A Brain MR Image Segmentation Approach Based on Local Intensity Fitting Curve Evolution

The segmentation of brain magnetic resonance (MR) images is of great significance in research and clinical applications, including diagnosis of pathology, presurgical planning and computer integrated surgery. To cope with the inevitable intensity in homogeneity problem of MR images, based on the techniques of curve evolution and level sets, a novel curve evaluation method employing neighboring information was proposed in this paper. An energy function was defined with a local intensity fitting term, which adopted the local information in the neighborhood of the interested pixels, and greatly improved the boundary recognition accuracy for low contrast areas. A numerical algorithm using finite difference is also presented. Finally, the method was tested by several experiments on MR images. Experimental results indicate the proposed algorithm is effective for segmenting MR images corrupted by intensity inhomogeneity, and usually outperforms the corresponding conventional methods. The proposed model could also be applied in segmentation of ordinary inhomogeneous images which is very common in reality.