隨著人工智慧的蓬勃發展和應用，深度學習技術已被廣泛地應用在醫學影像的分析。青光眼為前三大造成失明的眼科疾病，若能以深度學習的方法來協助分割眼底影像(Fundus Image)的視盤(Optic Disc)和視杯(Optic Cup)，將有助於眼科醫生使用電腦輔助診斷(Computer-Aided Diagnosis, CAD)。
雖然人工智慧可以有效地協助醫生作診斷，但訓練模型需依賴大量的專業性標註樣本，因此會消耗不少醫師人力與時間。為了解決此問題，本論文基於一種自我監督學習的卷積神經網路BT-Unet (Barlow Twins-Unet)，提出了一個Broaden Representation Module (BR Module)。其可以在預訓練階段利用不同感知域(Receptive Field)取得影像全局與局部的特徵並融合，藉此提升預訓練效果。最後，將此預訓練權重引入分割訓練以提升分割效果，並減少所需的標註樣本數量。
最後，本論文在以下三個公開資料庫進行了實驗，包括：RIM-ONE v2、RIM-ONE DL和ORIGA。在使用30%的標註樣本情況下，本論文提出的模型在ORIGA資料庫針對分割視盤，準確率可達99.8%；IoU (Intersection over Union)效果可達96.5%；Dice效果可達96.5%。在另外一方面，針對分割視杯時，準確率可達99.7%；IoU效果可達88.6%；Dice效果可達87.3%。與其他四個模型作比較，這樣的方法可減少標註樣本，也可以有效提升分割效果。

With the vigorous development and application of Artificial Intelligence (AI), the Deep Learning (DL) techniques have been widely used in the analysis of medical images. Glaucoma is one of the top three ophthalmic diseases causing blindness. If DL can be used to help divide the Optic Disc and Optic Cup of Fundus Images, it will be helpful for ophthalmologists to use the Computer-Aided Diagnosis (CAD).
Although AI can effectively assist doctors in diagnosis, training models rely on a large number of professional labeled samples, thus consuming a lot of physician manpower and time. To solve this problem, this paper proposes a Broaden Representation Module (BR Module) based on a self-supervised learning convolutional neural network BT-Unet (Barlow Twins Unet). It can take advantage of different receptive fields to acquire global and local features of the image and fuse them in the pre-training stage to improve the pre-training effect. Finally, the pre-training weight is introduced into the segmentation training to improve the segmentation effect and reduce the required number of labeled samples.
Finally, the experiments are carried out in the following three public datasets, including RIM-ONE v2, RIM-ONE DL, and ORIGA. In the case of using 30% labeled samples, the accuracy of the model proposed in this paper can reach 99.8% in the ORIGA database. The IoU (Intersection over Union) effect can reach 96.5%. Dice effect can reach 96.5%. On the other hand, the accuracy rate can reach 99.7% when aiming at the split vision cup. The IoU effect can reach 88.6%. Dice effect can reach 87.3%. In comparison with other four models, this method can reduce the number of labeled samples and effectively improve the segmentation effect.

[1] Y. C. Tham, X. Li, T. Y. Wong, H. A. Quigley, T. Aung, and C. T. Cheng, “Global prevalence of glaucoma and projections of glaucoma burden through 2040: A systematic review and meta-analysis,” Ophthalmology, vol. 121, no. 11, pp. 2081-2090 November 2014, DOI: 10.1016/j.ophtha.2014.05.013.
[2] B. Turgut, “Pearls for correct assessment of optic disc at glaucoma diagnosis,” US Ophthalmic Review, vol. 10, no. 2, pp. 104-110, November 2017, DOI: 10.17925/ USOR.2017.10.02.104.
[3] Z. Zhang, F. S. Yin, J. Liu, W. K. Wong, N. M. Tan, B. H. Lee, J. Cheng, and T. Y. Wong, “Origa-light: An online retinal fundus image database for glaucoma analysis and research,” in Proc. Annual International Conference of the IEEE Engineering in Medicine and Biology, Buenos Aires, Argentina, August 31-September 4, 2010, pp. 3065-3068, DOI: 10.1109/IEMBS.2010.5626137.
