Automated localization of anatomical landmark points in 3D medical images

Abstract

Anatomical landmark point are 3D points in a well-defined anatomical structure in which correspondences between and within the population of the anatomical structure are preserved. Accurate delineation of the landmark points is crucial task for many medical imaging applications. However, in most current clinical applications, the anatomical landmark points are usually manually delineated by experts, which is time-consuming and irreproducible. In this study, an automated method for identification of anatomical landmark points in 3D medical images is presented. A 3D rotationally-invariant image descriptor was adopted to extract image information of pre-defined landmark points in a template image, and then use the information to identify corresponding landmark points in transformed images. This method was implemented and tested on 3D magnetic resonance images of human brain. Experimental results suggested this method can be potentially useful for identification of anatomical landmark points in 3D medical images.