Biomedical image segmentation via constrained graph cuts and pre-segmentation

In this paper we present a high-fidelity method for 2D and 3D image boundary segmentation. The algorithm is a novel combination of graph-cuts and initial image segmentation. The pre-segmentation using anisotropic vector diffusion and the fast marching method is employed so that the size of the graph being considered is significantly reduced. To further improve the segmentation accuracy, some user guidance is taken into account in finding the minimal graph cut. To this end, a user-friendly graphical user interface (GUI) is developed not only for visualization purposes but for user input and editing as well. The approaches and tools developed are validated on a number of 2D/3D biomedical imaging data, showing the high efficiency and effectiveness of our method.