Chityala R, Hoffmann K R, Rudin S, Bednarek D R
Toshiba Stroke Research Center, Rm 445, Biomedical Research Building, State University of New York at Buffalo, 3435 Main Street, Buffalo, NY 14214.
Proc SPIE Int Soc Opt Eng. 2005 Feb;5747(3):2110-2117. doi: 10.1117/12.595450.
Truncation of projection data in CT produces significant artifacts in the reconstruction process due to non-locality of the Radon transform. In this paper, we present a method for reducing these truncation artifacts by estimating features that lie outside the region of interest (ROI) and using these features to complete the truncated sinogram.Projection images of an object are obtained. A sinogram is obtained by stacking profile data from all projection angles. A simulated truncated sinogram is generated by setting pixel values outside an ROI to zero. The truncated sinogram is then transformed into a (radius, phase) image, with pixel values in what we term as the Polar representation (PR) image corresponding to the minimum value along sine curves given by x = r*cos(projection angle + phase). The PR image contains data for radii greater than the ROI radius. Pixel values outside the ROI in the completed sinogram are determined as follows. For each pixel in the PR image, a sine curve is generated in the completed sinogram image outside the ROI, having the same pixel value as that of the PR image for that radius and phase. Successive sine curves are laid and the values of each are summed. The intensity outside is then equalized to the intensity inside the ROI. The completed sinogram is then reconstructed, to obtain completed reconstruction.The percentage error in the difference image between the full FOV reconstruction and the corresponding completed reconstruction and the extrapolated-average reconstruction are 1.1% and 3.3% respectively. This indicates that the completed reconstruction is closer to full FOV reconstruction. Thus, the sinogram completion can be used to improve reconstructions from truncated data.
由于拉东变换的非局部性，CT 中的投影数据截断会在重建过程中产生显著伪影。在本文中，我们提出了一种减少这些截断伪影的方法，即通过估计位于感兴趣区域（ROI）之外的特征，并使用这些特征来完成截断的正弦图。获取物体的投影图像。通过堆叠来自所有投影角度的轮廓数据获得正弦图。通过将 ROI 之外的像素值设置为零来生成模拟截断正弦图。然后将截断的正弦图转换为（半径，相位）图像，我们称之为极坐标表示（PR）图像中的像素值对应于由 x = r*cos（投影角度 + 相位）给出的正弦曲线沿其的最小值。PR 图像包含半径大于 ROI 半径的数据。完成的正弦图中 ROI 之外的像素值按以下方式确定。对于 PR 图像中的每个像素，在 ROI 之外的完成的正弦图图像中生成一条正弦曲线，其像素值与该半径和相位的 PR 图像的像素值相同。铺设连续的正弦曲线并对每个曲线的值求和。然后将外部强度与 ROI 内部强度进行均衡。然后对完成的正弦图进行重建，以获得完整的重建。全视野重建与相应的完整重建以及外推平均重建之间的差异图像中的百分比误差分别为 1.1%和 3.3%。这表明完整重建更接近全视野重建。因此，正弦图完成可用于改进截断数据的重建。
