Automatic Intensity Graph Method for Brain and Tumor Segmentation Using Density Measures

Abstract

Brain and tumor segmentation has been supported by different methodologies earlier, but struggles with identifying the spatial deformations and identifying the exact boundary regions of the tumor which affects the classification accuracy. We propose a new automatic intensity graph method for brain and tumor image segmentation. Despite its simplicity, this application utilizes the best features of combinatorial graph cuts methods in vision: global optima, practical efficiency, numerical robustness, ability to fuse a wide range of visual cues and constraints, unrestricted topological properties of segments, and applicability to N-D problems. The proposed graph based method also utilizes the region properties for the segmentation problem. At the first stage intensity normalization is performed with the histogram equalization techniques, and then the image is converted into gray scale and based on the gray values similar valued pixels are grouped and constructed as a graph. The segmentation performed with the intensity graph method produces efficient results. Once segmentation is performed, different segmented region is represented with different color values. Obviously the region affected with tumor has different intensity values, using which the tumor tissues are identified.