臺灣博碩士論文加值系統

淺層點狀角膜炎是現在眼科常見的疾病，然而傳統的診斷方式是透過螢光淚膜影像經由醫生參考相關量表來來分級診斷，而這個方法有個很大的問題，會受到醫師主觀解釋與判斷之螢光影像中的SPK的面積大小，導致每一位專業醫師診斷的等級會有所不同。本研究提出使用卷積神經網路(Convolutional Neural Network, CNN)模型來檢測螢光淚膜之SPK面積大小進而去分類其SPK等級，並將其定義為CNN淺層點狀角膜炎分級（Convolutional Neural Network Superficial Punctate Keratitis , CNN-SPK）。
裂隙燈錄製了102名受試者的標準螢光淚膜影像。其中有20位用於訓練CNN模型用以辨別眼表上各個特徵，其餘82位用來驗證CNN-SPK的效果，82位受試者我們提取他們左右眼的螢光影像所以總共有164張影像來做驗證，當中有109張影像是醫生判定為有SPK病徵，其餘56張為無SPK病徵之受試者。實驗結果，SPK患者的CNN-SPK顯著低於正常之受試者(P < 0.05)。而CNN-SPK和醫生判之SPK等級的相關性顯著(r = 0.8; P<0.05)。當CNN-BUT使用1020Pixel作為閥值來分級SPK等級，而分級之閥值靈敏性及特異性分別為0.84及0.79，代表CNN-SPK可以用在評估SPK及並且準確的自動評估眼睛表面之狀態。

Superficial punctate keratitis is a common disease in ophthalmology. However, the traditional diagnosis method is to use the doctor's reference to the relevant scale to diagnose the fluoroscopic tear film. This method has a big problem and will be subjectively interpreted by the doctor. The size of the SPK in the fluorescent image is determined to cause the level of diagnosis by each professional physician to be different. This study proposes to use the Convolutional Neural Network (CNN) model to detect the SPK area of the fluorescent tear film and then classify its SPK level, and define it as CNN shallow keratitis grading (Convolutional Neural) Network Superficial Punctate Keratitis, CNN-SPK). The slit lamp recorded a standard fluorescent tear film image of 102 subjects. Twenty of them were used to train the CNN model to identify the features on the ocular surface, the remaining 82 were used to verify the effect of CNN-SPK, and 82 subjects extracted the fluorescence images of their left and right eyes, so there were a total of 164 images. To verify, 109 images were diagnosed by the doctor as having SPK symptoms, and the remaining 56 were subjects without SPK symptoms. The results showed that CNN-SPK was significantly lower in SPK patients than in normal subjects (P < 0.05). The correlation between CNN-SPK and the SPK grade judged by the doctor was significant (r = 0.8; P < 0.05). When the CNN-BUT uses 1020Pixel as a threshold to grade the SPK level, the threshold sensitivity and specificity of the grading are 0.84 and 0.79, respectively, which means that the CNN-SPK can be used to evaluate the SPK and accurately and automatically assess the state of the surface of the eye.

