Segmentation of kidney by using a deformable model

This paper presents a deformable model based approach for segmenting three dimensional (3D) anatomic volumes (such as human organs) in medical images. The deformable model is represented by non-uniform rational B-splines surface and a priori knowledge of an organs shape, such as average and variation of the shape, is introduced as energy function for the model fitting process. To describe the shape, we used principal curvature and its statistical information which are invariant for rotation and translation. The proposed method is applied to the segmentation problems of the kidney in 3D CT images and the promising results are shown.

1. Introduction

With an increase of 3D medical images in the diagnosis and treatment of disease, automatic or semiautomatic segmentation procedures of human organs are required for supporting of observation and navigation by physicians. Segmenting a 3D image slice by slice by applying 2D techniques is a laborious processing and requires the process of connecting them each other to reconstruct continuous surface. Recently, the use of 3D deformable surface models [1] has offered accurate and robust approach which ensures adjacent extracted regions smoothly connected and the results seem to be superior to the conventional methods, such as region growing approach.