基于残差学习的CT稀疏重建伪影校正  

Artifacts correction of CT sparse reconstruction based on residual learning

作　　者：张鹏程 谢世朋[1] ZHANG Peng -cheng, XIE Shi- peng(College of Telecommunication and information Engineering, Nanjing Uinversity of Posts and Telecommunication, Nanjing 210003, Chin)

机构地区：[1]南京邮电大学通信与信息工程学院,江苏南京210003

出　　处：《计算机工程与设计》2018年第8期2557-2563,共7页Computer Engineering and Design

基　　金：国家自然科学基金项目(11547155);教育部-中国移动科研基金项目(MCM20150504);江苏省科技重点研发计划-产业前瞻与共性关键技术基金项目(BE2016001-4);南京邮电大学科研基金项目(NY214026;NY217035);江苏省高校自然科学基金项目(17KJB510038)

摘　　要：为减少CT扫描中X射线的辐射剂量,提出一种基于残差学习的CT稀疏重建伪影校正的方法。增大CT扫描间隔的角度,减少投影的个数;利用FBP方法对原始数据进行反投影重建;建立基于残差学习的卷积神经网络模型,对重建过程中的伪影特征进行学习;通过对稀疏重建的图像进行残差操作,得到去伪影的CT图像。实验结果表明,该方法降低了X射线的辐射剂量,减少了传统稀疏重建中正则化的影响,有效降低了条形伪影。To reduce the radiation dose of X ray in CT scanning,a method of artifacts correction of CT sparse reconstruction based on residual learning was proposed.CT scan interval angle was increased,reducing the number of projections.The original data were reconstructed using FBP method.A convolution neural network model based on residual learning was proposed to learn the artifacts in the reconstruction process.The CT image of the artifact-free was obtained by residual operation on the sparse reconstructed image.Experimental results show that the method reduces the radiation of X ray,while effectively reduces the influence of regularization in the traditional sparse reconstruction and the streak artifacts.

关 键 词：计算机断层扫描 稀疏重建 残差 卷积神经网络 伪影 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]