書名頁 …………………………………………………………………… i
一、 論文口試委員審定書 …………………………………………………… ii
二、 中文摘要 ………………………………………………………………… iv
三、 英文摘要 ………………………………………………………………… v
四、 誌謝 ……………………………………………………………………… vi
五、 目錄 ……………………………………………………………………… vii
六、 表目錄 …………………………………………………………………… vii
七、 圖目錄 …………………………………………………………………… viii
一、 序論
1.1 研究動機 …………………………………………………………… 1
1.2 論文架構 …………………………………………………………… 2
二、 SPK介紹
2.1.1 乾眼症介紹 ………………………………………………………… 3
2.1.2 乾眼症的定義……………………………………………………… 4
2.2.1 SPK介紹…………………………………………………………… 6
2.2.2 SPK定義…………………………………………………………… 6
三、 深度學習
3.1 人工神經網路………………………………………………………… 9
3.2 卷積神經網路………………………………………………………… 11
四、 統計方式
4.1 統計方式介紹………………………………………………………… 14
五、 研究與實驗方法
5.1 受試者說明…………………………………………………………… 17
5.2 裂隙燈螢光影像擷取……………………………………………… 18
5.3 眼表特徵分類限制………………………………………………… 19
5.4 卷積神經網路架構………………………………………………… 21
5.5 SPK面積計算與測量……………………………………………… 23
六、 結果與討論
6.1 實驗環境…………………………………………………………… 25
6.2 卷積神經網路分類結果…………………………………………… 25
6.3 卷積神經網淺層點狀角膜炎(CNN-SPK) ………………………… 27
6.4 受試者分析………………………………………………………… 30
七、 結論
7.1 結論與未來發展……………………………………………………… 31
參考文獻 …………………………………………………………………… 32

[1] A. R. Bentivoglio, S. B. Bressman, E. Cassetta, D. Carretta, P. Tonali, and A. J. M. d. Albanese, "Analysis of blink rate patterns in normal subjects," vol. 12, no. 6, pp. 1028-1034, 1997.
[2] J. I. Prydal, P. Artal, H. Woon, F. W. J. I. o. Campbell, and v. science, "Study of human precorneal tear film thickness and structure using laser interferometry," vol. 33, no. 6, pp. 2006-2011, 1992.
[3] M. A. J. A. j. o. o. Lemp, "Advances in understanding and managing dry eye disease," vol. 146, no. 3, pp. 350-356. e1, 2008.
[4] M. C. Olson, D. R. Korb, and J. V. Greiner, "Increase in tear film lipid layer thickness following treatment with warm compresses in patients with meibomian gland dysfunction," Eye & contact lens, vol. 29, no. 2, pp. 96-99, 2003.
[5] H.-B. Chen, S. Yamabayashi, B. Ou, Y. Tanaka, S. Ohno, and S. Tsukahara, "Structure and composition of rat precorneal tear film. A study by an in vivo cryofixation," Investigative ophthalmology & visual science, vol. 38, no. 2, pp. 381-387, 1997.
[6] F. J. Holly, "Formation and rupture of the tear film," Experimental eye research, vol. 15, no. 5, pp. 515-525, 1973.
[7] A. Galor et al., "Prevalence and risk factors of dry eye syndrome in a United States veterans affairs population," vol. 152, no. 3, pp. 377-384. e2, 2011.
[8] C. Baudouin et al., "Revisiting the vicious circle of dry eye disease: a focus on the pathophysiology of meibomian gland dysfunction," vol. 100, no. 3, pp. 300-306, 2016.
[9] S. E. Moss, R. Klein, and B. E. J. A. o. o. Klein, "Incidence of dry eye in an older population," vol. 122, no. 3, pp. 369-373, 2004.
[10] A. Behrens et al., "Dysfunctional tear syndrome: a Delphi approach to treatment recommendations," vol. 25, no. 8, pp. 900-907, 2006.
[11] K. Tsubota and K. J. N. E. J. o. M. Nakamori, "Dry eyes and video display terminals," vol. 328, no. 8, pp. 584-584, 1993.
[12] R. Saxena, S. Srivastava, D. Trivedi, E. Anand, S. Joshi, and S. K. J. A. O. S. Gupta, "Impact of environmental pollution on the eye," vol. 81, no. 5, pp. 491-494, 2003.
[13] P. Wolkoff, J. Nøjgaard, P. Troiano, B. J. O. Piccoli, and e. medicine, "Eye complaints in the office environment: precorneal tear film integrity influenced by eye blinking efficiency," vol. 62, no. 1, pp. 4-12, 2005.
[14] J. J. Nichols, L. T. J. I. o. Sinnott, and v. science, "Tear film, contact lens, and patient-related factors associated with contact lens–related dry eye," vol. 47, no. 4, pp. 1319-1328, 2006.
