Detecting open surfaces in three dimensions

Abstract

We present a novel level-set method for representing and detecting open surfaces embedded in three-dimensional image volumes. Open surfaces are two-dimensional manifolds with a one-dimensional boundary lying within a three-dimensional volume. Distinct portions of a closed surface can be modeled as open surfaces, as can very thin volumes with negligible thickness. To detect open surfaces, we propose an interface likelihood model that captures the image appearance along a profile normal to the open surface. This allows statistical modeling of more complex surface-appearance characteristics than just voxel intensities or gradients. Appearance models of the surface are used in the level-set framework in two ways: firstly, to evolve the open surface in the normal direction for the purpose of detecting the location and shape of the surface, and secondly, to evolve the boundary of the open surface in a direction tangential to the surface in order to delineate the extent of the surface. We show that our models are well suited to detecting structures of interest in three-dimensional medical and geological images, and demonstrate their utility on challenging structural magnetic resonance (MR) datasets and seismic-reflection volumes.