Correction of inhomogeneities in Magnetic Resonance Images

This paper presents a method by which two fast spoiled gradient echo image volumes can be used to estimate the slowly varying intensity variation frequently observed in Magnetic Resonance Images (MRI). We first present results for simulated brain MRI, where we compare the performance of a segmentation algorithm on uncorrected, corrected and artifact free images. Next, we present the results for a real uniform phantom and then for clinical spinal images, where our algorithm reduces the intensity variation within a tissue class from a factor of 30 to a factor of 2. The technique and algorithm has been successfully tested on over 1000 images acquired with different coils, sequences and anatomies.