[15] I. Toda, H. Fujishima, and K. J. A. o. Tsubota, "Ocular fatigue is the major symptom of dry eye," vol. 71, no. 3, pp. 347-352, 1993.
[16] K. Miyata, S. Amano, M. Sawa, and T. J. A. o. O. Nishida, "A novel grading method for superficial punctate keratopathy magnitude and its correlation with corneal epithelial permeability," vol. 121, no. 11, pp. 1537-1539, 2003.
[17] R. J. O. i. Read and inflammation, "Clinical mini-review: systemic lupus erythematosus and the eye," vol. 12, no. 2, pp. 87-99, 2004.
[18] C. Vitali et al., "Classification criteria for Sjögren's syndrome: a revised version of the European criteria proposed by the American-European Consensus Group," vol. 61, no. 6, pp. 554-558, 2002.
[19] A. J. E. Lemp and C. Lens, "Report of the National Eye Institute/Industry workshop on clinical trials in dry eyes," vol. 21, no. 4, pp. 221-232, 1995.
[20] "The Definition and Classification of Dry Eye Disease: Report of the Definition and Classification Subcommittee of the International Dry Eye Workshop (2007),The Ocular Surface,Volume 5, Issue 2,2007,Pages 75-92."
[21] C. G. Begley et al., "The relationship between habitual patient-reported symptoms and clinical signs among patients with dry eye of varying severity," vol. 44, no. 11, pp. 4753-4761, 2003.
[22] G. J. B. j. o. o. Rieger, "The importance of the precorneal tear film for the quality of optical imaging," vol. 76, no. 3, pp. 157-158, 1992.
[23] E. Goto, Y. Yagi, Y. Matsumoto, and K. J. A. j. o. o. Tsubota, "Impaired functional visual acuity of dry eye patients," vol. 133, no. 2, pp. 181-186, 2002.
[24] S. C. Pflugfelder, D. Jones, Z. Ji, A. Afonso, and D. J. C. e. r. Monroy, "Altered cytokine balance in the tear fluid and conjunctiva of patients with Sjögren's syndrome keratoconjunctivitis sicca," vol. 19, no. 3, pp. 201-211, 1999.
[25] L. Luo, D.-Q. Li, R. M. Corrales, S. C. J. E. Pflugfelder, and c. lens, "Hyperosmolar saline is a proinflammatory stress on the mouse ocular surface," vol. 31, no. 5, pp. 186-193, 2005.
[26] L. Luo et al., "Experimental dry eye stimulates production of inflammatory cytokines and MMP-9 and activates MAPK signaling pathways on the ocular surface," vol. 45, no. 12, pp. 4293-4301, 2004.
[27] J. Y. Niederkorn et al., "Desiccating stress induces T cell-mediated Sjögren’s syndrome-like lacrimal keratoconjunctivitis," vol. 176, no. 7, pp. 3950-3957, 2006.
[28] F. J. J. E. e. r. Holly, "Formation and rupture of the tear film," vol. 15, no. 5, pp. 515-525, 1973.
[29] "Superficial Punctate Keratitis By Melvin I. Roat , MD, FACS, Sidney Kimmel Medical College at Thomas Jefferson University."
[30] R. W. J. T. o. t. A. O. S. Darrell, "Thygeson's superficial punctate keratitis: natural history and association with HLA DR3," vol. 79, p. 486, 1981.
[31] P. J. J. o. t. A. M. A. Thygeson, "Superficial punctate keratitis," vol. 144, no. 18, pp. 1544-1549, 1950.
[32] P. J. A. o. O. THYGESON, "Further observations on superficial punctate keratitis," vol. 66, no. 2, pp. 158-162, 1961.
[33] C. S. De Paiva and S. C. J. A. j. o. o. Pflugfelder, "Corneal epitheliopathy of dry eye induces hyperesthesia to mechanical air jet stimulation," vol. 137, no. 1, pp. 109-115, 2004.
