浏览全部资源
扫码关注微信
网络出版日期： 2024-09-10 ，
移动端阅览
杨可欣,刘骊,付晓东等.视网膜血管图像分割的尺度特征表示学习网络[J].中国图象图形学报,
Yang Kexin,Liu Li,Fu Xiaodong,et al.Scale feature representation learning network for retinal vessels image segmentation[J].Journal of Image and Graphics,
目的
2
针对视网膜血管图像分割中血管特征尺度多变、毛细血管细节丰富以及视杯视盘、病变等特殊区域干扰导致的表征不精确、分割误差大以及结果不准确等问题，提出一种视网膜血管图像分割的尺度特征表示学习网络，包括尺度特征表示、纹理特征增强和双重对比学习三个模块。
方法
2
首先输入视网膜图像集中的图像，通过引入空间自注意力构建尺度特征表示模块，对视网膜血管进行分层尺度表征；然后，采用上下文信息引导的纹理滤波器对血管尺度特征进行纹理特征增强；最后，通过采样血管尺度特征和纹理增强特征，并定义联合损失进行双重对比学习，优化两种特征空间中视杯视盘、病变等特殊区域的血管。
结果
2
为了验证方法的有效性，在三个具挑战性的数据集进行对比实验，结果表明，构建的视网膜血管图像分割网络有助于准确表示血管尺度特征和纹理增强特征，能够较好获得完整的视网膜毛细血管等特殊区域的血管分割结果。本文方法在DRIVE数据集中较大多数方法，Acc（Accuracy）值提高了0.67%，Sp（Specificity）值提高了0.48%；在STARE数据集中较大多数方法Se（Sensitivity）值提高了6.01%，Sp值提高了6.86%；在CHASE_DB1数据集中较大多数方法，Se值提高了1.88%，F1（F1 score）值提高了1.98%。
结论
2
本文提出的视网膜血管图像分割网络，能精准分割多尺度血管、毛细血管和特殊区域的血管，有效辅助视网膜血管疾病诊断。
Objective
2
Retinal vessels image segmentation refers to the separation of vessel pixels in a color fundus image from the background pixels. The morphology of retinal vessels is closely associated with many kinds of ophthalmic diseases and plays a crucial role in computer aided diagnosis and smart medicine. Retinal vessels image is also an important type of biological information that can be used as a recognition condition to construct personal identification systems in the field of social security. Lastly， the segmented retinal vessels images can serve as a priori for other anatomical sites， such as the macula. At present， retinal image segmentation methods can be divided into traditional methods and deep learning methods. Current methods of retinal vessels image segmentation have a good performance in the segmentation of large-scale vessels， which attributes to the ease of capturing features related to these prominent vessels. U-Net can effectively handle the complicated anatomical semantics in retinal vessels segmentation task and fuse adjacent-level features to learn more local and global semantic information for more accurate segmentation. Although significant progress in retinal vessels segmentation has been achieved with the advancement of deep learning， some challenging issues for retinal vessels image segmentation still remain. Firstly， current methods do not adequately represent vessels feature at multiple scales， resulting in poor segmentation results for retinal vessels with large differences in size and shape. Secondly， thin vessels are easily missed by current methods， especially located at the end of the extremely low contrast branches， leading to incomplete vessels segmentation. In addition， the medical semantic around retinal vessels is complex， special regions such as the optic cup， optic disc， and lesions can interfere with vessels segmentation and seriously affect the accuracy of retinal vessels segmentation. Also， most images in the STARE dataset have severe lesions， and the image information in different datasets varies greatly， resulting in lower sensitivity of vessels segmentation results. We propose a scale feature representation learning network for retinal vessels image segmentation by introducing three modules： scale feature representation， texture feature enhancement and double contrastive learning.
