An effective reconstruction method for magnetic particle imaging based on deep neural networks constrained by a physical model

Abstract

Magnetic particle imaging (MPI) visualizes the spatial distribution of magnetic particles based on their nonlinear response signals. Such a process can be implemented by image reconstruction. In recent years, deep learning techniques supported by large amounts of training data have been widely used in various medical image reconstructions. However, the acquisition of MPI data requires a long time of obtaining and preprocessing. This makes it impractical to obtain a sufficient amount of data for training. In this work, we propose an MPI image reconstruction framework that incorporates physical model constraints into deep learning networks to overcome the limitation of dependence on training data. This framework optimizes the network parameters through constraints of the physical model rather than training with paired data. Simulation results show that this is an effective reconstruction strategy and has good reconstruction robustness.