Automated extraction of anatomic landmarks on vertebrae based on anatomic knowledge and geometrical constraints

摘要

The determination of anatomic landmarks is an essential step in modeling, shape analysis, segmentation, and registration of the spine, as well as in spine interventions. In this paper, we propose a method to automatically partition each vertebra into anatomic substructures and extract the relevant anatomic landmarks. Anatomic knowledge is integrated in geometrical modeling for the landmark localization. Each vertebra is automatically partitioned into four substructures, and 43 anatomical landmarks are identified. The locations were validated with manual annotation from medical experts. The results showed that the automatic landmarks agreed with manually labeled reference standard landmarks (1.9±1.3mm).