Neural Network for Reconstruction of MPI Images

In Magnetic Particle Imaging the reconstruction of the image given the voltage signal is not trivial. Since the system function cannot be measured in its entirety the reconstruction algorithms can only estimate the images. The standard reconstruction approaches usually rely on time consuming optimization concepts that involve the use of sophisticated priors. We studied the general possibility of learning the reconstruction with neural networks. The results reveal that the networks potentially can reconstruct the images while even learning priors. However, the structures used for harvesting the training data should be chosen wisely.