Coarse-to-Fine Deformable Model-Based Kidney 3D Segmentation

We present a 3D segmentation algorithm which can extract kidney from CT images. The contribution of this paper is twofold: one is a deformable model-based kidney segmentation algorithm which directly provides users with a surface mesh model of kidney; the other is a coarse-to-fine update method of deformable model by adding vertices which decreases computing resource consumption of the segmentation algorithm. This algorithm is based on a deformable mesh model that interacts with intensity value of CT images. The mesh model is firstly initialized as a tessellated sphere with whose center lying inside the kidney. After initialization, the mesh model evolves to fit the actual kidney surface over iterations. At the same time, as a coarse-to-fine process, the total of vertices in the mesh model gradually increases. The algorithm has been tested on data sets from a wide variety of scan parameters. Evaluation against manual segmentation was carried out. Jaccard similarity index and sensitivity are used to measure the performance of the algorithm. Experimental results show that the algorithm is accurate, yielding an average Jaccard index of 90%.