Meeting Banner
Abstract #2439

Evaluation of MR Image Intensity Inhomogeneity Correction Algorithms

Jinghua Wang1, Lili He2, Zhong-lin Lu3

1Center for Cognitive and Behavioral Brain Imaging, The Ohio State Univeristy, Columbus, OH, United States; 2Center for Perinatal Research, Nationwide Children's Hospital, Columbus, OH, United States; 3Center for Cognitive and Behavioral Brain Imaging, The Ohio State University, Columbus, OH, United States

MR image intensity inhomogeneity has rendered quantitative MRI analysis in anatomical studies a major challenge. Various methods for performing inhomogeneity correction have been proposed. Evaluation of these methods has often relied on subjective assessment because of the lack of the ground truth. Here, we present a new method to evaluate four popular inhomogeneity correction methods based on both uniform phantom and in vivo brain images. We found that the field map method outperformed the others. This method can be used to guide parameter optimization for existing correction methods, improve bias field modeling, and evaluate and optimize new correction methods.

Keywords

abstract anal anatomical applied artifacts assessment assumption become behavioral best bias binary blur blurry boundaries boundary brain carried challenge channel children coefficient cognitive coil come computed contain contrast corrected correction defined dependent deviation edge effective evaluate evaluation example existing fast field generally good gradient gray greatly ground guide hand head hospital human ignored imperfect improve improved in vivo inhomogeneity inhomogeneous intensities intensity introduce introduced inversion isolated lack lead limited location mainly major masks matrix minimal modeling multiplying nationwide noise obviously often optimal optimization optimize others outperformed particularly percent performance perinatal phantom popular procedures processing produce produced proposed qualitatively quality quantifications quantify quantifying quantitative quantitatively reasonable recovery reduced reliable relied remarkably rendered residue resolution satisfactory segmentation segmented severe sharp simulated slice smoothed software strategy studies subjective subjects synthetic system table tissue tissues trio truth type uniform uniformities uniformity unrealistic validation validations variation various volumes volunteers white whole wicks