[4] O. Ronneberger, P. Fischer, and T. Brox, “U-Net: Convolutional networks for biomedical image segmentation,” in Proc. Medical Image Computing and Computer-Assisted Intervention, University of Freiburg, Germany, October 5-9, 2015, pp. 234-241, DOI: 10.1007/978-3-319-24574-4.
[5] J. Long, E. Shelhamer, and T. Darrell, “Fully convolutional networks for semantic segmentation,” in Proc. IEEE Conference on Computer Vision and Pattern Recognition, Boston, MA, USA, June 7-12, 2015, pp. 3431-3440, DOI: 10.1109/CVPR.2015. 7298965.
[6] S. Feng, H. Zhao, F. Shi, X. Cheng, M. Wang, Y. Ma, D. Xiang, W. Zhu, and X. Chen, “CPFNet: Context pyramid fusion network for medical image segmentation,” IEEE Transactions on Medical Imaging, vol. 39, no. 10, pp. 3008-3018, October 2020, DOI: 10.1109/TMI.2020.2983721.
[7] J. Wang, C. Gou, T. Shen, and F. Y. Wang, “Global segmentation-aided local masses detection in X-ray breast images,” in Proc. Chinese Automation Congress, Xi'an, China, November 30-December 2, 2018, pp. 3655-3660, DOI: 10.1109/CAC.2018.8623575.
[8] J. Redmon and A. Farhadi, “YOLO9000: Better, Faster, Stronger,” in Proc. IEEE Conference on Computer Vision and Pattern Recognition, Honolulu, HI, USA, July 21-26, 2017, pp. 6517-6525, DOI: 10.1109/CVPR.2017.690.
[9] M. AlGhamdi, M. Abdel-Mottaleb, and F. Collado-Mesa, “DU-Net: Convolutional network for the detection of arterial calcifications in mammograms,” IEEE Transactions on Medical Imaging, vol. 39, no. 10, pp. 3240-3249, October 2020, DOI: 10.1109/TMI.2020.2989737.
[10] G. Huang, Z. Liu, L. Van Der Maaten, and K. Q. Weinberger, “Densely connected convolutional networks,” in Proc. IEEE Conference on Computer Vision and Pattern Recognition, Honolulu, HI, USA, pp. 2261-2269, July 21-26, 2017, DOI: 10.1109/ CVPR.2017.243.
[11] J. Wang, X. Li, and Y. Cheng, “Towards an extended EfficientNet-based U-Net framework for joint optic disc and cup segmentation in the fundus image,” Biomedical Signal Processing and Control, vol. 85, Article 104906, pp. 1-14, August 2023, DOI: 10.1016/j.bspc.2023.104906.
[12] M. Tan and Q. Le, “Efficientnet: Rethinking model scaling for convolutional neural networks,” arXiv:1905.11946, September 2020, DOI: 10.48550/arXiv.1905.11946.
[13] S. Zheng, S. Jayasumana, B. R. Paredes, V. Vineet, Z. Su, D. Du, C. Huang, and P. H. S. Torr, “Conditional random fields as recurrent neural networks,” in Proc. IEEE International Conference on Computer Vision, Santiago, Chile, December 7-13, 2015, pp. 1529-1537, DOI: 10.1109/ICCV.2015.179.
[14] R. Bhattacharya, R. Hussain, A. Chatterjee, D. Paul, S. Chatterjee, and D. Dey, “PY-Net: Rethinking segmentation frameworks with dense pyramidal operations for optic disc and cup segmentation from retinal fundus images,” Biomedical Signal Processing and Control, vol. 85, Article 104895, pp. 1-10, March 2023, DOI: 10.1016/j.bspc. 2023.104895.
[15] Á. S. Hervella, J. Rouco, J. Novo, and M. Ortega, “End-to-end multitask learning for simultaneous optic disc and cup segmentation and glaucoma classification in eye fundus images,” Applied Soft Computing, vol. 116, no. 108347, pp. 1-13, February 2022, DOI: 10.1016/j.asoc.2021.108347.
[16] O. J. Afolabi, G. P. Mabuza-Hocquet, F. V. Nelwamondo, and B. S. Paul, “The use of U-Net lite and extreme gradient boost (XGB) for glaucoma detection,” IEEE Access, vol. 9, pp. 47411-47424, March 2021, DOI: 10.1109/ACCESS.2021.3068204.
