gms | German Medical Science

58. Jahrestagung der Deutschen Gesellschaft für Neurochirurgie e. V. (DGNC)

Deutsche Gesellschaft für Neurochirurgie (DGNC) e. V.

26. bis 29.04.2007, Leipzig

The accuracy of registration algorithms and registration error estimators in neuronavigation

Genauigkeit von Registrierungsalgorithmen und der Registrierungsfehler-Schätzer in Neuronavigation

Meeting Abstract

  • corresponding author D. Paraskevopoulos - Neurochirurgische Universitätsklinik Heidelberg
  • P. Oberhammer - Medizinische Informatik, Universität Heidelberg
  • C. R. Wirtz - Neurochirurgische Universitätsklinik Heidelberg
  • U. Eisenmann - Medizinische Informatik, Universität Heidelberg
  • R. Metzner - Medizinische Informatik, Universität Heidelberg
  • H. Dickhaus - Medizinische Informatik, Universität Heidelberg
  • A. Unterberg - Neurochirurgische Universitätsklinik Heidelberg

Deutsche Gesellschaft für Neurochirurgie. 58. Jahrestagung der Deutschen Gesellschaft für Neurochirurgie e.V. (DGNC). Leipzig, 26.-29.04.2007. Düsseldorf: German Medical Science GMS Publishing House; 2007. DocP 015

Die elektronische Version dieses Artikels ist vollständig und ist verfügbar unter:

Veröffentlicht: 11. April 2007

© 2007 Paraskevopoulos et al.
Dieser Artikel ist ein Open Access-Artikel und steht unter den Creative Commons Lizenzbedingungen ( Er darf vervielfältigt, verbreitet und öffentlich zugänglich gemacht werden, vorausgesetzt dass Autor und Quelle genannt werden.



Objective: Registration is an essential part of navigation procedures and since precision is crucial in neurosurgery, the surgeon needs an accurate estimation of the achieved accuracy. Thus the main goals of this study were to compare algorithms, achieve an adequate estimation of registration error and optimize the occurred error.

Methods: A registration framework was implemented, allowing integration of different registration algorithms as well as estimation of registration accuracy. Two algorithms were tested using simulation and real world measurements with a phantom: a rigid one based on Horn reflecting many established algorithms and a not strictly rigid one supporting dilation based on Miller. For predicting registration error, an estimator based on Fitzpatrick's “target registration error” (TRE) was used, as well as the RMS error (“root mean square”). Furthermore, marker selection methods were tested: no selection, manual selection of the fiducial with the worst deviation and an iterative selection method. In addition, a graphical representation of the estimator was developed.

Results: The Horn algorithm achieved better accuracy (0,42±0,21mm) than the one based on Miller (0,69±0,40mm). Applying the Horn algorithm, the TRE estimator achieved a much more accurate estimation (0,03±0,38mm) than RMS (1,24±0,51mm). When using the Miller algorithm, the RMS showed better results, which could be explained by the fact, that the algorithm is not strictly rigid but supports dilation, while TRE was designed for rigid transformations. Selection schemes removing inaccurate fiducials (0,59±0,29mm) resulted in increased errors as compared to registration with all fiducials (0,42±0,21mm). An iterative leave-one-out algorithm proved only in phantom measurements but not in simulation measurements to be slightly more accurate than registration with all fiducials.

Conclusions: RMS is not a trustworthy estimator of accuracy, although widely regarded and used as such in clinical praxis. Hence, other types of accuracy estimators are necessary for evaluating accuracy. Results contradict the subjective feeling that removing the fiducial with the worst deviation increases accuracy; such selections ought to be carefully implemented. Systems need to be more transparent, regarding implemented algorithms and accuracy feedback, which is necessary for the surgeon. A 3D visualisation tool representing the area where estimated registration error is beneath a certain threshold could prove to be useful.