Automated registration of partially defective surfaces by local landmark identification

Abstract

Objective: In computer- and robot-assisted surgery, the term "registration" refers to the definition of the geometrical relationship between the coordinate system of a surgical planning system and that of the patient. Within the context of the development of a navigation and control system for computer- and robot-assisted surgery of the lateral skull base, it was desirable to realize an algorithm for automated registration of partially defective surfaces that is reliable and suitable for use in clinical practice. Materials and Methods: A registration algorithm based on the use of local fingerprints for specific points on a surface (so-called "spin images") was developed. Anatomical patient landmarks were identified automatically and assigned to CT data, performing a cross-correlation analysis and an investigation of the geometrical consistency. The algorithm was evaluated within the development of the navigation and robotic control system in a laboratory setting. Results: Under laboratory conditions it could be shown that partially defective surfaces (simulated by, for example, adding white noise, or reducing or smoothing the polygon data) were correctly recognized and thereby registered. In particular, the algorithm proved its excellence in interpreting partially modified topologies. Conclusions: The proposed procedure can be used to accomplish dynamic intra-operative registration of the skull bone by the generation of point relations to the CT images. © 2003 Informa UK Ltd All rights reserved.