ABSTRACT
GOAL: The purpose of this paper is to develop a computational approach to the segmentation of human orbits. METHODS: The first step is to perform Hounsfield units thresholding to segment the bony structure around the orbit. Then, a three-dimensional mesh model is generated. Poisson surface reconstruction is applied to a set of automatically screened vertices, which are facing the inner orbital walls. These procedures effectively close orbital fissures; various nerves foramina; and interpolate the broken surfaces due to thin bone structures around the orbit. We also developed validation models with five dried skulls, where the orbits were filled with dental impression. Validations on the proposed algorithm were performed with the corresponding CT images and verified by experienced radiographer. RESULTS: The mean volume differences are less than 0.3%. Surface differences are within 0.3 mm of root mean square. Both differences are not clinically significant. SIGNIFICANCE: Traditional approaches are slice-by-slice manual editing or shape interpolation with selected slices interactively. It is not only time consuming, but also inefficient, exhibits interoperator variability, and repeatability problems. In the proposed method, most of the manual processes are eliminated with adjustable vertex screening parameters. It makes the proposed method repeatable.
