Quantitative study of renormalization transformation method to correct the inhomogeneity in MR images

The purpose of this work is to evaluate the effectiveness of using a newly proposed renormalization transformation (RT) technique to correct nonuniformity in MR images. Simulated brain T1, T2 and PD weighted images with two types of bias fields and Gaussian white noise were created using the average signal intensities of white matter, gray matter, and CSF from segmented masks of actual patient examinations. These images were then corrected by the RT method and quantitatively compared with the original non-biased simulated images. This study demonstrated that a single optimal correction exists for the RT method. At the optimal correction, the RT method can remove more than 75 percent of the bias field without significant loss of useful contrast in the images. Unfortunately, this optimal correction can not be directly determined for actual patient images where the truth is not known. However, simulated images showed that the optimal correction could be estimated from changes in the contrast ratio map, where the contrast ratio is the ratio of the local intensity standard deviation and local average intensity. Using the contrast ratio map, the optimal correction can be reliably applied in patient images.