Dipole-Based Filtering for Improved Removal of Background Field Effects from 3D Phase Data

Introduction: Phase images of the brain generated using gradient echo sequences at high field strength show excellent contrast attributed to small field perturbations produced by susceptibility differences between tissues. In order to reveal the small phase variations related to the local anatomy it is however necessary to eliminate the large field offsets caused by remote tissue/air interfaces, such as those that are present in the sinuses. These unwanted field shifts are generally removed by filtering the phase images using a variety of methods, including: (i) Fourier-based filtering where the data is converted into complex form, low-pass filtered, and divided into the original data [1]; (ii) subtraction of a low-order polynomial fit to the data [2]; (iii) subtracting off an estimate of the field perturbation produced by the sinuses based on information from T1-weighted images [3]. Unfortunately, methods (i) and (ii) tend to remove the desired anatomical structure as well as the unwanted external fields, while method (iii) requires additional image data from which the location of the sinuses can be inferred. Here we present a method for selectively eliminating the externally generated field shifts without the need for additional scans, using one or more dipole point sources situated outside the region of interest (ROI) being considered. The method was tested on simulated and experimentally acquired phase data spanning the red nucleus (RN) and substantia nigra (SN), which show the effect of strong unwanted fields due to the close proximity of the sinuses.