根據中華民國衛生福利部統計處所公布的統計數據，每年有將近3000人死於口腔癌，更有將近7000名的新病患罹患口腔癌。然而，若是能在癌症的早期就能篩檢出來並給予適當的治療，便可以大幅提高癌症的存活率。因此，在初期階段診斷出口腔癌是本研究的主要目標.在先前的研究中，我們已經開發出了一種拍攝口腔自體螢光影像的儀器系統，此系統具有拍攝鏡頭以及兩種不同的激發光源，用於激發並拍攝口腔的自體螢光影像經過臨床收案與資料測試後，深度卷積網路模型會自動的圈選出影像中的可能為病灶的區域，我們透過低頻的賈伯濾波器提取出特徵圖並輸入到深度卷積網路模型中，運用初始模組（inception module）來預測口腔中可能為病灶區域。雖然我們透過儀器進行臨床收案已經持續兩年，但是礙於深度卷積網路需要龐大的資料量，我們目前資料庫中的口腔癌患者資料仍舊相當不足且遠小於正常人的口腔影像數量。此外，在拍攝影像的同時，由於人為操作手持儀器拍攝兩張影像，有可能會導致影像位移，造成FAD影像與NADH影像的拍攝位置有少許差異。
為了解決手持儀器造成的影像位移以及資料不足的問題，本研究運用多個生成對抗網路來生成癌症資料，並且藉此提升癌症可疑病徵自動圈選演算法的圈選效果。在我們所知的已發表研究中，這是深度卷積生成對抗網路第一次用來生成新的口腔癌影像。本研究會透過臨床收案所收到的正常的口腔FAD影像來生成一個新的人造口腔癌FAD影像。第二步，透過循環生成對抗網路（Cycle-Consistent Adversarial Networks）來將上述所生成的口腔癌FAD影像轉換成對應的口腔癌NADH影像。第三步，將生成的口腔癌NADH影像經過訓練過好的超解析度生成對抗網路（Super Resolution Generative Adversarial Networks）進行影像解析度強化的後處理。之後，從新生成的口腔癌資料中提取的特徵與從醫院臨床收案中收集到的真實口腔癌資料的特徵會用二次判別分析技術來比較並驗證。最後會將生成的口腔癌資料添加到口腔癌自動圈選演算法的訓練集中。通過以上步驟，我們可以解決演算法訓練數據集內癌症資料過少的問題，並使我們的數據更加多樣化。

Oral cancer causes almost 3000 deaths every year and more than 7000 new patients appear in the statistic published by the Health Promotion Administration. However, the survival rate is greatly improved if the cancer is detected sufficiently early and treated proper treatment. Accordingly, to diagnose oral cancer in the early stage is the primary goal of this study. In previous research, we had developed an oral cavity imaging system with two different excitation light sources for oral auto-fluorescence imaging. After clinical test and data collection, the region of interest in the images will automatically be marked by the deep convolutional networks model. Through the deep convolutional networks model, we input the texture feature maps which were extracted from Gabor filter with low frequency to predict the oral cancer ROI by inception module. Although we have been collected clinical data for two years, the amount of cancer data is still insufficient for deep convolutional network training, and it is far below normal data in our database. In addition, the handheld device of our instrument may cause the image displacement from FAD image to NADH image.

To address the image displacement caused by the handheld device and the data insufficient issue, this study presents an enhanced algorithm to generate synthetic cancer data for oral cancer region detection algorithm based on several generative adversarial network models. In the proposed approval, Deep Convolutional Generative Adversarial Networks is first used to produce a new image of the oral cancer. It will output a synthetic cancer data which was transferred from a normal data we collected from clinical screening. Second, Cycle-Consistent Adversarial Networks is using to transfer FAD images to NADH images. Third, let NADH images go through a pre-trained Super Resolution Generative Adversarial Networks as a post processing. After that, A Quadratic Discriminant Analysis technique is the used to compare the features extracted from the new data with those of the old data which were collected from the hospital clinical screening. The new data are then added to the training dataset of the oral cancer detection algorithm. Through above steps, we may address the problem of too few cancer data and make our data more diversity.

[1] "Health Promotion Administration, Ministry of Health and Welfare https://www.hpa.gov.tw/Pages/Detail.aspx?nodeid=204&pid=1118."
[2] S. Y. Kao and E. Lim, "An overview of detection and screening of oral cancer in Taiwan," The Chinese journal of dental research: the official journal of the Scientific Section of the Chinese Stomatological Association (CSA), vol. 18, no. 1, pp. 7-12, 2015.
[3] K. Thankappan and M. A. Kuriakose, "Sentinel Node Biopsy in Oral Cancer," in Contemporary Oral Oncology: Springer, 2017, pp. 211-233.
