Efficient cellular automaton segmentation supervised by pyramid on medical volumetric data and real time implementation with graphics processing unit

In surgery simulation, the extracted tissue data can be operated repeatedly in a Virtual-reality (VR) system which provides a good alternative to classical training method. Fully automated segmentation techniques cannot guarantee the efficiency and precision in general case. This paper describes a user interactive segmentation method: given a labeled 2D image plane in Multi-Planar Reformation (MPR), the rest tissues are segmented automatically by a cellular automaton in multi-scale domain. Labels image generated in higher level Gaussian pyramid can be extended to lower level ones according to selected resolution. An edge indicator function is also set to avoid over-segmentation in Laplacian pyramid. The evolution can be observed and guided with volume rendering results. The proposed method shows the merits of higher precision, real time response in GPU framework and few interactions are required.