臺灣博碩士論文加值系統

本篇研究針對糖尿病患者可能併發的糖尿病視網膜病變，根據其併發情況的嚴重性而分級為五類，透過目前主流於圖像分類辨識的卷積神經網路中的深度殘差網路，以本篇所提出的預訓練方法，加上以限制對比度自適應直方圖均衡算法事先處理視網膜病變圖片，學習糖尿病視網膜病變發作後各時期的病徵，且對視網膜病變的圖像資料集做出嚴重性的正確分類，最後與其他篇論文進行整體以及個別分類準確度的比較。
本篇論文使用的資料集來自Kaggle上eyePacs的視網膜圖片，在最後的分類任務中，平均各級標籤的分類準確率可以達到約96%，而初期的分類準確率則達到約97%，比起預處理前的約40%上升了2.425倍的幅度，此項研究不僅能在醫療上幫助醫師節省診斷上的人力與時間成本，更能搭配應用程式或是醫療系統，使患者可以自行做初步診斷。

In this paper, we focus on diabetic retinopathy, which may be complicated by diabetes, is classified into five classes according to the severity of the concomitant condition. Through the deep residual network, which is a kind of convolutional neural network, is currently used for image classification and identification, with the pre-training method proposed in this paper and adaptive contrast histogram equalization algorithm to learn the symptoms of various stages after the onset of diabetic retinopathy, and the severity of the retinopathy image data set will be correctly classified. Finally, the overall and individual classification accuracy will be compared with other papers.
The data set used in this paper is from the retina image of eyePacs on Kaggle website. In the final mission of classification, the average classification accuracy of labels can reach about 96%, and the classification accuracy of initial symptoms is about 97%. Compared with about 40% before preprocessing, the increase is 2.425 times. This research can not only help doctors save the labor and time cost of diagnosis, but also match the application or medical system, so that patients have the possibility of self-diagnosis.

中文摘要 i
Abstract ii
誌謝 iii
目錄 iv
圖目錄 vi
表目錄 viii

第一章 緒論 1
1-1 研究背景 1
1-2 研究動機 3
1-3 論文架構 5
第二章 相關研究 6
第三章 神經網路的發展與介紹 8
3-1 人工神經網路 8
3-1-1 激勵函數 10
3-1-2 反向傳播算法 16
3-2 卷積神經網路 20
3-2-1 AlexNet 23
3-2-2 VGGNet 25
3-2-3 GoogLeNet 26
3-2-4 ResNet 28
3-2-5 DenseNet 29
第四章 資料集的使用與預處理 32
4-1 各類DR資料集 32
4-2 資料預處理 36
4-3 資料平衡化 39
4-4 資料強化 40
第五章 模型權重的初始化 43
5-1 隨機初始化 43
5-2 Xavier初始化 43
5-3 He初始化 44
5-4 批量標準化 44
5-5 預訓練 45
第六章 實驗結果 48
6-1 實驗設備 48
6-2 不同預訓練方式之結果 48
6-3 與相關研究結果之比較 52
第七章 結論 54
參考文獻 55

