Tumor delineation from 3-D MR brain images.

In cancer research, automatic brain tumor detection from 3-D magnetic resonance (MR) images is an important pre-requisite. In this regard, the paper presents a new method for segmentation of brain tumor from MR volumes corrupted with different imaging artifact, such as bias field and noise. It addresses the problems of uncertainty and bias field artifact of brain MR images by means of a segmentation algorithm, termed as CoLoRS (Coherent Local Intensity Rough Segmentation). However, the lack of knowledge about the tumor intensity and the textural properties around tumor surface makes the exclusion or inclusion of tumor or healthy tissues, respectively, in the tumor region obtained by CoLoRS. Therefore, a post-processing technique is introduced for precise delineation of tumor region from healthy brain tissues. It mingles the benefits of morphological operations and theory of rough sets into the region growing approach to improve the result of tumor detection. Several publicly available MR brain tumor data sets, namely, BRATS 12, BRATS 14 and BRATS 19, are used to demonstrate the effectiveness of the proposed method with respect to existing approaches. For real high-grade and low-grade data sets of BRATS-12-14, the Dice coefficient of the proposed algorithm is 0.783443 and 0.787045, respectively, and for BRATS-19, 0.788849 and 0.808582, respectively. [ABSTRACT FROM AUTHOR]