CIOMP OpenIR  > 中科院长春光机所知识产出
基于压缩感知理论的锥束CT断层图像重建算法研究
杨宏成
学位类型博士
导师张涛
2014-07
学位授予单位中国科学院大学
学位专业光学工程
摘要计算机断层成像(Computed Tomography,CT)技术通过无损方式获取物体内部结构信息,广泛用于医学辅助诊断、工业无损检测、安全检查等领域。锥束CT体积小,重量轻,移动灵活,可在手术中快速拍片、定位,广泛用于介入手术治疗。受圆轨道几何扫描结构限制,锥束CT无法获取完备投影数据,制约了锥束CT成像质量。近年提出的压缩感知理论是一种信号高度不完备线性测量的高精确重建技术。在锥束CT图像重建领域,压缩感知理论利用图像的稀疏性先验信息,可从少视角投影数据中精确恢复原始图像。本文基于压缩感知理论,围绕重建精度、收敛速度和计算效率等问题展开研究工作,重点研究了基于FDK的反投影权重算法、基于投影收缩的压缩感知算法、快速自适应共轭梯度投影算法和锥束CT重建并行加速算法,其成果如下:    针对锥束CT系统锥角增大而导致的锥束伪影严重的问题,提出了一种基于FDK的反投影权重锥束CT重建算法(back projection weighted FDK,BPW-FDK)。分析了圆扫描轨迹远端伪影的成因,针对圆周扫描阴影区域导致的Radon空间数据缺失,提出了一种距离变量的反投影权重函数,并将其作为约束条件引入到FDK算法中,实现扫描轨迹远端区域的数据补偿,扩大图像重建区域。与Parker-FDK算法相比,BPW-FDK算法重建图像在目标区域的归一化均方距离判据d和归一化平均绝对距离判据r均下降了50%以上,信噪比提高了5db。    针对锥束CT成像系统中不完全投影数据重建问题,提出了一种基于投影收缩的压缩感知锥束CT短扫描重建算法(Projection-Contraction Barzilari-Borwein,PCBB)。针对BB梯度投影算法的非单调收敛,分析投影收缩法的预测校正特性,将校正过程引入压缩感知图像重建算法中,结合目标函数下降方向和凸集投影下降方向,对BB梯度投影算法进行校正,改善BB梯度投影算法的非单调特性。仿真结果表明,在25个采样角度下,PCBB算法重建图像的信噪比值比ASD-POCS算法、PC算法、GPBB算法的重建结果分别高出9.4870db、9.8027db、3.6159db。    针对锥束CT成像系统中压缩感知算法最速下降法收敛缓慢问题,提出了一种快速自适应共轭梯度压缩感知锥束CT重建算法(adaptive stepsize congjuate gradient, ASCG)。利用Lipschitz连续性求出下降步长,然后使用共轭梯度下降法迭代计算,最后采用联合代数重建算法更新重建图像。在每次迭代过程中自适应调整梯度下降步长,进一步加快重建算法的收敛速度。ASCG算法能够有效抑制条状伪影,极大提高少量投影数据时重建图像质量。在40投影角度下,ASCG算法重建结果的相对误差为0.1%,比GPBB算法重建结果的相对误差提高了一个数量级。    锥束CT图像重建的计算复杂度与被重建体数据量N、投影视图个数M的乘积成正比,CPU架构的锥束CT图像重建时间往往达到几十分钟,难以满足实时成像要求。针对锥束CT图像重建时间过长这一瓶颈问题,提出了高度优化的基于GPU的Parker-FDK算法和联合代数重建算法(Simultaneous Algebraic Reconstruction Technique,SART)。基于CUDA架构的Parker-FDK算法充分利用以下技术:(1) 优化使用线程块大小;(2) 提高常量存储器和共享存储器的重复利用率;(3) 使用纹理存储器的线性插值提高计算效率;(4) 使用多GPU进一步提高加速比。基于CUDA的SART代码实现:(1)基于射线驱动的正投影技术,使用纹理存储器三线性插值技术;(2) 基于体素驱动的反投影技术,使用共享存储器减少冗余计算。实验结果表明,基于 CUDA架构的Parker-FDK算法和SART算法的时间性能得到极大提升:Parker-FDK算法的重建时间减少为0.33s;与CPU相比,基于CUDA架构的SART算法加速比约为100倍。
其他摘要Computed tomography (CT) technique can get the internal information through a non-destructive way and has been widely used in a large number of applications in medical diagnosis, industrial non-destructive detection and other fields. Because of its small size, light weight, mobility and flexiblity, C-arm cone-beam CT is extensively applied to interventional surgery. However, C-arm cone-beam CT can not get sufficient projection data for exact reconstruction due to its short scanning and sparse views. Thus the quality of the reconstructed image is not satisfied. Recent developments in compressed sensing have enabled an accurate cone beam computed tomography (CBCT) reconstruction from highly undersampled projections. In this dissertation, we concentrate on the backprojection weightedFDK algorithm, projection contraction based compressed sensing algorithm, a fast adaptive conjugate gradient projection algorithm to improve the reconstruction accuracy and convergence speed. GPU is used to accelerate the image reconstruction. The results are as follows:    Cone beam artifacts increase along with the larger cone angle because the scanning trajectory of the C-arm cone beam does not sufficiently satisfy data conditions,. Aimed at the characteristics of missing data in Radon space, we propose a Backprojection Weighted-FDK (BPW-FDK) algorithm for C-arm cone beam. A new backprojection weight is presented to compensate for the missing data away from the rotating track for reconstruction region expansion. The images reconstructed from simulated noiseless projections, projections with noise, and real projections from an internally developed 3D scanner show that the proposed algorithm is  able to sufficiently suppress artifacts away from the rotating track for large cone angle and provide more homogeneous image contrast. Compared to Parker-FDK algorithm, normalized mean square distance criterion and normalized mean, absolute distance criterion of BPW-FDK decreased by 5%. Its accuracy and speed make BPW-FDK algorithm suitable for image reconstruction of real large C-arm cone beam.     