基于动态特征融合与区域增强 Transformer 的视网膜血管分割算法

Retinal Vessel Segmentation Algorithm Based on Dynamic Feature Fusion and Region-enhanced Transformer

摘要

针对视网膜血管分割中存在微细血管分割断裂、血管纹理特征丢失和病灶信息干扰等问题,提出一种基于动态特征融合(dynamic feature fusion,DFF)与区域增强 Transformer 的视网膜血管分割算法 DRET-Net。 首先,采用区域增强 Transformer 构建双边特征提取编码端,以缓解原始编码端血管信息损失,提升算法对全局特征的捕捉能力;其次,在原始编码端设计双重注意力模块,实现血管纹理的更精准识别,同时,增强模型对微细血管的提取能力;然后,在解码部分引入 DFF 模块,有效去除病理信息干扰,突出病灶区域血管信息;最后,在公共数据集 DRIVE、CHASE_DB1 和 STARE 上进行实验,F1 分数分别为 82.94%、81.06% 和 83.64%,准确率分别为 97.11%、97.60% 和 97.60%,灵敏度分别为 80.18%、81.25% 和 80.38%。 实验结果表明,该算法表现出较强的鲁棒性和泛化能力,并且整体性能优于现有大部分先进算法。

Abstract

To address various issues encountered by the existing methods, such as fragmentation of fine vessels, loss of vascular texture features, and interference from lesion information, a retinal vessel segmentation algorithm based on dynamic features fusion (DFF) and region-enhanced Transformer is proposed, named DRET-Net. First, a region-enhanced Transformer was used to construct a bilateral feature extraction encoding end, alleviating vessel loss in the original encoding end and improving the algorithm's ability to capture global features. Second, a dual attention module was embedded in the original encoding end to achieve more precise recognition of vascular textures and to enhance the model's capability to extract fine vessels. Third, a DFF module was introduced in the decoding part to effectively remove pathological information interference and highlight vascular information in lesion areas. Finally, experiments were conducted on the public datasets DRIVE, CHASE_DB1, and STARE, the F1 score reaches 82.94%, 81.06%, and 83.64%, the accuracy reaches 97.11%, 97.60%, and 97.60%, and the sensitivity reaches 80.18%, 81.25%, and 80.38%, respectively. Results indicate that the proposed algorithm demonstrates strong robustness and generalization capability, and overall performance superior to most existing advanced algorithms.

关键词

视网膜血管分割 / 区域增强 Transformer / 双重注意力 / 动态特征融合模块 / 全局特征 / 血管纹理

Key words

retinal vessel segmentation / region-enhanced Transformer / dual attention / dynamic feature fusion (DFF) module / global feature / vascular texture

参考文献

[1] 王素玉, 董政伦, 刘涵宇 . 基于注意力机制和稠密卷积的视网膜微血管分割算法[J]. 北京工业大学学报, 2021, 47(5): 500-507.

[2] Wang S Y, Dong Z L, Liu H Y . Retinal microvascular segmentation algorithm based on attention mechanism and dense convolution[J]. Journal of Beijing University of Technology, 2021, 47(5): 500-507. (in Chinese)

[3] 梁礼明, 阳渊, 何安军, 等 . 跨级可变形 Transformer 编解码视网膜图像分割算法[J]. 无线电工程, 2023, 53(9): 1990-2001.

[4] Liang L M, Yang Y, He A J, et al. Cross—stage deformable transformer encoding and decoding algorithm for retinal image segmentation[J]. Radio Engineering, 2023, 53(9): 1990-2001. (in Chinese)

[5] Ali A, Wan Zaki W M D, Hussain A . Blood vessel segmentation from color retinal images using K—means clustering and 2D Gabor wavelet[M]// Lecture Notes in Electrical Engineering. Cham: Springer, 2017: 221-227.

[6] Oliveira W S, Teixeira J V, Ren T I, et al. Unsupervised retinal vessel segmentation using combined filters[J]. PLoS One, 2016, 11(2): e0149943.

[7] Hassan G, El—bendary N, Hassanien A E, et al. Retinal blood vessel segmentation approach based on mathematical morphology[J]. Procedia Computer Science, 2015, 65: 612-622.

[8] Relan D, Macgillivray T, Ballerini L, et al. Automatic retinal vessel classification using a least square—support vector machine in VAMPIRE[C]// 2014 36th Annual International Conference of the IEEE Engineering in Medicine and Biology Society. Piscataway, NJ: IEEE, 2014: 142-145.

[9] Kuppusamy P, Basha M M, Hung C L . Retinal blood vessel segmentation using random forest with Gabor and Canny edge features[C]// 2022 International Conference on Smart Technologies and Systems for Next Generation Computing. Piscataway, NJ: IEEE, 2022: 1-4.

[10] Sindhusaranya B, Geetha M R . Retinal blood vessel segmentation using root guided decision tree assisted enhanced fuzzy C—mean clustering for disease identification[J]. Biomedical Signal Processing and Control, 2023, 82: 104525.

