L_1范数字典约束的感兴趣区域CT图像重建算法

A Reconstruction Algorithm of CT Images of Interested Regions Based on an L_1 Norm Dictionary Sparse Constraint

针对现有全变分(TV)约束感兴趣区域(ROI)重建方法易产生块状伪影、细小结构丢失的问题,提出了一种L1范数字典稀疏约束的ROI低剂量CT医学图像重建算法。首先将ROI医学图像重建问题转化为最优化问题,以罚加权最小二乘函数为保真项,L1范数字典稀疏表示为约束项构建目标函数;然后将目标函数分解为图像更新和字典稀疏表示两个子优化问题,并交替求解上述两个子优化问题,实现ROI图像重建。胸腔模体仿真实验结果表明,在分别添加光子数为1×105、5×104和1×104泊松噪声投影情况下,与TV约束重建方法相比,图像结构相似度(SSIM)分别提高约0.103 5、0.113 1和0.125 8,峰值信噪比分别提高4.88、4.93和5.44dB。山羊肺部实际CT扫描实验结果进一步证明,本文算法能够有效地去除块状伪影且较好的保留细小结构。

A medical image reconstruction algorithm of CT images of interested region with low does based on an L 1 norm dictionary sparse constraint is proposed to address the problem that the existing total variation(TV)regularization algorithms often suffer from patchy artifacts and losing fine structure.First,ROI image reconstruction is converted into an optimization problem by using a penalized weighted least squares function to establish a data-fitting term and the L 1 norm of sparse representation in terms of learned dictionary as a constraint term.Then,the objective function is split into an image updating sub-optimization problem and a sparse representation sub-optimization problem,and these two sub-problems are alternatively solved in a minimization manner.Chest simulation results and a comparison to the reconstruction algorithm with TV regularization show that in the cases of Poisson noise projection added 1×10^5,5×10^4 and 1×10^4 photons per detector element,respectively,the proposed algorithm decreases the structural similarity index metric by 0.103 5,0.113 1 and 0.125 8,respectively,and improves the peak signal to noise by 4.88,4.93 and 5.44 dB,respectivtly.Moreover,experiments with sheep Lung real CT validates that the proposed algorithm can effectively remove blocky artifacts and preserve low-contrast structures.