A Deep Prior Approach to Magnetic Particle Imaging

Magnetic particle imaging (MPI) is a tracer-based imaging modality with an increasing number of potential medical applications exploiting the nonlinear magnetization behavior of magnetic nanoparticles. The image reconstruction is obtained by solving an ill-posed inverse problem requiring regularization. The number of data-driven machine learning techniques applying to inverse problems is continuously increasing. While more classical regularization techniques, e.g., variational methods, are commonly used in MPI, we focus on a novel deep image prior (DIP) approach. Initially developed for image processing tasks, it has been shown to be applicable to inverse problems. In this work, we investigate the DIP approach in the context of MPI. Its behavior is illustrated and compared to standard reconstruction methods on a 2D phantom data set obtained from the Bruker preclinical MPI system.