Study Of Limited-Views CT Image Reconstruction

Computer tomography system acquire the projection data in different views,using the X-ray rotating scanner. High dose of X-ray is harmful to the patient, andmakes the low utilization of the X-ray. For another, the perspective projection wouldtake a long time. So, in recent years research on the incomplete data cause by lowdose and limited angular range have significantly meaning.Filter back projection(FBP) reconstruction in two-dimension and FDK inthree-dimensional could get a good reconstruction image based on the full projectiondata. But in limited angular range, when the projection data are not sufficient for exactreconstruction of tomography images, and such algorithm will not get a high qualityimage. While the iterative reconstruction algorithms have obvious advantages in theinsufficient projection data. However, the huge load of computation prevent it fromapplication in CT image reconstruction, especially for the reconstruction oflarge-scale CT system.TV-POCS image reconstruction algorithm is based on the total variation function,the iterative process includes two major phases: gradient descent on the TV function,and projection onto convex sets(POCS) for the constraint. In this algorithm, theiterative step and the number of iterations are fixed. But different projection methodsuse the different step and iteration number, so they must be determined throughseveral tests.ASD-POCS is an improved algorithm of TV-POCS. Different with the TV-POCSalgorithm, ASD-POCS algorithm use a adaptive step to control the convergence of theTV gradient descent algorithm. At beginning of the iteration, we could select a largerstep to accelerate the iteration speed. When the iterative image is close to the targetimage, we should select another parameter to reduce the iterative step to ensure theconvergence of the iteration.CCSD-POCS is another improved algorithm of TV-POCS. In ASD-POCS, theiteration step is controlled by many artificial parameter, so the scope of algorithmapplication has been limited. CCSD-POCS has proposed a nonparametric method forconstrained TV optimization. The method automatically updates the step-size of TViteration according to the changes in the consistency term defined by the constraintswithout introducing artificial parameters. In this paper, we propose a two-strategy algorithm. The objective function isoptimized using the first order primal dual algorithm. Physical constraints that cannotbe efficiently adopted in the objective are enforced separately using the projectiononto convex set (POCS) method. The FOPD and POCS operations are applied inalternation until the desired solution is obtained. The hybrid iterative algorithmprovides a flexible approach for CT reconstruction.To verify efficiency of the proposed method, comparative tests have beendesigned. Using the representative projection data of few-view CT reconstruction, theexperiment include two aspects. We investigated its performance on X-ray imagingwith a limited number of projection views and with different levels of radiationintensities.