An Automated Segmentation Method of Kidney Using Statistical Information

This paper presents a deformable model based approach for automated segmentation of kidneys from tree dimensional (3D) abdominal CT images. Since the quality of an input image is very poor and noisy due to the large slice thickness, we use a deformable model represented by NURBS surface, which uses not only the gray level appearance of the target but also statistical information of the shape. A shape feature vector is defined to evaluate geometric character of the surface and its statistical information is incorporated into the deformable model through an energy formulation for deformation. Principal curvature on the model surface, which is invariant to rotation and translation, is adopted as a component of the vector. Furthermore, automated positioning procedure of an initial model is presented in this paper. We applied the proposed method to the 33 abdominal CT images whose slice thickness is 10mm and evaluated the effectiveness of the proposing method.