Automatic Segmentation of Extraocular Muscle Using Level Sets Methods with Shape Prior

摘要

Extraocular muscle segmentation based on magnetic resonance images is an important and relatively challenging task for clinical diagnosis and basic research. In this paper, we present a fully automatic segmentation method for extraocular muscle based on the level set scheme while integrating the prior information of shape and pose. A set of training data is manually segmented by experts in advance. We align them to each other and extract the shape priors of the segmented muscles by performing principal component analysis on their signed distance functions. Then we include the shape priors and pose information into the geodesic active contour model to derive the level set evolution. With this method, we success to segment each extraocular muscle automatically by preventing their basic structural outlines, even under the condition of weak images while extraocular muscle and neighboring tissues having the similar intensity. The results demonstrate the efficiency of the proposed segmentation procedure for dealing with the extraocular muscle segmentation.