[17] A. L. Maas, A. Y. Hannun, and Y. A. Ng, “Rectifier nonlinearities improve neural network acoustic models,” in Proc. Proceedings of the 30th International Conference on Machine Learning, Atlanta, GA, USA, June 16-21, 2013, pp. 3-9. [Online]. Available: https://ai.stanford.edu/~amaas/papers/relu_hybrid_icml2013_final.pdf, Accessed on: July 2, 2023.
[18] R. Zheng, Y. Zhong, S. Yan, H. Sun, H. Shen, and K. Huang, “MsVRL: Self-supervised multiscale visual representation learning via cross-level consistency for medical image Segmentation,” IEEE Transactions on Medical Imaging, vol. 42, no. 1, pp. 91-102, January 2023, DOI: 10.1109/TMI.2022.3204551.
[19] H. Zhang, J. Liu, T. Wu, Y. Chen, C. Liu, and Y. Wang, “JIANet: Jigsaw-invariant self-supervised learning of autoencoder-based reconstruction for melanoma segmentation,” IEEE Transactions on Instrumentation and Measurement, vol. 71, Article 5025413, pp. 1-13, November 2022, DOI: 10.1109/TIM.2022.3218033.
[20] K. Yu, L. Sun, J. Chen, M. Reynolds, T. Chaudhary, and K. Batmanghelich, “DrasCLR: A self-supervised framework of learning disease-related and anatomy-specific representation for 3D medical images,” arXiv:2302.10390, March 2023, DOI: 10.48550/arXiv.2302.10390.
[21] X. Li, L. Yu, H. Chen, C. W. Fu, L. Xing, and P. A. Heng, “Transformation-consistent self-ensembling model for semisupervised medical image segmentation,” IEEE Transactions on Neural Networks and Learning Systems, vol. 32, no. 2, pp. 523-534, February 2021, DOI: 10.1109/TNNLS.2020.2995319.
[22] S. Dong and Q. Chen, “A ROI based self-supervised strategy for retinal image analysis,” in Proc. International Conference on Optical and Photonic Engineering, Nanjing, China, November 24-27, 2022, vol. 125501I, pp. 325-330, DOI: 10.1117/12.2667274.
[23] A. Ben-Cohen, E. Klang, M. M. Amitai, J. Goldberger, and H. Greenspan, “Anatomical data augmentation for CNN based pixel-wise classification,” in Proc. IEEE International Symposium on Biomedical Imaging, Washington, DC, USA, April 4-7, 2018, pp. 1096-1099, DOI: 10.1109/ISBI.2018.8363762.
[24] L. Jing and Y. Tian, “Self-supervised visual feature learning with deep neural networks: A survey,” IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 43, no. 11, pp. 4037-4058, November 2021, DOI: 10.1109/TPAMI.2020.2992393.
[25] Z. Dai, B. Cai, and J. Chen, “UniMoCo: Unsupervised, Semi-supervised and fully-supervised visual representation learning,” in Proc. IEEE International Conference on Systems, Man, and Cybernetics, Prague, Czech Republic, October 9-12, 2022, pp. 3099-3106, DOI: 10.1109/SMC53654.2022.9945500.
[26] S. Chakraborty, A. R. Gosthipaty, and S. Paul, “G-SimCLR: Self-supervised contrastive learning with guided projection via pseudo labelling,” in Proc. International Conference on Data Mining Workshops, Sorrento, Italy, November 17-20, 2020, pp. 912-916, DOI: 10.1109/ICDMW51313.2020.00131.
[27] Z. Jure, J. Li, M. Ishan, L. C. Yann, and D. Stéphane, “Barlow Twins: Self-Supervised Learning via Redundancy Reduction,” arXiv:2103.03230, June 2021, DOI: 10.48550/ arXiv.2103.03230
[28] L. Wang, J. Gu, Y. Chen, Y. Liang, W. Zhang, J. Pu, and H. Chen, “Automated segmentation of the optic disc from fundus images using an asymmetric deep learning network,” Pattern Recognition, vol. 112, Article 107810, pp. 1-12, April 2021, DOI: 10.1016/j.patcog.2020.107810.
