刘晓培 ,滕建辅 ,费 腾 ,孙云山
AuthorsHTML:刘晓培 ^{1, 2}，滕建辅 ³ ，费 腾 ² ，孙云山 ²
AuthorsListE:Liu Xiaopei,Teng Jianfu,Fei Teng,Sun Yunshan
AuthorsHTMLE:Liu Xiaopei^{1, 2}，Teng Jianfu³，Fei Teng²，Sun Yunshan²
Unit:1. 天津大学电气自动化与信息工程学院，天津 300072；
2. 天津商业大学信息工程学院，天津 300134；
3. 天津大学微电子学院，天津 300072

Unit_EngLish:1. School of Electrical and Information Engineering，Tianjin University，Tianjin 300072，China；
2. School of Communication，Tianjin University of Commerce，Tianjin 300134，China；
3. School of Microelectronics，Tianjin University，Tianjin 300072，China

Abstract_Chinese:本文提出了一种基于稀疏贝叶斯智能优化(SBL-RNAMBO)算法的低剂量医学 CT 图像的盲复原重建方 法．首先，利用群智能算法的全局搜索能力，同时引入稀疏贝叶斯方法进行训练，将大量全剂量 CT 图像作为先验 信息，改善图像重建过程中的欠定问题．其次，加入二次惩罚项约束解空间，构造了参数未知的后验概率目标函 数，采用 RNAMBO 算法优化稀疏贝叶斯超参数，建立优化后的稀疏贝叶斯模型．最后，用 SBL-RNAMBO 方法对 所有盲复原未知量进行估计并求解后验概率目标函数．将 SBL-RNAMBO 算法与其他 5 种对比算法进行 Shepp\u0002Logan 体膜、临床盆腔 CT、临床脑部 CT 的定性及定量实验，实验结果表明，在定性实验中该方法可以获得良好的 CT 重建图像，保留清晰的纹理细节和结构特征；在 145/20 mA 及 90/20 mA 定量实验中峰值信噪比(PSNR)、通用 图像质量指数(UIQI)、结构相似性指数(SSIM)、误差方差和(SSDE)指标均优于对比算法，算法复杂度最大减少 854.6 s，通过不同初始值 PSNR 实验，验证了该算法的稳定性及有效性．
Abstract_English:This paper proposes a blind reconstruction method for low-dose medical computer tomography(CT)images based on the sparse Bayesian intelligent optimization(SBL-RNAMBO)algorithm. First，we applied the global search capability of the swarm intelligence algorithm and introduced the sparse Bayesian method for training，and a large number of full-dose CT images were used as prior information to improve the underdetermined problem in the image reconstruction process. Second，a quadratic penalty term was added to constrain the solution space，and a posterior probability objective function with unknown parameters was constructed. The RNAMBO algorithm was used to opti\u0002mize the sparse Bayesian hyperparameters to establish the optimized sparse Bayesian model. Finally，the SBL\u0002RNAMBO method was used to estimate all unknown variables of blind restoration as well as to solve the posterior probability objective function. The SBL-RNAMBO algorithm was then applied with five other algorithms to perform qualitative and quantitative experiments on the Shepp-Logan body membrane，clinical pelvic CT，and clinical brain CT. The results showed that the SBL-RNMABO method could obtain good CT reconstruction images in qualitative experiments，while retaining clear textural details and structural features. In the 145/20 mA and 90/20 mA quantitative experiments，the peak signal-to-noise ratio(PSNR)，universal image quality index(UIQI)，structure similarity in-dex(SSIM)，sum of squared difference error(SSDE) indices were better than the contrast algorithm，and the algorithm complexity could be reduced by 854.6 s. The PSNR experiment with different initial values verified the stability and effectiveness of the algorithm.
Keyword_Chinese:低剂量 CT 图像；图像盲复原重建；群智能优化；稀疏贝叶斯算法
Keywords_English:low dose CT image；blind image restoration and reconstruction；swarm intelligence optimization； sparse Bayesian algorithm

PDF全文下载地址:http://xbzrb.tju.edu.cn/#/digest?ArticleID=6695