[34] A. J. Bron, V. E. Evans, and J. A. J. C. Smith, "Grading of corneal and conjunctival staining in the context of other dry eye tests," vol. 22, no. 7, pp. 640-650, 2003.
[35] O. J. A. o. o. Van Bijsterveld, "Diagnostic tests in the sicca syndrome," vol. 82, no. 1, pp. 10-14, 1969.
[36] J. D. Rodriguez et al., "Automated grading system for evaluation of superficial punctate keratitis associated with dry eye," vol. 56, no. 4, pp. 2340-2347, 2015.
[37] J. J. N. n. Schmidhuber, "Deep learning in neural networks: An overview," vol. 61, pp. 85-117, 2015.
[38] Y. LeCun, Y. Bengio, and G. J. n. Hinton, "Deep learning," vol. 521, no. 7553, p. 436, 2015.
[39] V. Nair and G. E. Hinton, "Rectified linear units improve restricted boltzmann machines," in Proceedings of the 27th international conference on machine learning (ICML-10), 2010, pp. 807-814.
[40] S. Sigtia and S. Dixon, "Improved music feature learning with deep neural networks," in 2014 IEEE international conference on acoustics, speech and signal processing (ICASSP), 2014, pp. 6959-6963: IEEE.
[41] M. A. Nielsen, Neural networks and deep learning. Determination press San Francisco, CA, USA:, 2015.
[42] Y. Le Cun et al., "Handwritten digit recognition: Applications of neural network chips and automatic learning," IEEE Communications Magazine, vol. 27, no. 11, pp. 41-46, 1989.
[43] G. Hinton et al., "Deep neural networks for acoustic modeling in speech recognition: The shared views of four research groups," IEEE Signal Processing Magazine, vol. 29, no. 6, pp. 82-97, 2012.
[44] S. Min, B. Lee, and S. J. B. i. b. Yoon, "Deep learning in bioinformatics," vol. 18, no. 5, pp. 851-869, 2017.
[45] D. H. Hubel and T. N. J. T. J. o. p. Wiesel, "Receptive fields of single neurones in the cat's striate cortex," vol. 148, no. 3, pp. 574-591, 1959.
[46] Y. LeCun, Y. J. T. h. o. b. t. Bengio, and n. networks, "Convolutional networks for images, speech, and time series," vol. 3361, no. 10, p. 1995, 1995.
[47] A. Krizhevsky, I. Sutskever, and G. E. Hinton, "Imagenet classification with deep convolutional neural networks," in Advances in neural information processing systems, 2012, pp. 1097-1105.
[48] K. J. C. j. o. i. m. Hajian-Tilaki, "Receiver operating characteristic (ROC) curve analysis for medical diagnostic test evaluation," vol. 4, no. 2, p. 627, 2013.
[49] K. H. Zou, A. J. O’Malley, and L. J. C. Mauri, "Receiver-operating characteristic analysis for evaluating diagnostic tests and predictive models," vol. 115, no. 5, pp. 654-657, 2007.
[50] S. M. J. S. S. Stigler, "Francis Galton's account of the invention of correlation," pp. 73-79, 1989.
[51] K. J. T. L. Pearson, Edinburgh,, D. P. Magazine, and J. o. Science, "X. On the criterion that a given system of deviations from the probable in the case of a correlated system of variables is such that it can be reasonably supposed to have arisen from random sampling," vol. 50, no. 302, pp. 157-175, 1900.
[52] W. M. A. J. B. o. t. W. H. Organization, "World Medical Association Declaration of Helsinki. Ethical principles for medical research involving human subjects," vol. 79, no. 4, p. 373, 2001.
[53] D. P. Strik, A. M. Domnanovich, L. Zani, R. Braun, P. J. E. M. Holubar, and Software, "Prediction of trace compounds in biogas from anaerobic digestion using the MATLAB Neural Network Toolbox," vol. 20, no. 6, pp. 803-810, 2005.