Brain Tumor Segmentation Using Geodesic Region-based Level Set without Re-initialization

Abstract

Region-based level set segmentation is a paradigm for the automatic segmentation of brain tumor image. Unfortunately, region-based segmentation, which is relied on the intensity difference of different regions, has been of limited used in presence of complex background. In fact, the evoluting curve may leak out the boundary of tumor to reach a steady state by the global region force. In this work, we propose a new hybrid approach for brain tumor segmentation, which is relied on the approach of global intensity difference, local edge properties, curve evolution, and level set method. The regional information drives the contour to converge to the global minimum. By combining the edge information into the region-based framework, the images with intensity inhomogeneity and complex background can be efficiently segmented. To improve the accuracy of brain tumor segmentation, a skull-stripped method for brain images is proposed by utilizing the new morphological process. In addition, a penalizing energy is used for avoiding the time-consuming re-initialization step of the level set method. Finally, experiments are preformatted on some synthetic and real images. By visually assessments, results on patients demonstrate the new method can segment tumors with few iteration times. Moreover, comparisons with the most similar methods also show that the proposed method is effective for the segmentation of tumor in MR image.