Image Reconstruction Based on a Cascaded neural network for Magnetic Particle Imaging

Abstract

The performance of image reconstruction method is critical in magnetic particle imaging (MPI), as it greatly affects the quality of MPI images. In this study, an image reconstruction method based on a cascaded neural network is proposed to obtain high-quality MPI images. In the initial stage, the measured signal is processed by the first network module to reconstruct a preliminary image that captures the global structural information of the imaging target. The preliminary image is then passed to the second network module to generate the final reconstructed output. The two-stage structure ensures that each subnetwork can focus on a specific subtask to improve overall reconstruction quality. Numerical simulations are performed to evaluate the performance of the proposed method based on different loss functions. We envisage that the proposed method is of great significance in advancing biomedical applications of MPI.