[29] Q. Zhu, X. Chen, Q. Meng, J. Song, G. Luo, M. Wang, F. Shi, Z. Chen, D. Xiang, and L. Pan, “GDCSeg-Net: General optic disc and cup segmentation network for multi-device fundus images,” Biomedical Optics Express, vol. 12, no. 10, pp. 6529-6544, October 2021, DOI:10.1364/BOE.434841
[30] H. Lei, W. Liu, H. Xie, B. Zhao, G. Yue, and B. Lei, “Unsupervised domain adaptation based image synthesis and feature alignment for joint optic disc and cup segmentation,” IEEE Journal of Biomedical and Health Informatics, vol. 26, no. 1, pp. 90-102, January 2022, DOI: 10.1109/JBHI.2021.3085770.
[31] Z. Chen, Y. Pan, and Y. Xia, “Reconstruction-driven dynamic refinement based unsupervised domain adaptation for joint optic disc and cup segmentation,” IEEE Journal of Biomedical and Health Informatics, April 2023, DOI: 10.1109/JBHI.2023. 3266576.
[32] D. P. Kingma and M. Welling, “Auto-encoding variational bayes,” arXiv:1312.6114, December 2022, DOI: 10.48550/arXiv.1312.6114.
[33] J. H. Gagan, H. S. Shirsat, Y. S. Kamath, N. I. R. Kuzhuppilly, and J. R. H. Kumar, “Automated optic disc segmentation using basis splines-based active contour,” IEEE Access, vol. 10, pp. 88152-88163, August 2022, DOI: 10.1109/ACCESS.2022.3199347.
[34] J. R. H. Kumar, S. Sachi, K. Chaudhury, S. Harsha, and B. K. Singh, “A unified approach for detection of diagnostically significant regions-of-interest in retinal fundus images,” in Proc. IEEE Region 10 Conference, Penang, Malaysia, November 5-8, 2017, pp. 19-24, DOI: 10.1109/TENCON.2017.8227829.
[35] L. Wang, H. Liu, Y. Lu, H. Chen, J. Zhang, and J. Pu, “A coarse-to-fine deep learning framework for optic disc segmentation in fundus images,” Biomed Signal Process Control, vol. 51, pp. 82-89, May 2019, DOI: 10.1016/j.bspc.2019.01.022.
[36] S. M. Shankaranarayana, K. Ram, K. Mitra, and M. Sivaprakasam, “Fully convolutional networks for monocular retinal depth estimation and optic disc-cup segmentation,” IEEE Journal of Biomedical and Health Informatics, vol. 23, no. 4, pp. 1417-1426, July 2019, DOI: 10.1109/JBHI.2019.2899403.
[37] S. Lu, H. Zhao, H. Liu, H. Li, and N. Wang, “PKRT-Net: Prior knowledge-based relation transformer network for optic cup and disc segmentation,” Neurocomputing, vol. 538, Article 126183, pp. 1-14, March 2023, DOI: 10.1016/j.neucom.2023.03.044.
[38] A. Zhao, H. Su, C. She, X. Huang, H. Li, H. Qiu, Z. Jiang, and G. Huang, “Joint optic disc and cup segmentation based on elliptical-like morphological feature and spatial geometry constraint,” Computers in Biology and Medicine, vol. 158, Article 106796, pp. 1-10, March 2023, DOI: 10.1016/j.compbiomed.2023.106796.
[39] R. Girshick, “Fast R-CNN,” in Proc. IEEE International Conference on Computer Vision, Santiago, Chile, December 7-13, 2015, pp. 1440-1448, DOI: 10.1109/ICCV. 2015.169.
[40] J. -D. Sun, C. Yao, J. Liu, W. Liu, and Z. -K. Yu, “GNAS-U2Net: A new optic cup and optic disc segmentation architecture with genetic neural architecture search,” IEEE Signal Processing Letters, vol. 29, pp. 697-701, February 2022, DOI: 10.1109/ LSP.2022.3151549.
