Rough segmentation of coherent local intensity for bias induced 3-D MR brain images

Segmentation of brain MR volumes into different meaningful tissue classes is an essential prerequisite for many clinical analyses. However, intensity inhomogeneity or bias field, present in MR volumes, considerably degrades the quality of segmentation. In this regard, the paper presents a new segmentation algorithm, termed as CoLoRS (Coherent Local Intensity Rough Segmentation), for brain MR volumes corrupted with bias field artifact. It judiciously integrates the merits of coherent local intensity clustering and the theory of rough sets for simultaneous segmentation and bias field correction of brain MR volumes. The proposed algorithm partitions the entire image space into a number of small overlapping neighborhood regions. The bias, in each neighborhood region, is assumed to be constant. For each individual region, an objective function is defined for coherent local intensity rough segmentation. The voxels near the center point have similar influences on local objective function. In addition, the smaller distance between center and neighboring voxels yields more contribution on the voxel of interest. The proposed algorithm uses the dual-region concept to represent the neighborhood structure more efficiently. It makes possible of separate modeling of the voxels within neighborhood, according to their locations. Each region is considered to have several tissue classes, where each tissue class consists of a core region and an overlapping region. The segmentation in fuzzy approximation spaces provides an effective mean for brain MR volume analysis, as it handles overlapping partitions and addresses vagueness in tissue class definition. The effectiveness of the proposed algorithm, along with a comparison with existing approaches, is demonstrated on several publicly available brain MR data. (C) 2019 Elsevier Ltd. All rights reserved.