An improved method for precise tomographic reconstruction with limited projections

by Xiaoli Yang

Tuesday May 26, 2015, 11:00 AM → 12:00 PM Europe/Berlin

X-ray computed tomography (CT) is a popular, non-invasive technique able to produce structural information with high resolution. In multiple applications more and more attention has been paid to the tomographic imaging method with limited amount of projections, which is important in limiting the radiation dose to a low level, or adapting a rapid imaging process or specific geometry. However, fewer projections normally imply worse quality of image reconstruction with artifacts, leading to difficulties in the following data analysis. A new reconstruction method other than the traditional filtered back-projection (FBP) method is demanded.

To overcome this challenge, a parameter-optimized ART reconstruction method is proposed for computed tomography with limited projections, subject to the minimization of the total variation (TV) in pursuing a precise three-dimensional (3D) reconstruction in the sense of compressive sampling (CS) theory. This problem is formulated in a Lagrangian multiplier fashion with the parameter optimization appealing to a discrete L-curve. The reconstruction is carried out using the corresponding developed conjugate gradient solver to minimize the TV-regularized formulation. All these methods are incorporated into an automatic framework of parallel 3D reconstruction in a computer cluster to achieve a near real time reconstruction performance.

The proposed method is evaluated with various datasets, including both simulated and experimental data produced by synchrotron X-ray tomography. The reconstructions from few-view projections demonstrate the promising performance of the proposed method. With the parallel reconstruction framework in the computer cluster, the 3D reconstruction can be finished in near real time.