Research On Image Segmentation Method Based On Level Set

Hippocampus, an important part in human brain, has a close relationship with the memory and emotion express. The geometric parameter changes in hippocampus volume and shape will result in some nervous system diseases, which makes whether the volume change occurs becomes the clinical criterion for the above disease, thus hippocampus’volume measuring has significant meaning. How to extract2D silhouette, the basis for hippocampus volume measuring and3D rebuilding, has become the highlight for many researchers. Considering level set is beyond any other methods, this paper exploits it to segment hippocampus from MRI images. The main jobs of the paper are as following:1) Hippocampus is only a small region in MRI image. In order to reduce the computation and improve the segment reliability, this paper only process the certain part, with the size of20*20, including the whole hippocampus and little other region.2) The original contour curve is defined as the uniform and simply closed graphic in traditional level set, which leads to many derivations. To solve this problem, adaptive region growing and morphological opening are adopted in this paper to realize the coarse segmentation for hippocampus MRI images, which generates a convex polygon including the entire coarse segmentation as our original outline. This will provide the prior-knowledge for us and improve the segmentation accuracy.3) In traditional method, the zero level set curve is apt to converge at the gray gradient extremum of background with salient features. An improved level set method based on global variance is proposed to solve this problem. The wave energy added to external energy function can make the zero level set curves beyond the untargeted zone with violent gray fluctuation and finally converge at the extremum with small gray fluctuation. Additional advantages are less computation time cost and speeder segment efficiency.4) Edge detection was used to validate the converge accuracy of level set. The original contour will be adjusted for new evolution when zero level set has un-segmented region, which makes the zero level set finally stay at the real borders of hippocampus and grantee the segmentation accuracy as well.