On the importance of MRI nonuniformity correction for texture analysis

With the great progress in magnetic resonance imaging of the last decades, computerized image texture analysis has become an appealing approach to image quantification - for more objective medical diagnosis and therapy outcome evaluation. The number of research papers on TA applications to MRI of the bone, brain, breast, liver, myocardium and prostate continues to increase. MRI texture descriptors are computed to characterize spatial patterns of the image intensity that originate in the structure of tissues. This is a rich source of information, not available through standard quantification, based on single-voxel intensity only. However, the large number of texture descriptors depend not only on the texture, but also on local average image intensity. Since variations of intensity, known as image nonuniformity (inhomogeneity) artifacts, often occur in MRI, their presence may severely interfere with texture analysis and, consequently, lead to errors in tissue description. The origin of this unwanted effect is explained using the popular gray-level-cooccurrence matrix texture description. The intensity-caused errors are discussed and illustrated using MRI slices of a normal and fibrotic liver. To reduce those errors, correction of image spatial nonuniformity prior to texture analysis is recommended as an image preprocessing step, especially for large-volume regions of interest, featuring homogeneous texture.