Extraction of Anatomic Structures from Medical Volumetric Images

Abstract

In this paper, we present the extraction method of anatomic structures from volumetric medical images using the level set segmentation method. The segmentation step using the level set method consists of two kinds of processes which are a pre-processing stage for initialization and the final segmentation stage. In the initial segmentation stage, to construct an initial deformable surface, we extract the two dimensional boundary of relevant objects from each slice image consisting of the medical volume dataset and then successively stack the resulting boundary. Here we adopt the statistical clustering technique consisting of the Gaussian mixture model (GMM) and the Deterministic Annealing Expectation Maximization (DAEM) algorithm to segment the boundary of objects from each slice image. Next, we use the surface evolution framework based on the geometric variation principle to achieve the final segmentation. This approach handles topological changes of the deformable surface using geometric integral measures and the level set theory. These integral measures contain the alignment term, a minimal variance term, and the mean curvature term. By using the level set method with a new defined speed function derived from geometric integral measures, the accurate deformable surface can be extracted from the medical volumetric dataset. And we also use the Fast Matching Method that can reduce largely the computing time required to deform the 3D object model. Finally, we use the marching cubes algorithm to visualize the extracted deformable models. The experimental results show that our proposed method can exactly extract and visualize the human organs from the CT volume images.