A Novel Automatic Liver Segmentation Technique for MR Images
Abstract—This paper presents an automatic liver segmentation algorithm based on fast marching and improved fuzzy cluster methods, which can segment liver from abdominal MR images accurately. The developed method is composed of three major steps: first, fast marching method and convex hull algorithm are applied to roughly extract the liver’s boundary and topology, which provides a basic estimation for the subsequent calculations; second, an improved fuzzy cluster method, combining with a multiple cycles processing, is designed to refine the segmentation result; third, based on the segmented results, the liver is visualized by Marching Cube method. There are two major difficulties in MRIs liver segmentation: one is that the boundaries between liver and adjacent tissues generally have uniform intensity distributions, which often leads to over segmentation of the liver; the other is that inner vascular inside the liver commonly leads to segmentation leakage. In order to solve these two problems, a prior knowledge based fuzzy cluster method is proposed. Experiments show that the developed method is effective and robust for liver segmentation of MR images.
