Background signal suppression using a transformer-based masked autoencoder for magnetic particle imaging

Abstract

Magnetic particle imaging (MPI), an emerging imaging technique, utilizes the nonlinear response of superparamagnetic iron oxide nanoparticles to generate an image of their spatial distribution. To obtain the high quality MPI images, it is necessary to suppress the background signal. In the previous work, we have proposed a deep learning based method, which can effectively suppress different background signal at different level simultaneously. In this work, we further designed a transformer-based masked autoencoder to learn the relationship between different harmonic components for better noise suppression. The experiments show that the proposed method can effectively suppress background noise at different levels. Besides, our method can reduce the network's dependence and demand on the amount of datasets.