Image segmentation of embedded shapes using constrained morphing

Abstract

An imaging system and method enables 3-D direct segmentation from a series of spatially offset 2-D image slices in a volume scan. The algorithm first smoothes and preserves the interface edges of the image volume using Bottom-Hat gray scale morphological transform followed by 3-D segmentation using fast 3-D level sets by preserving topology constraints, for example, cortical thickness in a brain volume. The method inputs opposite polarity spheres (contracting and expanding spheres) which morph into shapes within the volume using a surface propagation technique. The speed of propagation is controlled by the likelihood statistical component derived under constraints. During the propagation polygonalization extracts the zero-level surface set. The field distribution is computed using the improved shortest distance method or polyline distance method. The morphing algorithm then morphs the input concentric spheres into interface surfaces such as WM-GM and GM-CSF with cortical constraint. The system is optimized by computing the 3-D field in the narrow band on the morphing spheres.