簇细胞核分割算法[J]. 西安电子科技大学学报, 2022, 49(2):198-206.
	MA Sike, ZHAO Meng, SHI Fan, et al. Attention Driven Nuclei Segmentation Method for Cell Clusters[J]. Journal of Xidian University, 2022, 49(2):198-206.
[2]	刘博翀, 蔡怀宇, 杨诗远, 等. 一种用于自动驾驶场景的轻量级语义分割网络[J]. 西安电子科技大学学报, 2023, 50(1):118-128.
	LIU Bochong, CAI Huaiyu, YANG Shiyuan, et al. Lightweight Semantic Segmentation Network for Autonomous Driving Scenarios[J]. Journal of Xidian University, 2023, 50(1):118-128.
[3]	YIN Y, MOULOUD A, BOURENNANE S. Retinal Vessel Segmentation Using a Probabilistic Tracking Method[J]. Pattern Recognition, 2012, 45(4):1235-1244.
[4]	AKRAM M U, KHAN S A. Multilayered Thresholding-Based Blood Vessel Segmentation for Screening of Diabetic Retinopathy[J]. Engineering with Computers, 2013, 29:165-173.
[5]	GWETU M V, TAPAMO J R, VIRIRI S, et al. Segmentation of Retinal Blood Vessels Using Normalized Gabor Filters and Automatic Thresholding[J]. South African Computer Journal, 2014, 55(1):12-24.
[6]	TUBA E, MRKELA L, TUBA M. Retinal Blood Vessel Segmentation by Support Vector Machine Classification[C]//Proceedings of the 27th International Conference Radioelektronika(RADIOELEKTRONIKA). Piscataway:IEEE, 2017:1-6.
[7]	FAN Z, RONG Y, LU J, et al. Automated Blood Vessel Segmentation in Fundus Image Based on Integral Channel Features and Random Forests[C]//Proceedings of the 12th World Congress on Intelligent Control and Automation(WCICA). Piscataway:IEEE, 2016:2063-2068.
[8]	ASLANI S, SARNEL H. A New Supervised Retinal Vessel Segmentation Method Based on Robust Hybrid Features[J]. Biomedical Signal Processing and Control, 2016, 30:1-12.
[9]	RONNEBERGER O, FISCHER P, BROX T. U-Net:Convolutional Networks for Biomedical Image Segmentation[C]//International Conference on Medical Image Computing and Computer-Assisted Intervention. Berlin:Springer, 2015:234-241.
[10]	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.
[11]	LI J, GAO G, YANG L, et al. GDF-Net:A Multi-Task Symmetrical Network for Retinal Vessel Segmentation[J]. Biomedical Signal Processing and Control, 2023, 81:104426.
[12]	CHEN J, LU Y, YU Q, et al. TransUNet:Transformers Make Strong Encoders for Medical Image Segmentation(2021)[J/OL].[2021-02-08]. https://arxiv.org/abs/2102.04306?context=cs.
[13]	SHI Z, LI Y, ZOU H, et al. Tcu-Net:Transformer Embedded in Convolutional U-Shaped Network for Retinal Vessel Segmentation[J]. Sensors, 2023, 23(10):4897.
[14]	CHEN B, LIU Y, ZHANG Z, et al. TransAttUNet:Multi-Level Attention-Guided U-Net with Transformer for Medical Image Segmentation(2021)[J/OL].[2021-07-12]. https://arxiv.org/abs/2107.05274.
[15]	VALANARASU J M J, OZA P, HACIHALILOGLU I, et al. Medical Transformer:Gated Axial-Attention for Medical Image Segmentation[C]//Proceedings of the International Conference on Medical Image Computing and Computer-Assisted Intervention. Berlin:Springer, 2021:36-46.
[16]	ZHANG Y, CHUNG A C. GUNet:A GCN-CNN Hybrid Model for Retinal Vessel Segmentation by Learning Graphical Structures[C]//Proceedings of the International Workshop on Ophthalmic Medical Image Analysis. Berlin:Springer, 2022:33-42.
[17]	JALALI Y, FATEH M, REZVANI M. VGA-1Net:Vessel Graph Based Attentional U-Net for Retinal Vessel Segmentation[J]. IET Image Processing, 2024, 18(8):2191-2213.