[4] T. Remmerbach, N. Pomjanski, U. Bauer, and H. Neumann, "Liquid-based versus conventional cytology of oral brush biopsies: a split-sample pilot study," Clinical oral investigations, vol. 21, no. 8, pp. 2493-2498, 2017.
[5] X. Wang, K. E. Kaczor-Urbanowicz, and D. T. Wong, "Salivary biomarkers in cancer detection," Medical Oncology, vol. 34, no. 1, p. 7, 2017.
[6] Y.-W. Chen et al., "Use of methylene blue as a diagnostic aid in early detection of oral cancer and precancerous lesions," British Journal of Oral and Maxillofacial Surgery, vol. 45, no. 7, pp. 590-591, 2007.
[7] D. C. De Veld, M. Skurichina, M. J. Witjes, R. P. Duin, H. J. Sterenborg, and J. L. Roodenburg, "Autofluorescence and diffuse reflectance spectroscopy for oral oncology," Lasers in Surgery and Medicine: The Official Journal of the American Society for Laser Medicine and Surgery, vol. 36, no. 5, pp. 356-364, 2005.
[8] D. C. de Veld, M. Skurichina, M. J. Witjes, R. P. Duin, H. J. Sterenborg, and J. L. Roodenburg, "Clinical study for classification of benign, dysplastic, and malignant oral lesions using autofluorescence spectroscopy," Journal of biomedical optics, vol. 9, no. 5, pp. 940-951, 2004.
[9] L. Z. Li et al., "Quantitative magnetic resonance and optical imaging biomarkers of melanoma metastatic potential," Proceedings of the National Academy of Sciences, vol. 106, no. 16, pp. 6608-6613, 2009.
[10] M. Cicciù, A. S. Herford, G. Cervino, G. Troiano, F. Lauritano, and L. Laino, "Tissue fluorescence imaging (VELscope) for quick non-invasive diagnosis in oral pathology," Journal of Craniofacial Surgery, vol. 28, no. 2, pp. e112-e115, 2017.
[11] H. Teng-Yi, "Application of Autofluorescence Imaging on Clinical Oral Cancer Screening," pp. 1-56, 2014.
[12] C. Ming-Hsuan, "Texture Feature Analysis of Cancer Detection," pp. 1-58, 2016.
[13] C. Chih-Hung, "Using Deep Convolutional Neural Networks for Oral Cancer Detection Based on Texture Feature Images," pp. 1-69, 2017.
[14] D. K. Iakovidis, E. G. Keramidas, and D. Maroulis, "Fuzzy local binary patterns for ultrasound texture characterization," in International conference image analysis and recognition, 2008, pp. 750-759: Springer.
[15] J. Long, E. Shelhamer, and T. Darrell, "Fully convolutional networks for semantic segmentation," in Proceedings of the IEEE conference on computer vision and pattern recognition, 2015, pp. 3431-3440.
[16] T.-Y. Lin, P. Dollár, R. B. Girshick, K. He, B. Hariharan, and S. J. Belongie, "Feature Pyramid Networks for Object Detection," in CVPR, 2017, vol. 1, no. 2, p. 3.
[17] C. Ledig et al., "Photo-Realistic Single Image Super-Resolution Using a Generative Adversarial Network," in CVPR, 2017, vol. 2, no. 3, p. 4.
[18] J.-Y. Zhu, T. Park, P. Isola, and A. A. Efros, "Unpaired image-to-image translation using cycle-consistent adversarial networks," arXiv preprint, 2017.
[19] S. Reed, Z. Akata, X. Yan, L. Logeswaran, B. Schiele, and H. Lee, "Generative adversarial text to image synthesis," arXiv preprint arXiv:1605.05396, 2016.
[20] M. Frid-Adar, E. Klang, M. Amitai, J. Goldberger, and H. Greenspan, "Synthetic data augmentation using GAN for improved liver lesion classification," in 2018 IEEE 15th international symposium on biomedical imaging (ISBI 2018), 2018, pp. 289-293: IEEE.
[21] A. Radford, L. Metz, and S. Chintala, "Unsupervised representation learning with deep convolutional generative adversarial networks," arXiv preprint arXiv:1511.06434, 2015.
[22] P. Isola, J.-Y. Zhu, T. Zhou, and A. A. Efros, "Image-to-image translation with conditional adversarial networks," arXiv preprint, 2017.
[23] N. E. Huang and Z. Wu, "A review on Hilbert‐Huang transform: Method and its applications to geophysical studies," Reviews of geophysics, vol. 46, no. 2, 2008.
[24] A. v. d. Oord, N. Kalchbrenner, and K. Kavukcuoglu, "Pixel recurrent neural networks," arXiv preprint arXiv:1601.06759, 2016.
