摘要
在工业CT中,使用含噪的投影数据进行重建,会导致重建图像中的噪声增加,降低重建图像的信噪比(SNR)。当投影数据质量较差时,经典的降噪和重建算法无法有效的去除噪声。为了提高CT重建图像的质量,本文提出一种基于深度学习的去噪方法。该方法将通道注意力机制模块嵌入到解码器阶段,通过自适应地调整通道的权重,从而提高网络在去噪过程更好地保留图像的结构细节。实验结果表明,所提方法能够显著地去除噪声并有效保护边缘细节,且在视觉效果和定量指标结果上都要优于其他对比方法。
In industrial computed tomography(CT),using noisy projection data for reconstruction increases the noise in the reconstructed image and reduces the signal-to-noise ratio(SNR).When the quality of projection data is poor,classical denoising and reconstruction algorithms are ineffective in removing the noise.To improve the quality of low signal-to-noise CT reconstructed images,this study proposes a deep learning-based denoising method.The method integrates squeeze-and-excitation blocks into the decoder phase and adaptively adjusts the weights of the channels to better preserve structural details during the denoising process.Experimental results demonstrate that the proposed method significantly reduces the noise and effectively preserves edge details,outperforming other comparative methods in both visual quality and quantitative values.
作者
何雨
王成祥
余维
HE Yu;WANG Chengxiang;YU Wei(School of Mathematical Sciences,Chongqing Normal University,Chongqing 401331,China;Key Laboratory of Optoelectronic Sensing and Intelligent Control,Hubei University of Science and Technology,Xianning 437100,China;School of Biomedical Engineering and Imaging,Medical College,Hubei University of Science and Technology,Xianning 437100,China)
出处
《CT理论与应用研究(中英文)》
2025年第4期543-550,共8页
Computerized Tomography Theory and Applications
基金
国家自然科学基金(多段有限角光子计数能谱CT图像重建方法研究(62371184))
重庆市教育委员会科技研究计划(基于多目标优化的CL成像图像重建算法研究(KJQN202300535))
咸宁市科技人才服务专项(结合特征学习的有限角CT成像方法研究(2024RCFW007))
重庆市自然科学基金(基于无监督学习的低剂量CT图像重建算法研究(CSTB2023NSCQ-MSX0148))。