Method
2
In this study， first， we input the images in the retinal image set， and the first layer retinal vessel feature is extracted by U-Net-based backbone network. The hierarchical representation and stepwise fusion strategies from local details to global structure are used to fully represent the scale features of the vessels by introducing average pool operation and spatial self attention mechanism， to rich multi-scale information and obtain a vessel scale feature representation. Then， based on the vessel scale feature， the last four layers of scale encoding features are obtained by down-sampling， and in the process of skipping connection the scale encoding features combine the deeper features to obtain the intermediate features by using three types of convolutions. Intermediate features are enhanced using contextual information-guided texture filters， obtaining enhanced texture features， which effectively focus on the edge of the thin vessels by supplementing texture information. Finally， we sample the vessel scale feature and texture enhanced feature to obtain vessel pixels， background pixels near vessels and other background pixels in the two feature space domains. They are used as inputs for double constraint learning to calculate double constraint loss. Double constraint learning can reduce the intra-class distance， increase the variance between classes， and significantly improve the segmentation effect of thin vessels and vessels in special regions， such as optic cup， optic disc or lesions regions.
Result
2
The illustrated method is validated on three challenging retinal vessels image datasets （DRIVE， STARE and CHASE_DB1）. The Acc， Se and Sp on the STARE and CHASE_DB1 are 0.9761， 0.8191， 0.9866， 0.9765 0.8415 and 0.9874 respectively， which indicate that our proposed method outperforms the most competing methods and significantly improves the performance when extracting thin vessels in the regions of lesions or optic disc. Compared with other methods， the Accuracy （ Acc） value of the proposed method in the DRIVE dataset is increased by 0.67% and the Specificity （Sp） value is increased by 0.48% by average. In the STARE dataset， compared with other methods， the sensitivity （Se） value is increased by 6.01% and the Sp value is increased by 6.86% by average. In the CHASE_DB1 dataset， the Se value is increased by 1.88% and the F1 score （F1） value is increased by 1.98% compared to other methods. We visually analyze the advantages of our method by showing the vessels details and our method has achieved better results in the thin vessels with few vessel breaks. And we achieve better results in dark and unevenly illuminated images. Some methods have inaccurate segmentation results on crossed vessels at the optic disc and cannot easily segment thin vessels around the optic disc. Compared with other methods， the segmentation results of our methods on the cross vessels at the optic disc do not show the phenomenon of vessels rupture. It can be observed that we significantly improve the segmentation of vessels and thin vessels in the regions of optic disc， lesions or other regions. Our demonstrations show that the constructed retinal vessels image segmentation network is helpful to accurately represent the scale features of vessels and enhance the texture of vessels， it can effectively obtain complete retinal capillaries and vessel segmentation results in special regions. Lastly， among the ablation studies and analyses， these findings will doubtlessly prove that only a combination of three modules can simultaneously deal with the variable scale and anatomical semantics variations of retinal vessels.
Conclusion
2
In this paper， the retinal vessels image segmentation network can accurately segment multi-scale vessels， thin vessels and vessels in special regions by effectively representing the vessels scale features and enhancing texture features， and effectively assist in the diagnosis of vessel diseases.
