Daxin Shi

Multi-scale iterative reconstruction

Zoomed ROI reconstruction is a natural challenge for iterative reconstruction (IR), since the reprojection step requires knowledge of the entire object attenuating the x-ray beam. In contrast, analytic FBP can reconstruct images within the zoomed ROI without the need of the full FOV reconstruction, provided projection data is not truncated. To overcome this challenge in IR, the following approaches have been suggested: 1) Remove the part of the projection data corresponding to the outside-of-ROI part of the image volume. 2) Use two volume grids, one for full FOV and one for desired ROI. In the case when the size of desired ROI becomes very small, resolution of the fine grid becomes dominated by the coarse grid, and small features cannot be resolved. Our approach is inspired by multi-resolution framework. Image reconstruction is done in several steps. At each step we reduce the size of reconstruction FOV, for example by half, thus refining the scale, since matrix size remains constant. The proposed approach provides the following advantages compared to the standard 2-grid approach 1) Spatial resolution is preserved in zoomed ROI reconstructions. 2) Computational burden and computer memory requirement are reduced. 3) Parameter optimization is simplified, since only one volume is reconstructed at each iteration.

Extension of axial coverage and artifact reduction in iterative reconstruction in computed tomography

It is well known that iterative reconstruction (IR) of a long object is a challenge in case of circular trajectory. Axial artifacts occur naturally in IR due to data truncation. Because IR uses reprojection, axial coverage of the reconstruction field of view reduces with each iteration. In this work we present a method to extend axial coverage and reduce artifacts in the circular cone-beam scanning geometry. Our approach is based on our recently presented weighted iterative reconstruction algorithm. We use extended FDK reconstruction for FBP seed, expansion of detector rows in Z direction, and a special axial weight designed to reduce artifacts caused by data truncation. Evaluation shows that sharp streaks at the edge of the reconstruction volume are removed and good image quality in the extended region is achieved.

Cone beam artifact reduction in circular computed tomography

Cone beam artifact is a well known problem in circular computed tomography. The artifact can be removed by applying exact reconstruction, for example with circle and line trajectory. However, an additional scan often is not available for reconstruction, nor it is practical to collect. Then, methods that can accurately reconstruct image volume from circular data only are of particular interest. In this work we propose a new approach that combines elements of exact reconstruction and iterative reconstruction (SART). We apply iterative procedure to estimate missing line data; at each iteration line data becomes more accurate and cone beam artifacts are further reduced. Evaluation with numerical disc phantom and clinical head data shows effectiveness of the proposed approach in cone beam artifact reduction.

A simplified implementation of total variation iterative reconstruction algorithm suitable for parallel computation

In this work, we proposed a simplified implementation of the total variation minimization iterative reconstruction algorithm for the conventional x-ray CT application, which was originally developed by the University of Chicago group. The simplifications make it possible for completely parallel computation that is suitable to be implemented with a GPU board. Reconstructed images from mathematical and real phantoms are demonstrated to verify the validity of our simplified algorithm.

Application of anisotropic diffusion potential to sparse view reconstruction problem

In this work, a new constrained optimization problem associated with anisotropic diffusion potential was formulated to provide reconstructions from projection data acquired from limited number of view angles. Our results from both simulated data and physical phantom data showed that the anisotropic diffusion method can produce appealing images from sparse view data which suggests it could be potentially applied to lower the dose delivered to patient. Compared with the total variation minimization algorithm, the anisotropic diffusion method can be an alternative candidate to sparse view reconstruction problem. The similarities between anisotropic diffusion and TV based algorithms are also discussed. A possible combination of the anisotropic diffusion and total variation methods is suggested in this paper.