A fast iterative reconstruction method based on the selective total variation for sparse angular CT

Abstract

Sparse angular Computer Tomography (CT) is a rapidly developing imaging modality that reconstructs high-quality images from sparse data toward low-dose x-rays. The effectiveness of conventional total variation (TV) algorithm is limited by the over-smoothness on the edges and slow convergence. To mitigate this drawback, we proposed an improved fast iterative reconstruction method based on the minimization of selective image TV. The presented selective TV model is derived by linking the regularity metric to the local gradient of images, and selectively applies different degrees of regularization (the value of p) to background and potential signal locations for the purpose of preserving the edge details. In order to further speed up the convergence, we draws on a fast variant of The Iterative-Shrinkage-Thresholding Algorithm (ISTA), which uses a special linear combination of the two previous iterate results as the initial value of next iteration for more accurate correction. Experiments on simulated Shepp-Logan phantom are performed. The results demonstrated that the new method not only protected the edge of the image characteristics, but also significantly improved the convergence speed of the iterative reconstruction.