Data-driven image reconstruction for Low-field MRI

The purpose of this research is to investigate how much we are able to improve the reconstruction of images from a low-cost MRI-scanner prototype aimed to be employed in developing countries. We used a data-driven approach searching for linear combinations of signals from a small measured data-set to form reconstructed images. This was done by solving linear least-squares problems, both with and without regularization, with the help of the Moore-Penrose pseudo inverse. This small and unoptimized data-set shows promising results and expanding the data-set with simulated data suggests that further improvement of the results can be achieved. A model-based approach is impossible without sufficient knowledge of the magnetic field map. Using simulated data we have shown that inaccuracies in the field map can be corrected from the information within the data. When applied to a measured data-set, however, we were not able to improve magnetic field map enough to allow for successful reconstruction.Applied Mathematics | Applied Physic