Sparse-representation-based image reconstruction for magnetic particle imaging

Reconstruction methods play a significant role in magnetic particle imaging (MPI). In this study, a sparse-representation-based reconstruction method by utilizing the Gaussian radial basis functions (GRBFs) is proposed to improve the spatial resolution and artifacts of MPI images. Figure 1a shows the schematic of the proposed method. The spatial distribution of magnetic nanoparticles (MNPs) is sparsely represented by GRBFs. Consequently, the inverse problem is transformed to search for the optimal weight coefficient vector of the GRBFs. During iterative reconstruction, the center points of the GRBFs are adaptively selected. Simulation and experiments on single-harmonic-based narrowband MPI are performed to evaluate the performance of MPI images. Figure 1b shows the experimental MPI images reconstructed by the proposed method, the Kaczmarz method and the iterative Tikhonov method with gradients of 2.2 T/m and 4.4 T/m in x- and z-direction. The red dashed boxes represent the true distributions of two lines with a gap d ranging from 0.3 mm to 0.75 mm. The subfigures in Figure 1c show the three 1D curves along the white dashed lines in Figure 1b with d = 0.3 mm. It shows that the two lines with d = 0.3 mm gap can be distinguished by the proposed method, while they cannot be distinguished by the Kaczmarz and iterative Tikhonov methods. In addition, the images reconstructed by the proposed method show less artifacts compared with the other two methods. In conclusion, the proposed sparse-representation-based reconstruction method improves the spatial resolution and the artifacts for MPI