A basic study of an image reconstruction method using neural networks for magnetic particle imaging

In magnetic particle imaging (MPI) [1], image artifacts and blurring appear on a reconstructed image such that magnetization signals generated from magnetic nanoparticles (MNPs), which exist at the boundary of the field free point, are also detected. In order to overcome these problems, we propose a new reconstruction method using neural networks [2]. This proposed method can estimate MNP distribution based on a data set of input (system-functions) and desired-output (MNP-location) pairs used as the teaching data for a neural network. By employing neural networks, we expect to suppress image blurring by learning a sufficient number of data sets to indicate a relationship between image blurring and the corresponding MNP location. We perform numerical experiments to confirm the effectiveness of this method.