Improving the reconstruction accuracy of MR imaging using Zero-point Attracting Projection

In order to reduce the scanning time and improve the reconstruction accuracy, the application of compressive sensing in MR image reconstruction has been widely studied recently. Most of the literatures formulate the reconstruction as a convex ℓ
₁
-minimization problem and show considerable improvement. However, some other researches show that non-convex penalties may induce better sparsity than convex ones, and achieve better reconstruction accuracy. In this paper, an efficient algorithm based on Zero-point Attracting Projection is proposed. Specifically, the concept of weak convexity is used to characterize the sparsity of image, and the reconstruction is formulated as a non-convex optimization problem, which is solved by generalized gradient projection. Experimental results demonstrate that the proposed algorithm exhibits better performance in terms of signal-to-noise ratio and artifact power compared with previous methods.