Correction of geometric distortion in MRI

This Thesis presents an algorithm for correcting MR field inhomogeneities that result from the presence of tissues having different magnetic susceptibilities. To correct for distortions due to inhomogeneities, this thesis provides an algorithm that works in the following manner. Based on a phase field map, the magnetic field distortion DeltaB is determined. Phase unwrapping is used on the phase images. The phase images are then smoothed in order to remove noise and it is found that a gaussian filter using pixel connectivity gives the best performance. Alternatively, it is also possible to generate a smoothed version of the field map by fitting to it a third or fourth order 3-D polynomial surface. The main advantage of fitting over filtering is that the magnetic field can be estimated in regions where accurate measurements are not possible. Finally, using the estimated magnetic field distortion map, a map of one dimensional pixel shifts along the frequency encoding direction can be estimated and interpolation is used to correct for the distortion. Several interpolation paradigms are compared in their performance and it is found that the Windowed Sinc Interpolation and direct geometric interpolation gave the best results. Simulations with phantom data show that through the use of this algorithm it is possible to correct for the magnetic field distortions. We have showed that this method leads to improvement in the visibility of in vivo human brain images. Results presented here demonstrate that magnetic field distortion correction can be useful within the context of accurate MRI-based neurosurgery planning. We have also made an effort to quantify the remnant errors in the method and found them to be dependent on the geometry under consideration. In phantom studies, the error in position was corrected to 4%. We found this to be true in human images in regions of slow field changes with errors up to 30% in pixel location for rapid phase changes.