Vertebrae localization in CT using both local and global symmetry features

Abstract

Automatic vertebrae segmentation and localization in CT images are essential in many medical treatments such as disease diagnosis and surgical planning. However, vertebra is one of the most complex organs to locate precisely due to its complex shape, deformation and occlusion by other organs. In this paper, we propose to incorporate local appearance features with global translational symmetry and local reflection symmetry features. Symmetrical structure of each vertebra provides strong cue for accurate localization. In order to efficiently investigate 3-dimensional reflection symmetry in CT images, we propose a Sphere Surface Expansion method and iterative optimization framework. Quantitative and qualitative evaluations show that the proposed method outperforms existing localization method.