Brain Tumor Segmentation Using Random Walks from MRI Images

Magnetic resonance imaging (MRI) is a vital and universally recognized medium to assess brain neoplasms. This paper presents a study on brain tumor segmentation based on the random walk algorithm which is a graph-based method in which pixels of a brain MR image are treated as nodes. Segmentation is performed by interactively labeling certain nodes as foreground and background seeds, followed by computing the probability of each unlabeled node to reach all the labeled nodes using random paths. The method is applied on two different MR modalities viz. T2-weighted MRI with fluid attenuated inversion recovery (FLAIR), and T2 MRI to segment complete tumor, and tumor core regions, respectively, by utilizing visual traits of MRI images and identifying local and global brain tissues information. Efficacy is validated quantitatively as well as qualitatively through performing the experiments on publicly available brain tumor segmentation challenge (BRATS-2013) dataset. Results demonstrate that the proposed method performs favorable as compared to several existing methods.