Automatic brain tumor segmentation based on features of separated local square

Abstract

Brain tumor segmentation from magnetic resonance images is a critical step for early tumor diagnosis and treatment. However, accurate and general segmentation of brain tumor is still a challenging task due to complicated characteristics of brain tumor in magnetic resonance images. To solve this problem, we proposed a novel method for brain tumor segmentation based on features of separated local square. The proposed method basically consists of three steps: superpixel segmentation, feature extraction and segmentation model construction. In the first step, the superpixel segmentation algorithm was used to partition an image into homogeneous regions with approximately similar characteristics and size. In the second step, gray statistical characteristic and texture features were extracted. Finally, by applying support vector machine (SVM), superpixels were classified as either tumor regions or nontumor regions in the segmentation model construction step. The brain tumor segmentation method takes the advantage of the superpixel-wise features and the ability of SVM. For verifying the proposed method, we select 20 images from BRATS 2013 dataset. The experimental results demonstrate the effectiveness and robustness of our proposed method.