Cerebral edema segmentation using textural feature

Diagnostic imaging provides a vital tool in detection and analysis of Brain pathologies. Magnetic resonance imaging (MRI) provides an effective means for non-invasive mapping of anatomy and pathology in the brain. Pathologies like cerebral edema and tumors can spread in different tissues in the brain and can affect cognitive and other functions in the body. Accurate segmentation is therefore a challenging task. Human Brain consists of different soft tissues. These tissues can be characterized using different textures. The work presents an automatic method for segmentation using textural feature of the MR image. The texture of MR image is exploited using the gray co-occurrence matrix (GLCM). GLCM creates a textural feature map by taking into account the spatial dependence of the pixels and its angular relationship between the neighboring cell pairs. Local entropy as second order textural feature is used to capture the texture of MR image. Entropy computes the randomness in pixel intensities and helps in defining a unique texture of edema for segmentation. The marked contrast enhancement obtained in FLAIR sequence of the MR image is captured as textural information by local entropy and GLCM combination. The proposed method obtains a definite textural signature of edema as well as tumor for threshold selection. Experiments on publically available BRATS database yields an average accuracy of 96%, specificity of 97%, sensitivity of 61%, Dice Coefficient as 50% and structural similarity index of 0.88 for edema. The proposed method demonstrates encouraging results in automatic segmentation of edema as well as tumor core. (C) 2019 Nalecz Institute of Biocybernetics and Biomedical Engineering of the Polish Academy of Sciences. Published by Elsevier B.V. All rights reserved.