视网膜血管图像分割尺度特征表示纹理特征增强纹理滤波器双重对比学习
retinal vessels segmentationscale feature representationtexture feature enhancementtexture filtersdouble constraint learning
Abdulsahib A A， Mahmoud M A， Aris H， Gunasekaran S S and Mohammed M A. 2022. An automated image segmentation and useful feature extraction algorithm for retinal blood vessels in fundus images. Electronic11（9）： 1295 ［DOI： https：//doi.org/10.3390/electronics11091295http://dx.doi.org/https：//doi.org/10.3390/electronics11091295］
Chaudhuri S， Chatterjee S， Katz N， Nelson M and Goldbaum M. 1989. Detection of blood vessels in retinal images using two-dimensional matched filters. IEEE Transactions on medical imaging，8（3）： 263-269 ［DOI： 10.1109/42.34715http://dx.doi.org/10.1109/42.34715］
Guo C L， Szemenyei M， Yi Y G， Wang W L， Chen B E and Fan C Q. 2021. Sa-unet： spatial attention u-net for retinal vessel segmentation//2020 25th international conference on pattern recognition. Milan， Italy： IEEE：1236-1242 ［DOI： 10.1109/ICPR48806.2021.9413346http://dx.doi.org/10.1109/ICPR48806.2021.9413346］
Huang H M， Lin L F， Tong R F， Hu H J， Zhang Q W， Iwamoto Y， Han X H， Chen Y W and Wu J. 2020. Unet 3+： A full-scale connected unet for medical image segmentation//IEEE international conference on acoustics， speech and signal processing. Barcelona， SPAIN： IEEE：1055-1059 ［DOI： 10.1109/icassp40776.2020.9053405http://dx.doi.org/10.1109/icassp40776.2020.9053405］
Khandouzi A， Ariafar A， Mashayekhpour Z， Pazira M and Baleghi Y. 2022. Retinal vessel segmentation， a review of classic and deep methods. Annals of Biomedical Engineering. 50（10）： 1292-1314 ［DOI： 10.1007/s10439-022-03058-0http://dx.doi.org/10.1007/s10439-022-03058-0］
Li H J， Li H， Shu H， Chen J Y， Hu Y and Liu J. 2023. Self-supervision boosted retinal vessel segmentation for cross-domain data//IEEE 20th International Symposium on Biomedical Imaging. Cartagena， Colombia： IEEE：1-5 ［DOI： 10.1109/ISBI53787.2023.10230561http://dx.doi.org/10.1109/ISBI53787.2023.10230561］
Lyu J， Ma C， and Cheng C. 2023. Improved U-Net Network for Retinal Vascular Segmentation. Journal of Frontiers of Computer Science and Technology，17（3）： 657-666
吕佳， 马超， 程超. 2023. 改进的U-Net网络用于视网膜血管分割. 计算机科学与探索，17（3）： 657-666 ［DOI： 10.3778/j.issn.1673-9418.2107060http://dx.doi.org/10.3778/j.issn.1673-9418.2107060］
Li J Y， Gao G， Yang L， Bian G B and Liu Y H. 2023. DPF-Net： A Dual-Path Progressive Fusion Network for Retinal Vessel Segmentation. IEEE Transactions on Instrumentation and Measurement，PP（99）： 1-1 ［DOI： 10.1109/TIM.2023.3277946http://dx.doi.org/10.1109/TIM.2023.3277946］
Luo L， Chen D L and Xue D Y. 2018. Retina Blood Vessels Semantic Segmentation Method Based on Modified U-Net//2018 Chinese Control And Decision Conference. Shenyang， China： IEEE：1892-1895 ［DOI： 10.1109/CompComm.2016.7924779http://dx.doi.org/10.1109/CompComm.2016.7924779］
Li L， Verma M， Nakashima Y， Nagahara H and Kawasaki R. 2020. Iternet： retinal image segmentation utilizing structural redundancy in vessel networks//Proceedings of the IEEE/CVF winter conference on applications of computer vision. Snowmass， CO： IEEE：3656-3665 ［DOI： https：//doi.org/10.1109/wacv45572.2020.9093621http://dx.doi.org/https：//doi.org/10.1109/wacv45572.2020.9093621］
Li M X， Zhou S L， Chen C， Zhang Y Y， Liu D and Xiong Z W. 2022. Retinal vessel segmentation with pixel-wise adaptive filters//2022 IEEE 19th International Symposium on Biomedical Imaging. Kolkata， India： IEEE：1-5 ［DOI： 10.1109/ISBI52829.2022.9761634http://dx.doi.org/10.1109/ISBI52829.2022.9761634］