[1] Centers for Disease Control and Prevention(CDC), "Long-Term Trends in Diabetes," Americans diagnosed, 2016.
[2] Kaggle:eyePacs[Online], Avaliable:https://www.kaggle.com/c/diabetic-retinopathy-detection/overview/description.
[3] Nicolas Coudray, Paolo Santiago Ocampo, Theodore Sakellaropoulos, Navneet Narula, Matija Snuderl, David Fenyö, Andre L. Moreira, Narges Razavian & Aristotelis Tsirigos, “Classification and mutation prediction from non–small cell lung cancer histopathology images using deep learning,” Nature Medicine, p. 1559–1567, 17 9 2018.
[4] Harry Pratta , Frans Coenen, Deborah M Broadbent , Simon P Hardinga, Yalin Zheng, “Convolutional Neural Networks for Diabetic Retinopathy,” in Medical Imaging Understanding and Analysis, Loughborough, UK, 2016.
[5] Ms. SheetalMarutiChougule, Prof. A. L. Renke, “New Preprocessing approach for Images in Diabetic Retinopathy Screening,” in International Journal of Engineering Research and Technology, 2017.
[6] Shaohua Wana , Yan Lianga , Yin Zhang, “Deep convolutional neural networks for diabetic retinopathy detection by image classification,” Computers and Electrical Engineering, p. 274–282, 2018.
[7] Seyed Hossein Rasta, Mahsa Eisazadeh Partovi, Hadi Seyedarabi, Alireza Javadzadeh, “A comparative study on preprocessing techniques in diabetic retinopathy retinal images: illumination correction and contrast enhancement,” in Journal of Medical Signals & Sensors, 2015.
[8] Aliaa A. A. Youssif, Atef Z. Ghalwash, and Amr S. Ghoneim, “Comparative Study of Contrast Enhancement and Illumination Equalization Methods for Retinal Vasculature Segmentation,” in Cairo International Biomedical Engineering Conference, 2006.
[9] T. L. a. Y. Zheng, "Adaptive image enhancement for retinal blood vessel segmentation," Electronics Letters, vol. 38, no. 19, pp. 1090-1091, 12 9 2002.
[10] C Lian, Y Liang, Rui Kang, Yao Xiang, "Deep Convolutional Neural Networks for Diabetic Retinopathy Classification," in ICAIP18, 2018.
[11] Avula Benzamin, Chandan Chakraborty, “Detection of Hard Exudates in Retinal Fundus Images Using Deep Learning,” in IEEE International Conference on System, Computation, Automation and Networking (ICSCAN), 2018.
[12] M. Haloi, "Improved Microaneurysm Detection using Deep Neural Networks," in Published in ArXiv, 2015.
[13] Sreejini K S and V K Govindan, "Severity Grading of DME from Retina Images: A Combination of PSO and FCM with Bayes Classifier," International Journal of Computer Applications 81(16), pp. 11-17, 2013.
[14] G. Cybenko, “Approximation by Superpositions of a Sigmoidal Function,” Mathematics of Control, Signals, and Systems, pp. 303-314, 1989.
[15] Inductiveload, "A selection of Normal Distribution Cumulative Density Functions (CDFs). Both the mean, μ, and variance, σ², are varied. The key is given on the graph," 2008.
[16] K. He, X. Zhang, S. Ren, and J. Sun, "Delving deep into rectifiers: Surpassing human-level performance on imagenet classification," in ICCV, 2015.
[17] I. J. Goodfellow, D. Warde-Farley, M. Mirza, A. Courville, and Y. Bengio, “Maxout networks,” arXiv:1302.4389, 2013.
[18] LeCun, Y., Bottou, L., Bengio, Y., and Haffner, P, "Gradient-based learning applied to document recognition," Proceedings of the IEEE, 86(11), p. 2278–2324, 1998.
[19] A. Krizhevsky, I. Sutskever, and G. Hinton, "ImageNet classification with deep convolutional neural networks," in NIPS'12 Proceedings of the 25th International Conference on Neural Information Processing Systems, Lake Tahoe, Nevada, 2012.
[20] Karen Simonyan and Andrew Zisserman, "Very Deep Convolutional Networks for Large-Scale Image Recognition," ICLR, 2015.
[21] C. Szegedy, W. Liu, Y. Jia, P. Sermanet, S. Reed, D. Anguelov, D. Erhan, V. Vanhoucke, and A. Rabinovich, “Going deeper with convolutions,” CVPR, 2015.
[22] K. He, X. Zhang, S. Ren, and J. Sun, "Deep residual learning for image recognition," CVPR, 2016.
[23] G. Huang, Z. Liu, and K. Q. Weinberger, “Densely connected convolutional networks,” arXiv preprint arXiv:1608.06993, 2016.
[24] IDRiD:Indian Diabetic Retinopathy Image[Online], Avaliable:https://idrid.grand-challenge.org/.
[25] Messidor[Online], Avaliable:http://www.adcis.net/en/third-party/messidor/.
[26] C. P. Wilkinson, Frederick L. Ferris, Ronald E. Klein, Paul P. Lee, Carl David, “Proposed international clinical diabetic retinopathy and diabetic macular edema disease severity scales,” Ophthalmology, pp. 1677-1682, 9 2003.
[27] K. Zuiderveld, Contrast limited adaptive histogram equalization, San Diego, CA, USA: Academic Press Professional, Inc., 1994.
[28] X. Glorot and Y. Bengio, “Understanding the difficulty of training deep feedforward neural networks,” in AISTATS, 2010.
[29] S. Ioffe and C. Szegedy, “Batch normalization: Accelerating deep network training by reducing internal covariate shif,” in ICML, 2015.
[30] Ghosh, Ratul, Kuntal Ghosh, and Sanjit Maitra, “Automatic detection and classification of diabetic retinopathy stages using CNN,” In: Signal Processing and Integrated Networks (SPIN), pp. 550-554, 2017.
[31] Lam C, Yi D, Guo M, Lindsey T, “Automated detection of diabetic retinopathy using deep learning,” AMIA Summits on Translational Science Proceedings, pp. 147-155, 2018.
[32] B. Graham, “Kaggle diabetic retinopathy detection,” in Report of University of Warwick, 2015.
[33] 林柏諺，吳宗益,“基於卷積神經網路對糖尿病患者視網膜病變的自動診斷,” 於 第十七屆微電子技術發展與應用研討會, 高雄, 2019.