[18]	LI Q, HAN Z, WU X. Deeper Insights into Graph Convolutional Networks for Semi-Supervised Learning[C]// Proceedings of the AAAI Conference on Artificial Intelligence. Palo Alto: AAAI, 2018:1-8.
[19]	CHEN Z, ZHANG Y, GU J, et al. Dual Aggregation Transformer for Image Super-Resolution[C]//Proceedings of the IEEE/CVF International Conference on Computer Vision. Piscataway:IEEE, 2023:12312-12321.
[20]	CHEN L, FU Y, WEI K, et al. Instance Segmentation in the Dark[J]. International Journal of Computer Vision, 2023, 131(8):2198-2218.
[21]	BROWATZKI B, LIES J P, WALLRAVEN C. The Unreasonable Effectiveness of Encoder-Decoder Networks for Retinal Vessel Segmentation(2020)[J/OL].[2020-11-25]. https://arxiv.org/abs/2011.12643.
[22]	LIU W, YANG H, TIAN T, et al. Full-Resolution Network and Dual-Threshold Iteration for Retinal Vessel and Coronary Angiograph Segmentation[J]. IEEE J Biomed Health Inform, 2022, 26(9):4623-4634.
[23]	LI Y, ZHANG Y, LIU J Y, et al. Global Transformer and Dual Local Attention Network via Deep-Shallow Hierarchical Feature Fusion for Retinal Vessel Segmentation[J]. IEEE Transactions on Cybernetics, 2023, 53(9):5826-5839.
[24]	梁礼明, 阳渊, 朱晨锟, 等. 融合Mobile Vit和倒置门控编解码的视网膜血管分割算法(2023)[J/OL]. 北京航空航天大学学报,1-14[2024-06-18].https://doi.org/10.13700/j.bh.1001-5965.2023.0088.
	LIAN Liming, YANG Yuan, ZHU Chenkun, et al. Retinal Vessel Segmentation Algorithm Incorporating Mobile Vit and Inverted Gated Codec(2023)[J/OL]. Journal of Beijing University of Aeronautics and Astronautics,1-14[2024-06-18].https://doi.org/10.13700/j.bh.1001-5965.2023.0088.
[25]	YANG D, ZHAO H, YU K, et al. NAUNet:Lightweight Retinal Vessel Segmentation Network with Nested Connections and Efficient Attention[J]. Multimedia Tools and Applications, 2023, 82(16):25357-25379.
[26]	KUMAR A, AGRAWAL R K, JOSEPH L. IterMiUnet:A Lightweight Architecture for Automatic Blood Vessel Segmentation[J]. Multimedia Tools and Applications, 2023, 82:43207-43231.
[27]	ZHU W, LIU P X. RFENet:Recurrent Feature Extraction Module with Attention for Retinal Vessel Segmentation(2023)[J/OL].[2023-06-14]. https://assets-eu.researchsquare.com/files/rs-3052974/v1_covered_24271a4b-5c9f-4351-8cb8-8deeea250b6f.pdf?c=1686737504.
[28]	GUO S. DPN:Detail-Preserving Network with High Resolution Representation for Efficient Segmentation of Retinal Vessels[J]. Journal of Ambient Intelligence and Humanized Computing, 2023, 14(5):5689-5702.
[29]	SUN Z, WANG H, XIE Q, et al. FRS-Nets:Fourier Parameterized Rotation and Scale Equivariant Networks for Retinal Vessel Segmentation(2023)[J/OL].[2023-09-20]. https://arxiv.org/abs/2309.15638.
[30]	ZHANG Y, HE M, CHEN Z, et al. Bridge-Net:Context-Involved U-Net with Patch-Based Loss Weight Mapping for Retinal Blood Vessel Segmentation[J]. Expert Systems with Applications, 2022,195:116526.
[31]	LI J, GAO G, YANG L, et al. DEF-Net:A Dual-Encoder Fusion Network for Fundus Retinal Vessel Segmentation[J]. Electronics, 2022, 11(22):3810.
[32]	LIU Y, SHEN J, YANG L, et al. Wave-Net:A Lightweight Deep Network for Retinal Vessel Segmentation from Fundus Images[J]. Computers in Biology and Medicine, 2023, 152:106341.
[33]	HUANG Y, DENG T. STSANet:Retinal Vessel Segmentation via Spatial-Temporal and Self-Attention Encoding[C]//Proceedings of the 14th International Conference on Wireless Communications and Signal Processing(WCSP). Piscataway:IEEE, 2022:132-137.