[11] Ronneberger O, Fischer P, Brox T . U—Net: convolutional networks for biomedical image segmentation[M]// Lecture Notes in Computer Science. Cham: Springer, 2015: 234-241.

[12] Guo C L, Szemenyei M, Yi Y G, et al. SA—UNet: spatial attention U—Net for retinal vessel segmentation[C]// 2020 25th International Conference on Pattern Recognition. Piscataway, NJ: IEEE, 2021: 1236-1242.

[13] Zhu Y F, Xu X, Zhang X D, et al. CCS—UNet: a cross—channel spatial attention model for accurate retinal vessel segmentation[J]. Biomedical Optics Express, 2023, 14(9): 4739-4758.

[14] Wang B, Qiu S, He H G . Dual encoding U—Net for retinal vessel segmentation[M]// Lecture Notes in Computer Science. Cham: Springer, 2019: 84-92.

[15] Vaswani A, Shazeer N, Paramar N, et al. Attention is all you need[PP/OL]. V7. arXiv(2023—08—02)[2024—05—12]. https://doi.org/10.48550/arXiv.1706.03762.

[16] Dosovitskiy A, Beyer L, Kolesnikov A, et al. An image is worth 16×16 words: transformers for image recognition at scale[PP/OL]. V2. arXiv(2021—06—03)[2024—05—12]. https://doi.org/10.48550/arXiv.2010.11929.

[17] Liu Z, Lin Y T, Cao Y, et al. Swin Transformer: hierarchical vision Transformer using shifted windows[C]// 2021 IEEE/CVF International Conference on Computer Vision. Piscataway, NJ: IEEE, 2021: 10012-10022.

[18] Shi Z D, 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.

[19] Jia W R, Ma S M, Geng P, et al. DT—Net: joint dual—input transformer and CNN for retinal vessel segmentation[J]. Computers, Materials & Continua, 2023, 76(3): 3393-3411.

[20] Hu X L, Wang L J, Li Y M . HT—Net: a hybrid transformer network for fundus vessel segmentation[J]. Sensors, 2022, 22(18): 6782.

[21] Yang J, Qiu P J, Zhang Y C, et al. D—Net: dynamic large kernel with dynamic feature fusion for volumetric medical image segmentation[PP/OL]. V2. arXiv(2024—10—17)[2024—05—22]. https://doi.org/10.48550/arXiv.2403.10674.

[22] Lan L B, Cai P Z, Jiang L, et al. BRAU—net++: U—shaped hybrid CNN—Transformer network for medical image segmentation[PP/OL]. V2. arXiv(2024—09—30)[2024—05—24]. https://doi.org/10.48550/arXiv.2401.00722.

[23] Oktay O, Schlemper J, Le Folgoc L, et al. Attention U—Net: learning where to look for the pancreas[PP/OL]. V3. arXiv(2018—05—20)[2024—05—28]. https://doi.org/10.48550/arXiv.1804.03999.

[24] Wang C, Zhao Z Y, Ren Q Q, et al. Dense U—Net based on patch—based learning for retinal vessel segmentation[J]. Entropy, 2019, 21(2): 168.

[25] Liu W T, Yang H H, Tian T, et al. Full—resolution network and dual—threshold iteration for retinal vessel and coronary angiograph segmentation[J]. IEEE Journal of Biomedical and Health Informatics, 2022, 26(9): 4623-4634.

[26] Alvarado—Carrillo D E, Dalmau—Cedeño O S . Width attention based convolutional neural network for retinal vessel segmentation[J]. Expert Systems with Applications, 2022, 209: 118313.

[27] Zhang Y, He M, Chen Z N, 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.

[28] Kumar A, Agrawal R K, Joseph L . IterMiUnet: a lightweight architecture for automatic blood vessel segmentation[J]. Multimedia Tools and Applications, 2023, 82(28): 43207-43231.

[29] Zhang H B, Zhong X, Li Z J, et al. TiM—Net: transformer in M—Net for retinal vessel segmentation[J]. Journal of Healthcare Engineering, 2022, 2022(1): 9016401.

[30] Liu M T, Wang Y Y, Wang L, et al. IMFF—Net: an integrated multi—scale feature fusion network for accurate retinal vessel segmentation from fundus images[J]. Biomedical Signal Processing and Control, 2024, 91: 105980.

[31] Nikoloulopoulou N, Perikos I, Daramouskas I, et al. A convolutional autoencoder approach for boosting the specificity of retinal blood vessels segmentation[J]. Applied Sciences, 2023, 13(5): 3255.

[32] Ding W P, Sun Y, Huang J S, et al. RCAR—UNet: Retinal vessel segmentation network algorithm via novel rough attention mechanism[J]. Information Sciences, 2024, 657: 120007.

基金资助

国家自然科学基金资助项目(51365017)

国家自然科学基金资助项目(61463018)

江西省自然科学基金资助项目(20192BAB205084)

江西省教育厅科学技术研究青年项目(GJJ2200848)