Choi J-H, Keil A, Maier A, Pal S, McWalter E J, Fahrig R
Stanford University, Stanford, CA.
Siemens AG, Forchheim, Franken.
Med Phys. 2012 Jun;39(6Part28):3973. doi: 10.1118/1.4736213.
Imaging the knee under realistic load-bearing conditions can be carried out in a horizontal plane using a C-arm CT scanner. Human subjects can be scanned in a standing position and acquired data successfully reconstructed. However, reconstructing this data is a challenge due to significant artifacts that are induced due to involuntary motion. Here, we propose motion correction methods in 2D and 3D.
Four volunteers were scanned for 8 seconds while squatting with ∼30 degree flexion. Eight tantalum fiducial markers suitably attached around the knee were used to track motion. The marker position in each projection was semi- automatically detected. Each marker's static 3D position, which served as a reference to correct temporal motion, was estimated by triangulating each marker's 2D position from 248 projections using known projection matrices. Motion was corrected in 3 ways: 1) 2D projection shifting based on the mean position of markers, 2) 2D projection warping using approximate thin- plate splines, 3) 3D rigid body warping.
The original reconstruction was severely motion-corrupted which made it impossible to distinguish the boundaries of bones. Reconstruction with projection shifting and warping in 2D improved visualization of edges of soft tissue as well as bone. A simple numerical metric of residual bead deviation from static position was reduced from 3.2mm to 0.4mm. The 2D-based methods are inherently limited in that they cannot fully accommodate different 3D movements at different depths from the X-ray source. Reconstruction with 3D warping shows clearer edges and less streak artifact than the 2D methods.
The proposed three motion correction methods effectively reduced motion-induced artifacts in the reconstruction and are therefore suitable for weight-bearing scanning. Future work includes scanning patients in standing position after contrast injection for evaluating the soft tissue structure and constructing 3D finite element models for the estimation of joint cartilage stress. This study was supported by Center for Biomedical Imaging at Stanford, by Siemens AG, Healthcare Sector, and by the Lucas Foundation at Stanford. The concepts and contents proposed here are based on research and are not commercially available.
使用C形臂CT扫描仪可在水平面内对处于实际负重状态下的膝关节进行成像。人体受试者可在站立位进行扫描，采集的数据也能成功重建。然而，由于非自主运动导致的显著伪影，重建这些数据颇具挑战。在此，我们提出二维和三维的运动校正方法。
四名志愿者在约30度屈曲下蹲时扫描8秒。在膝关节周围适当附着八个钽基准标记物以跟踪运动。每个投影中标记物的位置通过半自动方式检测。通过使用已知投影矩阵从248个投影中对每个标记物的二维位置进行三角测量，估计每个标记物的静态三维位置，该位置用作校正时间运动的参考。运动通过三种方式进行校正：1）基于标记物平均位置的二维投影移位；2）使用近似薄板样条的二维投影变形；3）三维刚体变形。
原始重建严重受运动干扰，无法区分骨骼边界。二维投影移位和变形后的重建改善了软组织以及骨骼边缘的可视化。残余珠子偏离静态位置的简单数值指标从3.2毫米降至0.4毫米。基于二维的方法本质上存在局限性，因为它们无法完全适应距X射线源不同深度处的不同三维运动。与二维方法相比，三维变形重建显示出更清晰的边缘和更少的条纹伪影。
所提出的三种运动校正方法有效减少了重建中运动引起的伪影，因此适用于负重扫描。未来的工作包括在注射造影剂后对患者进行站立位扫描以评估软组织结构，并构建三维有限元模型以估计关节软骨应力。本研究得到了斯坦福大学的生物医学成像中心、西门子医疗部门以及斯坦福大学的卢卡斯基金会的支持。这里提出的概念和内容基于研究，尚未商业化。
