基于条件生成对抗网络的CT重建算法研究  

Research on CT Reconstruction Algorithm Based on Conditional Generative Adversarial Network

作　　者：商然 王明泉[1] 谢绍鹏 黄心玥 耿宇杰 SHANG Ran;WANG Mingquan;XIE Shaopeng;HUANG Xinyue;GENG Yujie(School of Information and Communication Engineering,North University of China,Taiyuan 030051,China)

机构地区：[1]中北大学信息与通信工程学院,太原030051

出　　处：《河南科学》2025年第2期157-163,共7页Henan Science

基　　金：国家自然科学基金(6171177)。

摘　　要：计算层析成像(CT)是一种被广泛应用于图像重建的方法。但由于CT成像系统的复杂性,在采集过程中不可避免地会受到噪声干扰导致重建图像的精度下降。为解决这一问题,构建了一种特定的条件生成对抗网络(cGAN),并在网络中引入了CBAM注意力机制,以便更高效地获取相关信息,从而实现包含噪声干扰的CT重建。为了验证所提出网络的有效性,使用了固体火箭发动机模拟件的数据集进行实验。数据集将从投影数据中连续提取出来的正弦图分为五组,并为每组添加随机噪声,且每组之间的噪声程度逐渐增加。实验结果表明,本文提出的网络框架在CT重建中具有可行性和优越的性能。与传统方法相比,深度学习方法在面对逐渐增加的噪声时,重建结果几乎不受影响,显著提升了图像质量和重建精度。Computed tomography(CT)is a widely used method for image reconstruction.However,due to the complexity of CT imaging systems,noise interference during acquisition is inevitable,which reduces reconstruction precision.To address this issue,this study proposed a generative adversarial network(cGAN)based on specification,and introduced the CBAM attention mechanism in the network to better extract key information,enabling reconstruction of string diagrams with noise interference.To verify the performance of the proposed network,a solid rocket motor simulation dataset was used for experiments.Experiments show that the proposed network architecture is not only feasible but also superior for CT reconstruction.Compared to traditional methods,deep learning approaches show minimal impact on reconstruction results as noise increases,significantly enhancing image quality and reconstruction accuracy.

关 键 词：工业CT图像 注意力机制 CT重建 cGAN 深度学习 

分 类 号：TP311[自动化与计算机技术—计算机软件与理论]