近年來，口腔癌是好發癌症中五年存活率較低之一且其發生及死亡率皆為台灣男性癌症第4名，而口腔癌在初期做小手術後五年存活率約為77%。因此，口腔癌的提早診斷與治療是重要的一個議題，為了達到目的，許多口腔癌篩檢方法已經發展出來並實際應用於社會大眾，其中利用光學的方法來篩檢口腔癌有即時診斷和非侵入式的優點，其原理為利用口腔自體螢光在正常組織和惡性組織不同細胞代謝下會呈現不同的強度反應，而先前的研究中我們已經開發出自體螢光應用於口腔癌檢測系統，此系統包括收集自體螢光影像的手持儀器、口腔癌自動偵測演算法、雲端統整資料庫，並已實際用於臨床收案和測試。然而分析資料時必須手動圈選影像中該部位，並依照各個口腔部位來做後續分析，對於大量篩檢上，分類影像部位以及圈選每張影像中該部位透過人工的方式會非常的耗時和不便，近年來隨著GPU效能的增進、大數據的蓬勃發展，同時也帶動了深度學習的興起。本研究希望透過深度學習中的深度卷積神經網路來建立自動分類與分割系統，並且將此自動化系統加入於目前的口腔癌檢測系統，目的為能夠將大量的篩檢影像先依照部位分類並標記，接著依照標記來對該部位進行分割，使口腔癌檢測系統更有效率以及準確地對社會大眾進行口腔癌篩檢。
為了建立出自動分類與分割系統，本研究透過兩種不同目的的深度神經網路架構來組成此系統：第一種為分類網路架構，基於LeNet網路架構建立出最佳化分類模型來對影像進行部位的分類；第二種為分割網路架構，基於U-Net網路架構建立出最佳化分割網路模型來對影像進行部位的分割。本研究中更透過不同的技術來優化兩種網路架構的準確率並縮減訓練神經網路的時間。同時，研究中也提出來新的深度神經網路架構，結合U-Net和Inception block的網路架構來進行影像分割，最後比較多種分類與分割的結果。經實驗結果顯示，本論文所建立的自動化分類與分割系統中，分類影像中部位的準確率可達到93%，分割頰黏膜和舌頭的準確率分別可達到79.1%和90%。

In recent years, oral cancer is one of the five-year low survival rate in cancer. Its occurrence and mortality rate both are the fourth place in Taiwan. However, the five-year survival rate in the early oral cancer surgery can reach 77 percent. Therefore, the early diagnosis and treatment of oral cancer is an important issue. In order to achieve the goal, many cancer screening methods have been developed and applied to the community. Which uses optical methods to screen oral cancer with immediate diagnosis and non-invasive advantages. The principle of the use oral autofluorescence in normal tissue and malignant tissue metabolism under different cells will show different intensity response. In our previous research, using autofluorescence in the oral cancer detection system have been developed. The system includes a hand-held instrument for collecting autofluorescence images and automatic detection algorithm for oral cancer, which has been actually used in clinical cases and testing. However, the position of image still needs to be manually selected. And according to the various position to do follow-up analysis. For the large number of screening, classification of oral position on images and position selection of each image are time-consuming and inconvenient. In recent years, with the GPU performance improvement and development of big data, which led to the rise of the deep learning. This research purpose is that establish the automatic classification and segmentation system through the deep convolutional neural networks, and join this automatic system to the current oral cancer detection system. The purpose is to be able to classify a large number of screening images and segment position of images, which can let the oral cancer detection system more efficient and accurate to oral cancer screening.
In order to establish an automatic classification and segmentation system, the research consists of two different purpose of deep convolutional neural networks architecture to compose the system. The first is a classification network architecture, which based on LeNet network architecture to build an optimal classification model to classify the position of images. The second is a segmentation network architecture, which based on U-Net network architecture to build an optimal segmentation network model to segment the position of images；Moreover, we adopt different technologies to optimize the accuracy of the two network architecture and reduce the training time of the neural network. In this research, we also propose a new deep neural network architecture, which combine U-Net and Inception block network architecture for image segmentation. Furthermore, we compare the results of a variety of classification and segmentation. The experimental results show accuracy rate of classification position can reach 93 percent. The accuracy rate of buccal and tongue segmented can reach 79.1 percent and 90 percent.

[1] 口腔癌防治，台灣衛生福利部國民健康署
http://www.hpa.gov.tw/Pages/Detail.aspx?nodeid=204&pid=1118
[2] 口腔癌臨床診療指引，國家衛生研究院
http://www.nhri.org.tw/NHRI_ADM/userfiles/file/tcog/100oralpg.pdf
[3] 口腔癌檢資訊，國民健康署
http://oralscreen.hpa.gov.tw/announcement.aspx?
[4] 口腔癌篩檢，台北榮民總醫院
http://www.aoms.url.tw/joomla/attachments/article/16/chapter7.pdf
[5] 螢光光譜與影像在口腔癌前病變之診斷，科學發展月刊
http://catalog.digitalarchives.tw/item/00/61/83/55.html
[6] H-R Hung, “Application of Autofluorescence Imaging on Oral Cancer Screening”
[7] Chen, Chen Tin, et al. "Light-induced fluorescence spectroscopy: a potential diagnostic tool for oral neoplasia." Proceedings of the National Science Council, Republic of China. Part B, Life sciences 20.4 (1996): 123-130.
