3D Shape Reasoning for Identifying Anatomical Landmarks

Orthopaedic surgeries require identification of anatomical landmarks on skeletal tissue. Manual palpation methods may not be accurate, especially in resecting deceased tissue and positioning custom implants or mega endo-prostheses. This article describes a computer-aided method for automatic identification of anatomical landmarks on a 3D model reconstructed from CT images of the patient. This is based on computing curvature values, identifying different regions of the surface (generated from the segmented volume data), and classifying the regions as peaks, ridges, pits and ravines. The landmarks are identified by walking along all vertices, extracting gradients from curvature map, and checking against a set of rules. The approach has been successfully demonstrated on a pelvis model reconstructed from 443 slices. The locations of the landmarks are determined, useful for dimension measurements and pre-operative planning.