Automated Localization of Pelvic Anatomical Coordinate System from 3D CT Data of the Hip Using Statistical Atlas

Abstract

Localization of the pelvic anatomical coordinate system is a prerequisite for patient-specific preoperative planning and joint motion simulation for hip surgery. Our aim is to automate localization of the pelvic anatomical coordinate system from 3D CT data. In this paper, we propose a statistical atlas-based method that consists of three steps. The first step is spatial normalization using a probabilistic atlas. The second step is feature point recognition using a statistical landmark model. The final step is coordinate system refinement using a standard CT atlas. We applied the proposed method to 39 datasets. Compared to manual localization by an experienced surgeon, the average positional error was 2.37 ± 1.30 mm and the average orientation error was 1.07 ± 0.50 degrees. These results demonstrate the usefulness of the proposed method.