A comparison of retrospective intensity non-uniformity correction methods for MRI

Abstract

Intensity non-uniformity is an artifact often seen in MRI data that can significantly degrade automatic segmentation and prevent quantitative analysis. Using simulated MI~I data, we have quantitatively compared the performance of three methods of correcting for intensity non-uniformity. In this analysis, novel stereotaxic space techniques were employed so that all three methods were fully automatic and capable of correcting volumes of any pulse sequence. Our results, based on correcting simulated T1, T2, and proton density (PD) weighted MRI scans with various levels of noise and non-uniformity, showed no method to be superior in all categories. However, the N3 technique distinguished itself for uniformly good performance without the limitation of assuming a sophisticated model of the data.