Liskowski P and Krawiec K. 2016. Segmenting retinal blood vessels with deep neural networks. IEEE Transactions on Medical Imaging， 35（11）： 2369-2380 ［DOI： 10.1109/TMI.2016.2546227http://dx.doi.org/10.1109/TMI.2016.2546227］
Laibacher T， Weyde T and Jalali S. 2019. M2u-net： Effective and efficient retinal vessel segmentation for real-world applications// Proceedings of the IEEE/CVF conference on computer vision and pattern recognition workshops. Long Beach， CA： IEEE：115-124 ［DOI： 10.1109/CVPRW.2019.00020http://dx.doi.org/10.1109/CVPRW.2019.00020］
Liu Y H， Shen J， Yang L， Yu H N and Bian G B. 2023. Wave-Net： A lightweight deep network for retinal vessel segmentation from fundus images. Computers in Biology and Medicine，152（2）： 106341 ［DOI： 10.1016/j.compbiomed.2022.106341http://dx.doi.org/10.1016/j.compbiomed.2022.106341］
Liu Y H， Shen J， Yang L， Bian G B and Yu H N. 2023. ResDO-UNet： A deep residual network for accurate retinal vessel segmentation from fundus images. Biomedical Signal Processing and Control， 79： 104087 ［DOI： 10.1016/j.bspc.2022.104087http://dx.doi.org/10.1016/j.bspc.2022.104087］
Li Y， Zhang Y， Cui W G， Lei B Y， Kuang X H and Zhang T. 2022. Dual encoder-based dynamic-channel graph convolutional network with edge enhancement for retinal vessel segmentation. IEEE Transactions on Medical Imaging，41（8）： 1975-1989 ［DOI： 10.1109/TMI.2022.3151666http://dx.doi.org/10.1109/TMI.2022.3151666］
Mou L， Zhao Y T， Fu H Z， Liu Y H， Cheng J， Zheng Y L， Su P， Yang J L， Chen L， Frangi A F， Akiba M and Liu J. 2021. CS2-Net： deep learning segmentation of curvilinear structures in medical imaging. Medical Image Analysis，47： 101874 ［DOI： 10.1016/j.media.2020.101874http://dx.doi.org/10.1016/j.media.2020.101874］
Ronneberger O， Fischer P and Brox T. 2015. U-net： Convolutional networks for biomedical image segmentation// Medical Image Computing and Computer-Assisted Intervention–MICCAI 2015. Munich， Germany： Springer International Publishing：234-241 ［DOI： 10.1007/978-3-319-24574-4_28http://dx.doi.org/10.1007/978-3-319-24574-4_28］
Soares J V B， Leandro J J G， Cesar R M， Jelinek H F and Cree M J. 2006. Retinal vessel segmentation using the 2-d gabor wavelet and supervised classification. IEEE Transactions on Medical Imaging，25（9）： 1214-1222 ［DOI： 10.1109/TMI.2006.879967http://dx.doi.org/10.1109/TMI.2006.879967］
Toptaş B and Hanbay D. 2021. Retinal blood vessel segmentation using pixel-based feature vector. Biomedical Signal Processing and Control， 70： 103053 ［DOI： 10.1016/j.bspc.2021.103053http://dx.doi.org/10.1016/j.bspc.2021.103053］
Tan Y B， Yang K F， Zhao S X and Li Y J. 2022. Retinal vessel segmentation with skeletal prior and contrastive loss. IEEE Transactions on Medical Imaging，41（9）： 2238-2251 ［DOI： 10.1109/TMI.2022.3161681http://dx.doi.org/10.1109/TMI.2022.3161681］
Wang D Y， Haytham A， Pottenburgh J， Saeedi O and Tao Y. 2020. Hard attention net for automatic retinal vessel segmentation. IEEE Journal of Biomedical and Health Informatics，24（12）： 3384-3396 ［DOI： 10.1109/JBHI.2020.3002985http://dx.doi.org/10.1109/JBHI.2020.3002985］
Wu H S， Wang W， Zhong J F， Lei B Y， Wen Z K and Qin J. 2021. SCS-Net： A scale and context sensitive network for retinal vessel segmentation. Medical Image Analysis，70（10）： 102025 ［DOI： 10.1016/j.media.2021.102025http://dx.doi.org/10.1016/j.media.2021.102025］