[25] G. E. Hinton, S. Osindero, and Y.-W. Teh, "A fast learning algorithm for deep belief nets," Neural computation, vol. 18, no. 7, pp. 1527-1554, 2006.
[26] Y. LeCun et al., "Comparison of learning algorithms for handwritten digit recognition," in International conference on artificial neural networks, 1995, vol. 60, pp. 53-60: Perth, Australia.
[27] N. Srivastava, G. Hinton, A. Krizhevsky, I. Sutskever, and R. Salakhutdinov, "Dropout: a simple way to prevent neural networks from overfitting," The Journal of Machine Learning Research, vol. 15, no. 1, pp. 1929-1958, 2014.
[28] 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.
[29] S. Ioffe and C. Szegedy, "Batch normalization: Accelerating deep network training by reducing internal covariate shift," arXiv preprint arXiv:1502.03167, 2015.
[30] J. Duchi, E. Hazan, and Y. Singer, "Adaptive subgradient methods for online learning and stochastic optimization," Journal of Machine Learning Research, vol. 12, no. Jul, pp. 2121-2159, 2011.
[31] M. D. Zeiler, "ADADELTA: an adaptive learning rate method," arXiv preprint arXiv:1212.5701, 2012.
[32] T. Tieleman and G. Hinton, "Lecture 6.5-rmsprop: Divide the gradient by a running average of its recent magnitude," COURSERA: Neural networks for machine learning, vol. 4, no. 2, pp. 26-31, 2012.
[33] D. P. Kingma and J. Ba, "Adam: A method for stochastic optimization," arXiv preprint arXiv:1412.6980, 2014.
[34] 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.
[35] K. Simonyan and A. Zisserman, "Very deep convolutional networks for large-scale image recognition," arXiv preprint arXiv:1409.1556, 2014.
[36] K. He, X. Zhang, S. Ren, and J. Sun, "Deep residual learning for image recognition," in Proceedings of the IEEE conference on computer vision and pattern recognition, 2016, pp. 770-778.
[37] G. Huang, Y. Sun, Z. Liu, D. Sedra, and K. Q. Weinberger, "Deep networks with stochastic depth," in European conference on computer vision, 2016, pp. 646-661: Springer.
[38] C. Szegedy et al., "Going deeper with convolutions," in Proceedings of the IEEE conference on computer vision and pattern recognition, 2015, pp. 1-9.
[39] V. Badrinarayanan, A. Kendall, and R. Cipolla, "Segnet: A deep convolutional encoder-decoder architecture for image segmentation," IEEE transactions on pattern analysis and machine intelligence, vol. 39, no. 12, pp. 2481-2495, 2017.
[40] O. Ronneberger, P. Fischer, and T. Brox, "U-net: Convolutional networks for biomedical image segmentation," in International Conference on Medical image computing and computer-assisted intervention, 2015, pp. 234-241: Springer.
[41] D. P. Kingma and M. Welling, "Auto-encoding variational bayes," arXiv preprint arXiv:1312.6114, 2013.
[42] I. Goodfellow et al., "Generative adversarial nets," in Advances in neural information processing systems, 2014, pp. 2672-2680.
[43] M. Arjovsky and L. Bottou, "Towards principled methods for training generative adversarial networks. arXiv," 2017.
[44] M. Arjovsky, S. Chintala, and L. Bottou, "Wasserstein gan," arXiv preprint arXiv:1701.07875, 2017.
[45] V. Mnih et al., "Asynchronous methods for deep reinforcement learning," in International conference on machine learning, 2016, pp. 1928-1937.
[46] T. Karras, T. Aila, S. Laine, and J. Lehtinen, "Progressive growing of gans for improved quality, stability, and variation," arXiv preprint arXiv:1710.10196, 2017.
[47] M. Mirza and S. Osindero, "Conditional generative adversarial nets," arXiv preprint arXiv:1411.1784, 2014.
[48] E. L. Denton, S. Chintala, and R. Fergus, "Deep generative image models using a laplacian pyramid of adversarial networks," in Advances in neural information processing systems, 2015, pp. 1486-1494.
[49] C. Dong, C. C. Loy, K. He, and X. Tang, "Learning a deep convolutional network for image super-resolution," in European conference on computer vision, 2014, pp. 184-199: Springer.
[50] V. L. D. U. A. Vedaldi, "Instance Normalization: The Missing Ingredient for Fast Stylization," arXiv preprint arXiv:1607.08022, 2016.
[51] G. Rilling, P. Flandrin, P. Gonçalves, and J. M. Lilly, "Bivariate empirical mode decomposition," IEEE signal processing letters, vol. 14, no. 12, pp. 936-939, 2007.
[52] P. Kruizinga and N. Petkov, "Nonlinear operator for oriented texture," IEEE Transactions on image processing, vol. 8, no. 10, pp. 1395-1407, 1999.