Restructuring centipedes and their applications to fast extraction of structures in Electron Microscope tomography images

摘要

We present a class of region-based deformable contour models dubbed restructuring centipedes, R-centipedes. R-centipedes can automatically change their structures in order to extract multiple objects and their internal parts from complex medical and Electron Microscope (EM) tomography images. On the contrary to the topologically flexible edge-based T-snakes and the implicit level-set snakes, 1) R-centipedes do not rely on any image preprocessing such as the isotropic Gaussian smoothing and/or threshold setting needed for removing spurious and weak edges, and 2) they don't get stuck with textural artifacts which are abundantly found in EM tomography images. The R-centipedes have several desirable features, including geometrical flexibility, high speed, accuracy and insensitivity with respect to their initial positions, which make them useful tools in complex applications where demand is high for automatic object extraction and 3D visualization with non or minimal user interaction.