臺灣博碩士論文加值系統

本研究旨在探索並比較兩種方法，即Pix2pix和Vislab，使用深度學習來消除光學同調斷層掃描（Optical Coherence Tomography,OCT）中的單層和多層樣品偽影。在Pix2pix方法中，我們設計了一個基於條件GAN的模型，將真實的OCT圖像作為輸入，並以對應的無偽影圖像作為目標。通過對抗訓練和像素級別的差異損失，我們的模型學習如何生成消除多層樣品偽影的圖像。另一方面，Vislab方法利用多種AI深度學習技術來偵測實像和虛像的位置，只要透過簡單的操作就可以訓練好所需的模型，用訓練好的模型來解決樣品偽影的問題。我們使用大量的真實OCT圖像和相應的多層樣品偽影圖像來訓練和評估這兩種方法。實驗結果顯示，Pix2pix方法能夠生成具有高質量的消除偽影的OCT圖像。Vislab方法則通過物件偵測功能在有偽影的圖像中找出實像。綜合比較，兩種方法都在消除多層樣品偽影方面取得了顯著的改善。Pix2pix方法通過學習圖像生成，能夠生成無偽影的OCT圖像。Vislab方法則運用了視覺分析技術，以更直觀的方式消除偽影。總體而言，我們的研究表明使用生成對抗網路在光學同調斷層掃描中消除多層樣品偽影是可行且有效的。這兩種方法對於改善OCT圖像質量並提高診斷準確性具有潛力，並可在臨床和科學研究中得到廣泛應用。

This study aims to explore and compare two methods, namely Pix2pix and Vislab, that utilize Generative Adversarial Networks (GAN) to eliminate single-layer and multi-layer sample artifacts in Optical Coherence Tomography (OCT) scans.In the Pix2pix method, we designed a conditional GAN-based model that takes real OCT images as input and uses corresponding artifact-free images as targets. Through adversarial training and pixel-level difference loss, our model learns to generate images that eliminate multi-layer sample artifacts.On the other hand, the Vislab method utilizes various AI deep learning techniques to detect the positions of real and virtual images. By training the required models through simple operations, this method addresses the issue of sample artifacts.We trained and evaluated both methods using a large number of real OCT images and their corresponding multi-layer sample artifact images. The experimental results show that the Pix2pix method is capable of generating OCT images with high-quality artifact removal. The Vislab method identifies real images in artifact-containing images through object detection.In summary, both methods have significantly improved the elimination of multi-layer sample artifacts. The Pix2pix method achieves artifact-free OCT image generation through image learning, while the Vislab method utilizes visual analysis techniques for a more intuitive artifact removal approach.Overall, our research demonstrates the feasibility and effectiveness of using GAN to eliminate multi-layer sample artifacts in Optical Coherence Tomography. Both methods have the potential to enhance OCT image quality and improve diagnostic accuracy, with broad applications in clinical and scientific research.

摘要 ....i
Abstract ....ii
誌謝 ....iii
目錄 ....iv
表目錄 ....v
圖目錄 ....vi
第一章 緒論 ....1
1.1 前言 ....1
1.2 研究動機和目的 ....2
1.3 論文架構 ....3
1.4 文獻探討 ....3
第二章 光學同調斷層掃描技術背景與原理介紹 ....9
2.1 光學同調斷層掃描 ....9
2.2 時域光學同調斷層掃描 ....12
2.3 頻域光學同調斷層掃描 ....13
第三章 生成對抗網路原理 ....18
3.1 生成對抗網路介紹 ....18
3.2 Pix2pix介紹 ....19
3.3 生成器架構 ....20
3.4 鑑別器架構 ....21
3.5 卷積層結構 ....22
3.6 Vislab ....23
第四章 實驗架構與結果 ....25
4.1 生成單層和多層的FDOCT圖片 ....25
4.2 實驗一Pix2pix GAN ....28
4.3 實驗二Vislab ....32
4.4 Pix2pix跟Vislab實驗結果討論 ....37
第五章 結論 ....38
參考文獻 ....39
Extended Abstract ....41

