Reduction of Noise and Artefacts in MRIScan of Metal Implant Reconstructed by Frequency Mapping

Abstract: Metal objects can cause significant artefacts in magnetic resonance imaging (MRI) scans, cause loss of signal and what is known as geometric warping. There are ways to address these problems using different MRI sequences. This project looks at a newer method of artefact correction proposed by Månsson and Morin. The aim is to improve upon the main two faults of the method. The method makes use of a frequency map taken alongside the image dataset which describes the warping of the magnetic field caused by the metal object. This frequency map is then used to estimate how far data should be shifted back to correct for the geometric warping. The data is then shifted back using Fourier Shift. The main limitations of the method are a noticeable increase in noise present in the image post correction as well as the unexpected appearance of zebra-stripe artefacts. The improvements are made from two parts: filtering and interpolation. These methods were tested on both synthetic datasets generated using MATLAB, as well as a previously created phantom. By filtering the frequency map using a three-dimensional median filter prior to applying the correction method, the noise could be reduced by about a factor 2 in the final image. Additionally, interpolating both datasets prior to applying the method caused a significant reduction to the zebra-stripe artefacts.