数据集	方法	准确率	灵敏度	特异性	F1	AUC
	U-Net^[9]	96.97	79.22	98.74	82.40	98.68
	Vlight^[21]	96.73	73.81	98.93	79.82	98.28
DRIVE	FR U-Net^[22]	97.01	79.62	98.68	82.35	98.69
	GT-DLA-dsHFF^[23]	97.10	78.96	98.84	82.68	98.71
	FMVG-Net^[24]	97.08	79.98	98.72	82.76	98.65
	DGRT-Net(文中)	97.01	82.51	98.43	82.86	98.74

数据集	方法	准确率	灵敏度	特异性	F1	AUC
	U-Net^[9]	97.32	79.32	98.72	80.00	98.74
	Vlight^[21]	96.37	75.76	98.83	78.46	98.53
CHASE-DB1	FR U-Net^[22]	97.43	79.57	98.63	79.62	98.67
	GT-DLA-dsHFF^[23]	97.25	83.59	98.17	79.32	98.71
	FMVG-Net^[24]	97.49	79.50	98.70	80.01	98.75
	DGRT-Net(文中)	97.37	84.47	98.25	80.22	98.83

数据集	方法	准确率	灵敏度	特异性	F1	AUC
	U-Net^[9]	97.42	79.40	98.98	82.67	98.92
	Vlight^[21]	97.19	76.07	98.93	80.49	98.71
STARE	FR U-Net^[22]	97.45	80.98	98.77	82.72	98.99
	GT-DLA-dsHFF^[23]	97.50	79.30	98.99	82.81	99.01
	FMVG-Net^[24]	97.45	79.86	98.89	82.66	98.99
	DGRT-Net(文中)	97.42	81.21	98.79	82.89	98.94

数据集	方法	准确率	灵敏度	特异性	F1	AUC-ROC
	NAUNet^[25]	96.71	79.99	98.31		98.31
	IterMiUnet^[26]	95.68	80.53	97.89	82.62	98.10
	RFENet^[27]	95.76	80.17	98.12		98.16
	DPN^[28]	95.71	79.34	98.10	82.89	98.16
	ResDO-UNet^[10]	95.61	79.85	97.91	82.29
DRIVE	FRS-Net^[29]	95.74	79.97	98.04	82.69	98.16
	Bridge-Net^[30]	95.65	78.53	98.18	82.03	98.34
	DEF-Net^[31]	95.56	81.38	97.63	82.36	97.89
	Wave-Net^[32]	95.61	81.64	97.64	82.54
	STSANet^[33]	95.60	78.90	98.04		97.94
	DGRT-Net(文中)	97.01	82.51	98.43	82.86	98.74

数据集	方法	准确率	灵敏度	特异性	F1	AUC-ROC
	NAUNet^[25]	96.83	80.71	98.20		98.41
	IterMiUnet^[26]	95.91	84.43	97.04	78.75	98.12
	DPN^[28]	96.50	76.45	98.46	80.21	98.40
	ResDO-UNet^[10]	96.72	80.20	97.94	82.36
CHASE-DB1	FRS-Net^[29]	96.71	81.55	98.22	81.79	98.75
	Bridge-Net^[30]	96.67	81.32	98.40	82.93	98.93
	DEF-Net^[31]	96.26	80.53	98.35	80.76	98.57
	Wave-Net^[32]	96.64	82.84	98.21	83.49
	STSANet^[33]	96.49	79.74	98.20		98.48
	DGRT-Net(文中)	97.37	84.47	98.25	80.22	98.83