[41] X. Qin, Z. Zhang, C. Huang, M. Dehghan, O. R. Zaiane, and M. Jagersand, “U2-Net: Going deeper with nested U-structure for salient object detection,” Pattern Recognition, vol. 106, Article 107404, pp. 1-12, May 2020, DOI: 10.1016/j.patcog.2020.107404.
[42] B. Baker, O. Gupta, N. Naik, and R. Raskar, “Designing neural network architectures using reinforcement learning,” arXiv:1611.02167, March 2017, DOI: 10.48550/arXiv.1611.02167.
[43] A. Sevastopolsky, “Optic disc and cup segmentation methods for glaucoma detection with modification of U-Net convolutional neural network,” Pattern Recognition and Image Analysis, vol. 27, pp. 618-624, November 2017, DOI: 10.1134/ S1054661817030269.
[44] N. S. Punn and S. Agarwal, “BT-Unet: A self-supervised learning framework for biomedical image segmentation using Barlow twins with U-Net models,” Machine Learning, vol. 111 pp. 4586-4600, August 2022, DOI: 10.1007/s10994-022-06219-3.
[45] F. Francisco, D. A. Tinguaro, S. Jose, S. Alayon, R. Arnay, and A. P. Denisse, “RIM-ONE DL: A unified retinal image database for assessing glaucoma using deep Learning,” Image Analysis & Stereology, vol. 39, no. 3, pp. 161-167, November 2020, DOI: 10.5566/ias.2346
[46] Szegedy, S. Ioffe, V. Vanhoucke, and A. Alemi, “Inception-v4, inception-resnet and the impact of residual connections on learning,” in Proc. Proceedings of the AAAI Conference on Artificial Intelligence, San Francisco, California, USA, February 4-9, 2017, vol. 31, pp. 4278-4284, DOI: 10.5555/3298023.3298188.
[47] S. Xie, R. Girshick, P. Dollár, Z. Tu, and K. He, “Aggregated residual transformations for deep neural networks,” in Proc. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, Honolulu, HI, USA, July 21-26, 2017, pp. 1492-1500, DOI: 10.1109/CVPR.2017.634.
[48] K. He, X. Zhang, S. Ren, and J. Sun, “Deep residual learning for image recognition,” in Proc. IEEE Conference on Computer Vision and Pattern Recognition, Las Vegas, NV, USA, June 27-30, 2016, pp. 770-778, DOI: 10.1109/CVPR.2016.90.
[49] R. R. Selvaraju, M. Cogswell, A. Das, R. Vedantam, D. Parikh, and D. Batra, “Grad-CAM: Visual explanations from deep networks via gradient-based localization,” in Proc. IEEE International Conference on Computer Vision, Venice, Italy, October 22-29, 2017, pp. 618-626, DOI: 10.1109/ICCV.2017.74.
[50] F. Yu and V. Koltun, “Multi-scale context aggregation by dilated convolutions,” arXiv:1511.07122, April 2016, DOI: 10.48550/arXiv.1511.07122.
[51] A. Krizhevsky, I. Sutskever, and G. Hinton, “Imagenet classification with deep convolutional neural networks,” Communications of the ACM, vol. 60, no. 6, pp. 84-90, June 2017, DOI:10.1145/3065386
[52] F. Abdullah, R. Imtiaz, H. A. Madni, H. A. Khan, T. M. Khan, M. A. U. Khan, and S. S. Naqvi, “A review on glaucoma disease detection using computerized techniques,” IEEE Access, vol. 9, pp. 37311-37333, February 2021, DOI: 10.1109/ACCESS.2021. 3061451.
[53] F. Fumero, S. Alay, J. L. Sanchez, J. Sigut, and M. Gonzalez-Hernandez, “RIM-ONE: an open retinal image database for optic nerve evaluation,” in Proc. International Symposium on Computer-Based Medical Systems, Bristol, UK, June 27-30, 2011, pp. 1-6, DOI:10.1109/CBMS.2011.5999143.
[54] N. Inoue, K. Yanashima, K. Magatani, and T. Kurihara, “Development of a simple diagnostic method for the glaucoma using ocular Fundus pictures,” in Proc. IEEE Engineering in Medicine and Biology Annual Conference, Shanghai, China, January 17-18, 2006, pp. 3355-3358, DOI: 10.1109/IEMBS.2005.1617196.