MRI Tumor Image Segmentation by Parametric Kernel Graph Cut with Deformable Shape Prior

Abstract

Kernel graph cuts is one of the most efficient methods for image segmentation. However, kernel graph cuts for medical image segmentation without prior information is inefficient, especial for MRI tumor image segmentation. This paper presents a kernel graph cuts algorithm with deformable priors, which can successfully seize clinical MIR image features. The proposed networks for graph cuts are tailored to model the glioblastomas (both low and high grade) pictured in MR images for improvement accuracy performance. The experiment shows the success of the proposed approach.