Image Reconstruction in Magnetic Particle Imaging Based on Gaussian Weighted Laplace Prior Regularization

Magnetic particle imaging (MPI) is an emerging tomographic imaging modality with high spatial and temporal resolution. The image reconstruction in MPI needs to solve an ill-posed inverse problem. Tikhonov regularization is known to help solve this kind of problems. However, the traditional Tikhonov regularization (L2 regularization) guides the reconstruction to be over smooth and tends to generate a lot of low-value scattered signals around the real reconstructed objects. In this work, we develop an efficient and noise reducing reconstruction method for MPI. We propose a Gaussian weighted Laplace regularization which assumes that the correlation between any two voxels inside the field of view (FOV) has a non-linear inverse relationship with their spatial distance. Experimental results show that the proposed method can provide a more accurate MPI reconstruction.