一种低剂量锥束CT三维图像重建快速GPU并行算法

A FAST GPU PARALLEL ALGORITHM FOR LOW DOSE CONE BEAM CT 3D IMAGE RECONSTRUCTION

鉴于非局部平均NL-Means(Nonlocal Means)算法的高性能图像去噪表现,并考虑到重建速度这一重要因素,提出一种低剂量锥束CT稀疏角度3D图像迭代重建算法。首先,采用最小二乘方法进行图像重建以满足投影数据一致性,再对重建图像进行非负约束;然后,利用非局部平均算法对以上非负约束后的图像进行滤波处理,起到去噪保边缘的作用。以上各步骤均可以进行并行化处理,交替执行直至满足迭代终止条件。实验结果表明,该迭代重建算法获得了满意的3D图像质量,尤其适合并行化GPU加速,重建速度大幅度提升。

In view of the denoising performance of high performance image in non-local means（ NL-Means） algorithm,and taking into account the important factor of reconstruction speed,we present an iterative reconstruction method for low-dose cone-beam CT 3D image with sparse-view.First,we utilise the least square method for image reconstruction to meet the consistency of projection data,and then apply nonnegativity constraints on the reconstructed images; Secondly,we use non-local means algorithm to make filtration processing on the images treated with nonnegativity constraints,which plays the role of denoising and edge preserving.The above steps can be parallelly processed and executed alternatively until the program meets the iteration termination condition.Experimental results show that the iterative reconstruction method achieves satisfied 3D image quality,it is especially suitable for paralleled GPU acceleration,and the reconstruction speed is greatly improved as well.