To solve the problem of image reconstruction of incomplete projection data from short-scan cone-beam CT, a novel cone-beam CT short-scan reconstruction algorithm based on projection-contraction method was proposed. Aiming at the non-monotonic convergence of Gradient-Projection Barzilari-Borwein algorithm (GPBB), the predictor-corrector feature of projection-contraction method was analyzed and was incorporated into compressed sensing image reconstruction algorithm. The objective function descent direction and the projection onto convex sets descent direction were combined to correct the results of GPBB algorithm to improve the non-monotonic convergence of GPBB algorithm. The experiments were conducted on simulated projection data and phantom scanning data. The simulated results show that, for 25 sampling angles, signal-to-noise ratio of images reconstructed by PCBB algorithm is 9.4870db, 9.8027db, 3.6159db higher than those of images reconstructed by Adaptive Steepest Descent-Projection Onto Convex Sets algorithm, projection contraction algorithm and GPBB algorithm, respectively. The results of Phantom indicate that even when a small amount of projections are acquired, the new algorithm can effectively suppress strip artifacts and the reconstructed images show clear edge. The algorithm can greatly improve qualify of images reconstructed from few projection data.      Gradient descent commonly used in these reconstruction methods has a slow convergence speed. Thus we propose a novel CBCT reconstruction algorithm based on adaptive stepsize conjugate gradient (ASCG) method, which overcomes the drawback of the gradient descent methods. The Forbild numerical phantom is used to evaluate the performance of ASCG. Image relative error of reconstructed images and computation efficiency were assessed and the behavior of ASCG are compared with simultaneous algebraic reconstruction technique (SART), gradient projection Barzilai-Borwein (GPBB) and another conjugate gradient projection using fixed stepsize which is referred to FSCG. Under the condition of 50-view projections, the ASCG algorithm showed convergence about 600 iterations whereas other algorithms need more than 1000 iterations to reconstruct the Forbild phantom image. For the same number of iteration, the computation time of ASCG is less than half of those of GPBB algorithm. we propose a novel adaptive stepsize conjugate gradient projection algorithm for sparse-view CBCT reconstruction. Compared to GPBB algorithm, ASCG algorithm has better performance both in convergence speed and reconstruction accuracy. These advantages have been demonstrated on Forbild phantom studies.      To accelerate the Parker-FDK algorithm and SART for speedy and quality CT reconstruction by exploiting CUDA-enabled GPU, these techniques are proposed: (1) optimizing thread block size, (2) maximizing data reuse on constant memory and shared memory, (3) exploiting texture memory interpolation capability to increase efficiency, and (4) using multiply GPUs. Two core techniques are proposed to use SART into the CUDA architecture: (1) a ray-driven projection along with hardware interpolation, and (2) a voxel-driven back-projection that can avoid redundant computation by combining CUDA shared memory. Extensive experiments demonstrate the proposed techniques can provide faster reconstruction with satisfied image quality.
语种中文
文献类型学位论文
条目标识符http://ir.ciomp.ac.cn/handle/181722/41491
专题中科院长春光机所知识产出
推荐引用方式
GB/T 7714
杨宏成. 基于压缩感知理论的锥束CT断层图像重建算法研究[D]. 中国科学院大学,2014.
条目包含的文件
文件名称/大小 文献类型 版本类型 开放类型 使用许可
杨宏成.pdf(5208KB) 开放获取CC BY-NC-SA请求全文
个性服务
推荐该条目
保存到收藏夹
查看访问统计
导出为Endnote文件
谷歌学术
谷歌学术中相似的文章
[杨宏成]的文章
百度学术
百度学术中相似的文章
[杨宏成]的文章
必应学术
必应学术中相似的文章
[杨宏成]的文章
相关权益政策
暂无数据
收藏/分享
所有评论 (0)
暂无评论
 

除非特别说明,本系统中所有内容都受版权保护,并保留所有权利。