Deconvolution of direct reconstructions in 3D

Abstract

Recently, a direct reconstruction method for multi-dimensional magnetic particle imaging was proposed, which is based on the summation of weighted frequency components of the measured voltage signals with Chebychev polynomials of second kind. The method works fast but leads to reconstructions of the convolved spatial distribution of magnetic nanoaparticles. In a previous workwewere able to showthat using a neural network model to deconvolve these reconstruction leads to high quality images in the two-dimensional case. In this work, we take this approach one step further and demonstrate that this also applies to three-dimensional data. Therefore, in this work, we apply a neural network model on a simulated data set consisting of three-dimensional volumes containing blood vessel like structures. We show that the proposed network produces high quality deconvolution results and outperforms
conventional methods on the data set.