臺灣博碩士論文加值系統

頻域光學同調斷層掃描(Fourier Domain Optical Coherence Tomography, FD-OCT)是一種高解析度的成像技術，但常面臨共軛偽影、DC偽影和自相關偽影等問題導致成像失真。近年來，深度學習技術展現出優異的影像處理能力，其中生成對抗網路(Generative Adversarial Network, GAN)被廣泛應用於影像生成與轉換任務，但傳統GAN在處理複雜多層樣品FD-OCT影像時，效能仍有改善空間。
本文提出一種結合自動編碼器和條件生成對抗網路的新架構，並運用至多層全域式FD-OCT系統以消除偽影。首先藉由自動編碼器將原始FD-OCT數據進行潛在特徵擷取，再利用這些潛在特徵來訓練專為多層樣本設計的條件生成對抗網路，以獲得無偽影之潛在特徵。最後，再透過自動編碼器將無偽影之潛在特徵進行解碼，以重建高品質的FD-OCT影像。實驗結果顯示，本論文提出的網路架構相較於僅使用生成對抗網路，在複雜多層樣本的偽影消除和影像重建方面有顯著的改善。

Fourier Domain Optical Coherence Tomography (FD-OCT) is a high-resolution imaging technique, but often faces issues such as conjugate artifacts, DC artifacts, and autocorrelation artifacts, leading to distorted imaging. In recent years, deep learning techniques have demonstrated outstanding image processing capabilities, among which Generative Adversarial Networks (GANs) have been widely applied to image generation and transformation tasks. However, the performance of traditional GANs still has room for improvement when handling complex multi-layer sample FD-OCT images.
This proposal proposes a new architecture that combines an autoencoder and a conditional generative adversarial network to eliminate the artifacts in a multilayer full-range FD-OCT system. First, the raw FD-OCT data is processed by an autoencoder for latent feature extraction, and these latent features are then used to train a conditional generative adversarial network explicitly designed for multi-layer samples to obtain artifact-free latent features. Finally, the autoencoder decodes the artifact-free latent features to reconstruct high-quality FD-OCT images. Experimental results show that compared to using only a generative adversarial network, the proposed network architecture significantly improves artifact removal and image reconstruction for complex multi-layer samples.

摘要...i
Abstract...ii
誌謝...iii
目錄...iv
表目錄...vi
圖目錄...vii
1 第一章 緒論...1
1.1 前言...1
1.2 動機...2
1.3 論文架構...3
2 第二章 研究背景及相關知識...4
2.1 頻域光學同調斷層掃描...4
2.2 條件式生成對抗網路...5
2.3 基於條件式生成對抗網路(Conditional Generative Adversarial Net, cGAN)的頻域光學同調斷層掃描...6
2.3.1 U-Net...7
2.3.2 PatchGAN...8
2.3.3 卷積神經網路(Convolutional Neural Network, CNN)...8
2.3.4 類神經網路介紹...10
2.3.5 激活函數（Activation Function）...11
2.3.6 損失函數（Loss Function）...13
2.3.7 最佳化器(Optimizer)...14
3 第三章 研究內容與方法...16
3.1 FD-OCT資料集生成與處理...16
3.2 輸入資料型態...16
3.2.1 傳統的FD-OCT訓練資料...17
3.2.2 無偽影FD-OCT訓練資料...17
3.3 實驗架構...18
3.3.1 Autoencoder...18
3.3.2 AE-cGAN架構...19
3.4 調整Encoder和Decoder的輸出資料大小...20
3.5 驗證成功率說明...22
4 第四章 模擬結果與討論...26
4.1 cGAN...26
4.2 Encoder Channel 256...28
4.2.1 Autoencoder Size256...28
4.2.2 Autoencoder Size128...29
4.2.3 Autoencoder Size64...29
4.2.4 Autoencoder Size32...30
4.2.5 Autoencoder Size64addLayer...31
4.2.6 Autoencoder Size128addLayer...32
4.3 Encoder Channel 512...33
4.3.1 Autoencoder Size 256...33
4.3.2 Autoencoder Size128...33
4.3.3 Autoencoder Size64...34
4.3.4 Autoencoder Size32...35
4.3.5 Autoencoder Size64addLayer...36
4.3.6 Autoencoder Size128addLayer...37
4.4 實驗結果比較...38
5 第五章 結論...41
5.1 論文結論...41
5.2 未來研究...41
參考文獻...42
Extended Abstract...44

