Research On Cone-Beam Reconstruction Algorithms Based On DSA Scan Trajectory

Computed tomography (CT) is a technology which yields series of cross-sectional images inside of the object by nondestructive method, and has been widely applied to medical diagnostics, industrial nondestructive detection, aerospace and other fields. Due to the advantages of the cone-beam scan on spatial resolution, acquired speed, utilization ratio of X-ray, many researchers have been devoted to studying the scientific issues about cone-beam CT. In medical imaging field, digital subtraction angiography (DSA) used during interventional procedures can provides a basic imaging pipeline for more advanced cone-beam acquirsiton data, which has been used to visualize high contrast object. Recently, the advent of high quality flat panel detector has attracted more attention to the low contrast resolution imaging, which further enhance the diagnostic and therapeutic capabilities. Effective cone-beam reconstruction algorithm is a bottleneck for development of cone-beam CT, which is also the key factor improving the quality of reconstructed images. The main purpose of this dissertation is to investigate the algorithms of cone-beam reconstruction oriented to the projection data obtained by the DSA equipment so as to optimize the low-contrast imaging quality. The main research topics are included as follows:DSA system mainly utilizes a single arc trajectory and the short-scan FDK algorithm to reconstruct cone-beam projection data. Cone-beam artifacts are inevitable for image voxels out of the scanning plane. In this dissertation, a modified method is firstly presented based on the Slant-FDK algorithm, which avoids the process step of rebinning the cone-beam data to 3D parallel-beam data. Because of keeping filtering together as much as possible for the same voxels at different projection angle, this approach enlarges reconstructed volume opposite the FDK algorithm. Inspired by the successful above algorithm, the heuristic derivation of the FDK algoirithm is reconsidered on the family of planes with the same tilted angle angle relative to the trajectory plane. And then, a rotation-angle-dependent weighting scheme is increasing developed to modify the classifical FDK algorithm. Improved idea is also sucessfully generalized into the T-FDK algorithm, which can better suppress the intensity drop artifact inherent to the FDK algorithm.Modified T-FDK algorithm can obtain better reconstruction result, however it involves rebinning rebinning precess before recontructing. To avoid it, super-short-scan algorithm is studied. This kind method can be roughtly divided into the two classes according to whether reconstruction formula includes partial derivative with respect to the trajectory variable. The algorithm excluding derivative along the source trajectory has smaller reconstruction volume relative to other class method. Based on the decomposition of 3D Radon inversion transform, this dissertation derives a new reconstruction formula to solve the difference between two class algorithms. In addition, the Radon data which don't offer in the circular trajectory are compensated on some extent by the correction factor of rotation angle from the scheme in the modified T-FDK algotirhm, which make the proposed algorithm achieve the similar image quality as the modified T-FDK algotirhm.Severe cone artifacts appear in the images reconstructed by the FDK-Type algorithms along the arc trajectory for object including homogeneous component with embedded low contrast and high contrast structures. To address this problem, the Katsevich reconstruction theory is applied into the arc trajectory, in which this dissertation mainly detemines the value of structure factor using the geometric method, explores the distribute rule of the structure factor on the detector plane and derives reconstruction formulas satifying the sampling rate of prejection rate along filtering line. However, unlike the FDK- type algorithm, Katsevich-type algorithm involves non-horizontal filtering, and moreover projection data acquired from DSA equipment are truncated in the axial and transverse direction. Therefore, piecewise cubic Hermite interpolation is introduced to extrapolate the truncated projections, which can improve the truncation artifacts in the Katsevich-type algorithm.Due to the incompleteness of Radon data from the single arc trajectory, there exists some cone artifacts in reconstructed results, which significantly degrade image quality and can't be suppressed by the current algorithtms. To assure the data completeness, a reconstruction algorithm is investigated based on the katsevich's result for the arc-plus-line orbit adapting to the DSA equipment. In contrast to the original Katsevich's algorithm, this dissertation first studies the condition needing to be satisfied for this orbit in order to exact cone-beam reconstruction and compensates for the contribution of projection data on the plane tangental to arc orbit at the intersection point of two source orbits. And then, weight fuction of border transition is used to smooth step across the border of the backprojection region, which improving the quality of reconstructed image. Last, a new reconstruction formula is derived that avoids explicit computationof the patial derivative with respect to the variable of trajectory. The proposed algorithm is quantitatively more accurate than the original Katsevich's algorithm in the case of reconstructing intensive distribution small objects .