[1]D. Huang et al., "Optical coherence tomography," (in eng), Science, vol. 254, no. 5035, pp. 1178-81, Nov 22 1991.
[2]W. Drexler, M. Liu, A. Kumar, T. Kamali, A. Unterhuber, and R. A. Leitgeb, "Optical coherence tomography today: speed, contrast, and multimodality," (in eng), J Biomed Opt, vol. 19, no. 7, p. 071412, 2014.
[3]J.-F Huang, H.-C Cheng, C.-W Chang, "Group delay based phase shifting algorithm on frequency domain optical coherence tomography," in 2008 10th Anniversary International Conference on Transparent Optical Networks, 2008, vol. 4, pp. 367-370.
[4]H.-C. Cheng, J.-F. Huang, and Y.-H. Hsieh, "Numerical analysis of one-shot full-range FD-OCT system based on orthogonally polarized light," Optics Communications, vol. 282, pp. 3040-3045, 07/15 2009.
[5]A. Ozcan, M. J. F. Digonnet, and G. S. Kino, "Minimum-phase-function-based processing in frequency-domain optical coherence tomography systems," Journal of the Optical Society of America A, vol. 23, no. 7, pp. 1669-1677, 2006/07/01 2006.
[6]J. Lee, T. Yoon, and B. H. Lee, "Post-Processing Method for Image Reconstruction Enhancement in Integrating-Bucket-Based Full-Field Optical Coherence Tomography," vol. 10, no. 3, p. 830, 2020.
[7]Y. Zhu and W. Gao, "Single-shot wavelength-independent phase-shifting method for full-field optical coherence tomography," Applied Optics, vol. 58, no. 4, pp. 806-813, 2019/02/01 2019.
[8]X. Liu, M. Ke, X. Yao, J. Chua, L. Schmetterer, and B. Tan, "Stable complex conjugate artifact removal in OCT using circularly polarized light as reference," Optics Letters, vol. 45, no. 14, pp. 3977-3980, 2020/07/15 2020.
[9]K. T. Islam, S. Wijewickrema, and S. O. Leary, "Identifying Diabetic Retinopathy from OCT Images using Deep Transfer Learning with Artificial Neural Networks," in 2019 IEEE 32nd International Symposium on Computer-Based Medical Systems (CBMS), 2019, pp. 281-286.
[10]G. N. Girish, B. Thakur, S. R. Chowdhury, A. R. Kothari, and J. Rajan, "Segmentation of Intra-Retinal Cysts From Optical Coherence Tomography Images Using a Fully Convolutional Neural Network Model," IEEE Journal of Biomedical and Health Informatics, vol. 23, no. 1, pp. 296-304, 2019.
[11]J. Chi, C. Wu, X. Yu, P. Ji, and H. Chu, "Single Low-Dose CT Image Denoising Using a Generative Adversarial Network With Modified U-Net Generator and Multi-Level Discriminator," IEEE Access, vol. 8, pp. 133470-133487, 2020.
[12]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.
[13]J. Chhablani, T. Krishnan, V. Sethi, and I. Kozak, "Artifacts in optical coherence tomography," (in eng), Saudi J Ophthalmol, vol. 28, no. 2, pp. 81-7, Apr 2014.
[14]E. Götzinger, M. Pircher, R. A. Leitgeb, and C. K. Hitzenberger, "High speed full range complex spectral domain optical coherence tomography," Optics Express, vol. 13, no. 2, pp. 583-594, 2005/01/24 2005.
[15]P. Isola, J.-Y. Zhu, T. Zhou, and A. A. Efros, "Image-to-Image Translation with Conditional Adversarial Networks," presented at the 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2017. Available: https://doi.ieeecomputersociety.org/10.1109/CVPR.2017.632
[16]A. Mino and G. Spanakis, "LoGAN: Generating Logos with a Generative Adversarial Neural Network Conditioned on Color," in 2018 17th IEEE International Conference on Machine Learning and Applications (ICMLA), 2018, pp. 965-970.
[17]O. Ronneberger, P. Fischer, and T. Brox, U-Net: Convolutional Networks for Biomedical Image Segmentation. 2015, pp. 234-241.
[18]廖柏閎, "使用生成對抗網路技術於頻域光學同調斷層掃描圖像加強之研究," 碩士, 光電工程系光電與材料科技碩士班, 國立虎尾科技大學, 雲林縣, 2022.
[19]Q. Hao et al., "High signal-to-noise ratio reconstruction of low bit-depth optical coherence tomography using deep learning," (in eng), J Biomed Opt, vol. 25, no. 12, Nov 2020.
[20]U. Demir and G. Unal, "Patch-Based Image Inpainting with Generative Adversarial Networks," 03/20 2018.