A. Shrestha and A. Mahmood, "Review of Deep Learning Algorithms and Architectures," IEEE Access, vol. 7, pp. 53040-53065, 2019, doi: 10.1109/ACCESS.2019.2912200.
[2]Y. P. Liu, W. Zuo, R. Liang, H. Sun, and Z. Li, "Prototype-Guided Autoencoder for OCT-Based Fingerprint Presentation Attack Detection," IEEE Transactions on Information Forensics and Security, vol. 18, pp. 3461-3475, 2023, doi: 10.1109/TIFS.2023.3282386.
[3] A. ElTanboly et al., "Diabetic Retinopathy Early Detection Based on OCT and OCTA Feature Fusion," in 2018 IEEE International Symposium on Signal Processing and Information Technology (ISSPIT), 6-8 Dec. 2018 2018, pp. 607-611, doi: 10.1109/ISSPIT.2018.8642744.
[4] X. Liu, R. Wang, and J. Tang, "Graph Convolutional Networks with Feature Enhancement for Choroidal Neovascularization Segmentation in OCT Images," in 2023 IEEE International Conference on Systems, Man, and Cybernetics (SMC), 1-4 Oct. 2023 2023, pp. 3986-3990, doi: 10.1109/SMC53992.2023.10394074.
[5] D. W. K. Wong, A. P. Yow, B. Tan, Y. Xinwen, J. Chua, and L. Schmetterer, "Localization of Anatomical Features in Vascular-enhanced Enface OCT Images," in 2020 42nd Annual International Conference of the IEEE Engineering in Medicine & Biology Society (EMBC), 20-24 July 2020 2020, pp. 1875-1878, doi: 10.1109/EMBC44109.2020.9175868.
[6] D. Ravenscroft et al., "Learning feature extractors for AMD classification in OCT using convolutional neural networks," in 2017 25th European Signal Processing Conference (EUSIPCO), 28 Aug.-2 Sept. 2017 2017, pp. 51-55, doi: 10.23919/EUSIPCO.2017.8081167.
[7]I. Goodfellow et al., "Generative adversarial networks," Commun. ACM, vol. 63, no. 11, pp. 139–144, 2020, doi: 10.1145/3422622.
[8] I. J. Goodfellow et al., "Generative Adversarial Nets," in Neural Information Processing Systems, 2014.
[9]P. Isola, J.-Y. Zhu, T. Zhou, and A. A. Efros, "Image-to-Image Translation with Conditional Adversarial Networks," 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pp. 5967-5976, 2016.
[10]J.-Y. Zhu, T. Park, P. Isola, and A. A. Efros, "Unpaired Image-to-Image Translation Using Cycle-Consistent Adversarial Networks," 2017 IEEE International Conference on Computer Vision (ICCV), pp. 2242-2251, 2017.
[11]M. Mirza and S. Osindero, "Conditional Generative Adversarial Nets," ArXiv, vol. abs/1411.1784, 2014.
[12]Y. Zhou et al., "Speckle Noise Reduction for OCT Images Based on Image Style Transfer and Conditional GAN," IEEE Journal of Biomedical and Health Informatics, vol. 26, no. 1, pp. 139-150, 2022, doi: 10.1109/JBHI.2021.3074852.
[13] J. Kugelman, D. Alonso-Caneiro, S. A. Read, S. J. Vincent, and M. J. Collins, "Semi-supervised learning with cross-localisation in shared GAN latent space for enhanced OCT data augmentation," in 2022 International Conference on Digital Image Computing: Techniques and Applications (DICTA), 30 Nov.-2 Dec. 2022 2022, pp. 1-7, doi: 10.1109/DICTA56598.2022.10034570.
[14] D. S. Ujjwal et al., "GAN-based OCT Image Quality Enhancement: Mapping from Low Quality Cirrus OCT to High Quality EDI OCT," in 2023 IEEE 20th India Council International Conference (INDICON), 14-17 Dec. 2023 2023, pp. 1277-1281, doi: 10.1109/INDICON59947.2023.10440843.
[15] M. J. Hasan, M. S. Alom, U. Fatema, and M. F. Wahid, "Deep Learning Based Retinal OCT Image Denoising using Generative Adversarial Network," in 2021 International Conference on Automation, Control and Mechatronics for Industry 4.0 (ACMI), 8-9 July 2021 2021, pp. 1-6, doi: 10.1109/ACMI53878.2021.9528116.
[16]C. M. Huang, E. Wijanto, and H. C. Cheng, "Applying a Pix2Pix Generative Adversarial Network to a Fourier-Domain Optical Coherence Tomography System for Artifact Elimination," IEEE Access, vol. 9, pp. 103311-103324, 2021, doi: 10.1109/ACCESS.2021.3098865.
[17]J. A. Izatt and M. A. Choma, "Theory of Optical Coherence Tomography," in Optical Coherence Tomography: Technology and Applications, W. Drexler and J. G. Fujimoto Eds. Berlin, Heidelberg: Springer Berlin Heidelberg, 2008, pp. 47-72.
[18]O. Ronneberger, P. Fischer, and T. Brox, "U-Net: Convolutional Networks for Biomedical Image Segmentation," ArXiv, vol. abs/1505.04597, 2015.
[19] C. Yu, Y. H. Li, Y. H. Tseng, and Y. M. Chen, "Low-Latency LeNet-5 Architecture for Handwritten Digit Recognition," in 2023 IEEE International Conference on Consumer Electronics-Asia (ICCE-Asia), 23-25 Oct. 2023 2023, pp. 1-3, doi: 10.1109/ICCE-Asia59966.2023.10326419.
[20]C.-Y. Liou, W.-C. Cheng, J.-W. Liou, and D.-R. Liou, "Autoencoder for words," Neurocomputing, vol. 139, pp. 84-96, 2014/09/02/ 2014, doi: https://doi.org/10.1016/j.neucom.2013.09.055.