Tumor detection on brain MR images using regional features: Method and preliminary results

This paper presents a novel approach to detecting tumor in the brain magnetic resonance images using regional features. First, the proposed algorithm segments head area and skull area using average of brain magnetic resonance images and local adaptive threshold technique. Next, super-pixel segmentation algorithm is applied in order to generate categorized regions on the segmented brain image. Second, we extract regional features, which are texture feature and intensity. Finally, the support vector machine classifier detects the tumor regions by integrating candidates of tumor, which are computed from categorized regions according to different super-pixel parameters. The scheme successfully detects tumor region on the 60 brain magnetic resonance dataset.