MRI brain tumor segmentation using GLCM cellular automata-based texture feature

Brain tumor segmentation is an importance process in surgical and treatment planning in medicine. There are various intensity based techniques which have been proposed to segment homogeneous tumors on magnetic resonance images (MRI). Those still fail to segment homogeneous tumor against similar background, isointense signal, weak edges or diffused edges. These problems lead to oversegmentation by intensity based techniques. This paper presents a cellular automaton (CA) based on Gray-level co-occurrence matrix (GLCM) for determining the local transition function. This aims to extract the texture feature of MRI brain tumor being used for brain tumor segmentation. The state-of-art segmentation methods, namely, Tumor-Cut (TC) and active contours driven by local Gaussian distribution fitting energy (LGD), are compared the results between intensity image and the proposed texture-based image. For performance evaluation, MRI tumor datasets acquired from virtual skeleton database (VSD) are experimented throughout. Dice similarity coefficient (DSC) and Jaccard coefficient (JC) are employed to measure the overlapping region between tumor ground truth and segmentation results. In this regard, TC and LGD algorithms using the proposed texture feature provide the better results. TC with the proposed texture feature provides the best result with DSC and JC at 91.89% and 85.08%, respectively.