%0 Journal Article %T 分层极化注意力网络的OCT视网膜图像降噪研究
Research on De-Noising of OCT Retinal Images in Polarified U-Net %A 吴桐 %A 陈明惠 %A 胡亚兰 %A 周旭东 %A 马文飞 %A 陈思思 %A 李家昱 %J Modeling and Simulation %P 419-429 %@ 2324-870X %D 2025 %I Hans Publishing %R 10.12677/mos.2025.141039 %X 目的：光学相干断层成像技术(OCT)是视网膜疾病诊断中最常用的方法之一，由于OCT设备的光波多次前向和后向散射引起散斑噪声，斑点噪声的存在一般会模糊细微但重要的形态学细节，最终影响临床诊断。本文提出一种新的基于CNN的降噪网络，在对OCT图像进行降噪处理的同时，使得图像在保留空间结构细节的基础上能展示更多的信息，提高图像质量。方法：提出了一种基于改进的编码器–解码器(U-Net)分层网络的新型OCT图像降噪网络——DRS-Unet，加入由密集连接和局部残差连接组成的密集局部残差块和极化注意力机制，并且在保留图像空间结构细节的基础上降低图像噪声。结果：实验表明，DRS-Unet在客观评价指标方面要优于传统方法与其他深度学习方法，使用DRS-Unet对OCT图像进行降噪比其他深度学习降噪模型提高了2.7%左右，且拥有更强的边缘保持和泛化能力。证明该方法可以有效地保留OCT图像中的边缘结构信息，同时可以有效地抑制噪声，提高图像质量。
OBJECTIVE: Optical coherence tomography (OCT) is one of the most commonly used methods in the diagnosis of retinal diseases, due to the scattering noise caused by multiple forward and backward scattering of light waves from the OCT device, the presence of speckle noise generally blurs subtle but important morphological details, which ultimately affects the clinical diagnosis. In this paper, a new CNN-based noise reduction network is proposed to improve the image quality by enabling the image to display more information while preserving the spatial structural details while performing noise reduction processing on OCT images. METHODS: A novel OCT image noise reduction network based on improved encoder-decoder (U-Net) hierarchical network, DRS-Unet, is proposed to incorporate a dense local residual block consisting of dense connections and local residual connections and a polarized attention mechanism, and to reduce the image noise based on the preservation of the spatial structural details of the image. CONCLUSION: Experiments show that DRS-Unet outperforms traditional methods and other deep learning methods in terms of objective evaluation indexes, and noise reduction of OCT images using DRS-Unet improves by about 2.7% compared with other deep learning noise reduction models, and possesses stronger edge preservation and generalization abilities. It has been proved that the method can effectively retain the edge structure information in OCT images, and at the same time, it can effectively suppress the noise and improve the image quality. %K 光学相干断层成像， %K 图像降噪， %K 编码器– %K 解码器， %K 注意力机制
Optical Coherence Tomography %K Image Denoising %K Encoder-Decoder %K Attentional Mechanism %U http://www.hanspub.org/journal/PaperInformation.aspx?PaperID=105095