Automatic Brain Tumor Segmentation using Symmetric Superpixels and Deformable Model in MR Images

Brain tumor segmentation plays an important role for urgent diagnosis and radiotherapy planning, but automatic segmentation of brain tumor because of natural heterogeneity, various shapes and ambiguous boundaries of tumors has still remained as a challenging task. In this paper, an automatic method based on symmetric superpixels (SPs) and deformable model has been proposed for brain tumor segmentation from magnetic resonance (MR) images. We create symmetric SPs using symmetry axis and then we compare each superpixel (SP) with its corresponding symmetric SP to find asymmetric areas which can be used as initial set for the deformable model. Utilizing SPs as atomic units in the symmetry analysis yields robust segmentation results. This hybrid method is evaluated using glioma images from BRATS 2012 dataset and the results showed that it has reasonable accuracy and similarity to the ground truth data.

Keywords:

Brain Tumor Segmentation , Symmetric Superpixels , Region-Based Active Contour , Magnetic Resonance Imaging (MRI)