Wang K， Zhang X， Huang S，Wang Q and Chen F. 2020. Ctf-net： retinal vessel segmentation via deep coarse-to-fine supervision network//2020 IEEE 17th International Symposium on Biomedical Imaging. Lowa， IA， USA： IEEE：1237-124 ［DOI： 10.1109/ISBI45749.2020.9098742http://dx.doi.org/10.1109/ISBI45749.2020.9098742］
Wu Y C， Xia Y， Song Y， Zhang Y N and Cai W D. 2018. Multiscale network followed network model for retinal vessel segmentation// Medical Image Computing and Computer Assisted Intervention. Granada， Spain： Springer International Publishing：119-126 ［DOI： https：//doi.org/10.1007/978-3-030-00934-2_14http://dx.doi.org/https：//doi.org/10.1007/978-3-030-00934-2_14］
Xu R， Liu T T， Ye X C， Lin L and Chen Y W. 2020. Boosting connectivity in retinal vessel segmentation via a recursive semantics-guided network//International Conference on Medical Image Computing and Computer-Assisted Intervention. Cham： Springer International Publishing：786-795 ［DOI： 10.1007/978-3-030-59722-1_76http://dx.doi.org/10.1007/978-3-030-59722-1_76］
Yang L， Wang H X， Zeng Q S， Liu Y H and Bian G B. 2021. A hybrid deep segmentation network for fundus vessels via deep-learning framework. Neurocomputing，448（11）： 168-178 ［DOI： 10.1016/j.neucom.2021.03.085http://dx.doi.org/10.1016/j.neucom.2021.03.085］
Zhou J C， Hao M L， Zhang D H， Zou P Y and Zhang W S. 2019. Fusion pspnet image segmentation based method for multi-focus image fusion. IEEE Photonics Journal，11（6）： 1-12 ［DOI： 10.1109/JPHOT.2019.2950949http://dx.doi.org/10.1109/JPHOT.2019.2950949］
Zhang M， Yu F， Zhao J， Zhang L and Li Q Z. 2020. Befd： boundary enhancement and feature denoising for vessel segmentation // Medical Image Computing and Computer Assisted Intervention–MICCAI 2020： 23rd International Conference. Lima， Peru： Springer International Publishing：775-785 ［DOI： 10.1109/CompComm.2016.7924779http://dx.doi.org/10.1109/CompComm.2016.7924779］
Zhang S H， Fu H Z， Yan Y G， Zhang Y B， Wu Q Y， Yang M， Tan， M K and Xu Y W. 2019. Attention guided network for retinal image segmentation// /Medical Image Computing and Computer Assisted Intervention–MICCAI 2019： 22nd International Conference. Shenzhen， China： Springer International Publishing：797-805 ［DOI： 10.1007/978-3-030-32239-7_88http://dx.doi.org/10.1007/978-3-030-32239-7_88］
Zhou T， Dong Y L， Huo B Q， Liu S and Ma Z J. 2021. U-Net and its applications in medical image segmentation： a review. Journal of Image and Graphic，26（9）： 2058-2077
周涛， 董雅丽， 霍兵强， 刘珊， 马宗军. 2021.U-Net网络医学图像分割应用综述. 中国图像图形学报，26（9）： 2058-2077 ［DOI： 10.11834/jig.200704http://dx.doi.org/10.11834/jig.200704］
Zhong X， Zhang H B， Li G L and Ji D H. 2022. Do you need sharpened details？ asking mmdc-net： multi-layer multi-scale dilated convolution network for retinal vessel segmentation. Computers in Biology and Medicine，150（3）： 106198 ［DOI： 10.1016/j.compbiomed.2022.106198http://dx.doi.org/10.1016/j.compbiomed.2022.106198］
Zhou Z W， Siddiquee M M R， Tajbakhsh and Liang J M. 2018. Unet++： a nested u-net architecture for medical image segmentation//Deep Learning in Medical Image Analysis and Multimodal Learning for Clinical Decision Support： 4th International Workshop， DLMIA 2018， and 8th International Workshop. Granada， Spain： Springer International Publishing：3-11 ［DOI： 10.1007/978-3-030-00889-5_1http://dx.doi.org/10.1007/978-3-030-00889-5_1］
相关文章
相关作者
相关机构
京公网安备11010802024621