| 摘 要: 光学相干断层扫描(Optical Coherence Tomography,OCT)图像在采集过程中常遭受噪声影响,导致成像结构模糊和失真。为有效消除OCT图像中的噪声并提高图像清晰度,基于CycleGAN网络架构,通过加入SE模块、DSC模块和优化损失函数,并采用无监督学习方式处理OCT图像。实验结果表明,这些方法在去噪和去模糊方面优于传统方法和其他无监督深度学习技术,尤其在图像清晰度方面,比传统降噪方法的PSNR值高了10%以上。本研究突显了深度学习技术在医学图像处理中的潜力与实用价值,为未来的临床应用提供了新的指导方法。 |
| 关键词: OCT图像去模糊 OCT图像去噪 无监督学习 CycleGAN网络 |
|
中图分类号:
文献标识码: A
|
| 基金项目: 贵州省科技计划项目(黔科合支撑[2022]一般184) |
|
| Deblurringand Denoising OCT Images Based on CycleGAN Network |
|
FAN Xinghong1, CHEN Xiangping1, GU Hao2, ZHAO Su2, JIANG Hao2
|
(1.College of Electrical Engineering, Guizhou University, Guiyang 550025, China;
2.Guizhou Medical University, Guiyang 550002, China)
18586378117@163.com; ee.xpchen@gzu.edu.cn; guhao@gmc.edu.cu; ximi520@sina.com; kidd5jh0@sina.com
|
| Abstract: Optical Coherence Tomography (OCT) images are often affected by noise during acquisition, leading to blurred structures and distortions. To effectively eliminate noise and enhance image clarity in OCT images, this study leverages the CycleGAN network architecture, incorporating SE (Squeeze-and-Excitation) modules, DSC(Depthwise Separable Convolution) modules, and optimized loss functions, while employing unsupervised learning for OCT image processing. Experimental results demonstrate that these methods outperform traditional approaches and other unsupervised deep learning techniques in denoising and deblurring, particularly in image clarity, achieving a PSNR value over 10% higher than conventional denoising methods. This study highlights the potential and practical value of deep learning in medical image processing, providing new guidance for future clinical applications. |
| Keywords: OCT image deblurring OCT image denoising unsupervised learning CycleGAN network |
