Objective Evaluation of Three-dimensional Image Registration Algorithms – Tools for Optimization and Evaluation

F. Uhlemann; U. Morgenstern; R. Steinmeier

doi:10.1055/s-0038-1633879

Methods of Information in Medicine, Table of Contents

Methods Inf Med 2004; 43(04): 367-370
DOI: 10.1055/s-0038-1633879

Original Article

Schattauer GmbH

Objective Evaluation of Three-dimensional Image Registration Algorithms – Tools for Optimization and Evaluation

Authors

F. Uhlemann

¹Image Processing and Recognition Group, Institute of Artificial Intelligence, Department of Computer Science, Dresden University of Technology, Dresden, Germany
U. Morgenstern

²Institute of Biomedical Engineering, Department of Electrical Engineering and Information Technology, Dresden University of Technology, Dresden, Germany
R. Steinmeier

³Department of Neurosurgery, University Hospital “Carl Gustav Carus”, Dresden University of Technology, Dresden, Germany

Abstract

Summary

Objective: The registration of medical volume data sets plays an important role when different images or modalities are used during computer-assisted surgical procedures. Nevertheless, it is often questionable how robust and accurate the underlying algorithms really are. Therefore, the goal is to foster the establishment of methods for an objective evaluation.

Method: To reliably calculate the accuracy of registration algorithms, a reference transformation must be known. Due to the unknown perfect registration for real clinical data, the simulation of realistic data and successive affine transformations are employed. The simulation is based on models of the respective imaging modality where the dominant physical effects are taken into account. This gives the user full control over all simulation and transformation parameters. Finally, suitable quality measures are applied which allow a systematic evaluation of image registration accuracy by comparing the known theoretical result and the transformation calculated by the algorithm under investigation.

Results: During the development of a new registration algorithm, the presented method proved to be a very valuable tool for optimization and evaluation of registration accuracy, since it allows objective numerical comparison of the calculated results.

Conclusions: The presented method can be used during the development of algorithms for optimization and for quantitative comparison of different registration schemes. The respective software tool can automatically generate and transform simulated but realistic data. Employing suitable numerical quality measures, an objective evaluation of registration results can be easily obtained. Still, the validity of the relatively simple models has to be verified to draw reliable conclusions with respect to real data.

Keywords

Quantitative evaluation - computer-assisted image processing - image registration - reference standards - software validation

Full Text

References

References
1 Sobottka S, Uhlemann F, Beuthin-Baumann B. et al. Genauigkeit der digitalen Bildfusion von PET und MRT/CT eines automatischen Matching-Algorithmus. 1. Jahrestagung der Sektion Neurochirurgie und -navigation der DGNC; 2001, Oct 5-6; Tübingen. 2001: 60
2 West J, Fitzpatrick JM, Wang MY. et al. Comparison and Evaluation of Retrospective Intermodality Brain Image Registration Techniques. SPIE Medim ‘96 conference; 1996, Febr 10-15; Newport Beach; SPIE. 2017: 332-47.
3 Lehmann TM. From plastic to gold: A unified classification scheme for reference standards in medical image processing. SPIE 2001; 4684: 212-31.
4 Uhlemann F. Segmentierung, Volumenbestimmung und Visualisierung medizinischer Daten verschiedener Modalitäten [diploma thesis]. TU Dresden. 2000
5 Redondo G. Investigation and Simulation of characteristic Effects of three-dimensional Ultrasound Imaging [diploma thesis]. TU Dresden. 2003
6 Chodkowski B. A Simulation of Clinically Realistic PET Data and its Use in the Evaluation of Image Registration Algorithms [master thesis]. George Washington University. 1996
7 Holm S. Ultrasim – a toolbox for ultrasound field simulation. Nordic MATLAB conference; Oct. 2001
8 Kwan RKS, Evans AC, Pike GB. An Extensible MRI Simulator for Post-Processing Evaluation. VBC’96, Lecture Notes in Computer Science, Springer 1996; 1131: 135-40.
9 Lehman TM, Hemler PF, Webber RL. Virtual radiographs computed from TACT volume data as a gold standard for image registration prior to subtraction. Dentomaxilliofacial Radiology 2002; 31: 187-92.
10 Steinbrecher R. Bildverarbeitung in der Praxis. München, Wien, Oldenbourg: Oldenbourg Verlag GmbH; 1993
11 Bergen, et al. Hierarchical model-based motion estimation. Proceedings of Second European Conference on Computer Vision, Springer. 1992: 237-52.
12 Viola P. Alignment by Maximization of Mutual Information [PhD thesis]. Massachusetts Institute of Technology. 1995