协同感知损失和注意力机制的低剂量CT去噪
LOW-DOSE CT DENOISING BASED ON THE SYNERGISTIC NETWORK BETWEEN PERCEPTION LOSS AND ATTENTION MECHANISM
邓杰航 1吕伟考 1钟韬 1顾国生 1丁磊1
作者信息
- 1. 广东工业大学计算机学院 广东广州 510006
- 折叠
摘要
由于存在特有的量子噪声,低剂量CT去噪是一项艰巨的任务.当前主流研究使用的深度学习方法存在定性和定量指标不匹配的问题,实验结果的定量指标高,但视觉效果不好.为此,提出一种感知损失和注意力机制的低剂量CT协同去噪网络.该协同机制能够在保证视觉效果的基础上明显改善现有方法定量指标低的问题.模型在网络输入端还引入8方向的边缘检测层,可提取更丰富的纹理与结构信息,进一步提升了网络效果.针对体模数据集和真实临床数据集的实验对比结果表明,该方法相比主流工作,在视觉感受和PSNR以及SSIM指标上,均有更优异表现.
Abstract
Due to the unique quantum noise,low-dose CT denoising is a difficult task.Most of the deep learning methods for denoising have the problem of mismatch between visual inspection and quantitative indicators in which the quantitative indicators of experimental results are high,but the visual inspection is not well.Therefore,this paper proposes a low-dose CT synergistic denoising network between perceptual loss and attention mechanism.This synergistic mechanism could significantly improve the problem of low quantitative indicators in existing methods.The model also introduced an 8-direction edge detection layer to the beginning of the network,which could extract richer textures and structure information,further improving the network performance.The experimental comparison results based on the phantom data sets and the clinical data sets show that the proposed method has better performance in visual inspection,PSNR and SSIM indicators than the state-of-the-art methods.
关键词
低剂量CT/注意力机制/感知损失/去噪/多方向边缘提取Key words
eywordsLow-dose CT/Attention mechanism/Perception loss/Denoising/Multi-directional edge extraction引用本文复制引用
基金项目
国家自然科学基金项目(61202267)
广州市科技计划项目(201902020007)
广州市科技计划项目(202007010004)
广州市科技计划项目(201807010058)
出版年
2024
国家科技期刊平台
NSTL
维普
万方数据