[8] Chen, Chin‐Tin, et al. "Autofluorescence in normal and malignant human oral tissues and in DMBA‐induced hamster buccal pouch carcinogenesis." Journal of oral pathology & medicine 27.10 (1998): 470-474.
[9] T-Y Huang, “Application of Autofluorescence Imaging On Clinical Oral Cancer Screening”
[10] M-H Chang, “Texture Feature Analysis for Oral Cancer Detection”
[11] C-H Chan, “Using Deep Convolutional Neural Networks for Oral Cancer Detection Based on Texture Feature Images”
[12] Yann LeCun, Leon Bottou, Yoshua Bengio, and Patrick Haffner.”Gradient-based learning applied to document recognition”
[13] Ronneberger, Olaf, Philipp Fischer, and Thomas Brox,“U-net: Convolutional networks for biomedical image segmentation.”International Conference on Medical Image Computing and Computer-Assisted Intervention. Springer International Publishing, 2015.
[14] Szegedy, Christian, et al. "Going deeper with convolutions." Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. 2015.
[15] Wikipedia. (2017) Deep learning [online] Available at:
https://en.wikipedia.org/wiki/Deep_learning [Accessed May 2017].
[16] Mikolov, Tomas, et al. "Recurrent neural network based language model." Interspeech. Vol. 2. 2010.
[17] Hinton, Geoffrey E. "Deep belief networks." Scholarpedia 4.5 (2009): 5947.
[18] Fischer, Asja, and Christian Igel. "An introduction to restricted Boltzmann machines." Progress in Pattern Recognition, Image Analysis, Computer Vision, and Applications (2012): 14-36.
[19] Bengio, Yoshua. "Learning deep architectures for AI." Foundations and trends® in Machine Learning 2.1 (2009): 1-127.
[20] Arnold, Ludovic, et al. "An introduction to deep learning." European Symposium on Artificial Neural Networks (ESANN). 2011.
[21] Lee, Honglak. "Tutorial on deep learning and applications." NIPS 2010 Workshop on Deep Learning and Unsupervised Feature Learning. 2010.
[22] Deng, Li, and Dong Yu. "Deep learning: methods and applications." Foundations and Trends® in Signal Processing 7.3–4 (2014): 197-387.
[23] Wikipedia. (2017) AlphaGo [online] Available at:
https://zh.wikipedia.org/wiki/AlphaGo [Accessed May 2017].
[24] Garcia-Garcia, Alberto, et al. "A Review on Deep Learning Techniques Applied to Semantic Segmentation." arXiv preprint arXiv:1704.06857 (2017).
[25] Long, Jonathan, Evan Shelhamer, and Trevor Darrell. "Fully convolutional networks for semantic segmentation." Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. 2015.
[26] Simonyan, Karen, and Andrew Zisserman. "Very deep convolutional networks for large-scale image recognition." arXiv preprint arXiv:1409.1556 (2014).
[27] He, Kaiming, et al. "Deep residual learning for image recognition." Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. 2016.
[28] Krizhevsky, Alex, Ilya Sutskever, and Geoffrey E. Hinton. "Imagenet classification with deep convolutional neural networks." Advances in neural information processing systems. 2012.
[29] Badrinarayanan, Vijay, Alex Kendall, and Roberto Cipolla. "Segnet: A deep convolutional encoder-decoder architecture for image segmentation." arXiv preprint arXiv:1511.00561 (2015).
[30] Huang, Gao, et al. "Densely connected convolutional networks." arXiv preprint arXiv:1608.06993 (2016).
[31] Jégou, Simon, et al. "The One Hundred Layers Tiramisu: Fully Convolutional DenseNets for Semantic Segmentation." arXiv preprint arXiv:1611.09326 (2016).
[32] Srivastava, Nitish, et al. "Dropout: A simple way to prevent neural networks from overfitting." The Journal of Machine Learning Research 15.1 (2014): 1929-1958.
[33] Ioffe, Sergey, and Christian Szegedy. "Batch normalization: Accelerating deep network training by reducing internal covariate shift." arXiv preprint arXiv:1502.03167 (2015).
[34] Xu, Bing, et al. "Empirical evaluation of rectified activations in convolutional network." arXiv preprint arXiv:1505.00853 (2015).
[35] Clevert, Djork-Arné, Thomas Unterthiner, and Sepp Hochreiter. "Fast and accurate deep network learning by exponential linear units (elus)." arXiv preprint arXiv:1511.07289 (2015).
[36] Kingma, Diederik, and Jimmy Ba. "Adam: A method for stochastic optimization." arXiv preprint arXiv:1412.6980 (2014).
[37] Cs231n.github.io. (2017). CS231n Convolutional Neural Networks for Visual Recognition. [online] Available at : http://cs231n.github.io/
[38] Suzuki, Satoshi. "Topological structural analysis of digitized binary images by border following." Computer vision, graphics, and image processing 30.1 (1985): 32-46.
[39] Farah, Camile S., et al. "Efficacy of tissue autofluorescence imaging (VELScope) in the visualization of oral mucosal lesions." Head & neck 34.6 (2012): 856-862.
[40] Dean, Jeff. "Large-scale deep learning for intelligent computer systems." BayLearn keynote speech (2015).