Our fully automated registration method utilizes mutual information as
similarity measure [1][2]. The CT or PET image is resampled to the voxel size of
the MR image. The intensities are scaled to [0,127] for both images, then a
multiresolution pyramid is built [3] from the 2D slices. A variant of Powell's
multidimensional minimization technique [4] is used to find the transformation.
At lower resolution nearest neighbour, at the final level bilinear iterpolation
between slices is used. The registration takes 8-15 minutes on a PII-266 PC
(Linux).
1. "Automated multi-modality image registration based on
information theory", A. Collignon, F. Maes, D. Delaere, D. Vandermeulen, P.
Suetens, and G. Marchal, In Information processing in medical imaging 1995, Y.
Bizais, C. Barillot, and R. Di Paola, eds., Kluwer Academic, Dordrecht, 1995, pp
263-274.
2. "Multi-modal volume registration by maximization of mutual
information", W. M. Wells, III, P. Viola, and R. Kikinis. In Medical robotics
and computer assisted surgery 1995, Wiley-Liss, New York, 1995, pp. 155-162.
3. "The Laplacian Pyramid as a Compact Code", P. J. Burt, E. H. Adelson
IEEE Trans. on Comm., 31:532-540, 1983
4. "Numerical Recipes in C: The
Art of Scientific Computing", W. H. Press, B. P. Flannery, S. A. Teukolsky, W.
T. Vetterling, Cambridge University Press, Cambridge, England, second edition,
1992
Results
| Modality
| Surface-based
(6 methods)
error mean (std.dev.)
| Volume-based
(8 methods)
error mean (std.dev.)
| Our method
error mean
|
|---|
| CT-T1 | 5.7 (7.8)
| 2.9 (2.4)
| 1.63
|
| CT-PD | 5.8 (8.0)
| 2.9 (2.5)
| 2.25
|
| CT-T2 | 6.3 (7.9)
| 2.4 (1.4)
| 2.02
|
| CT-T1 rect. | 6.1 (8.3)
| 2.0 (2.5)
| 1.66
|
| CT-T2 rect. | 5.7 (7.8)
| 1.8 (2.0)
| 1.45
|
| CT-PD rect. | 6.1 (7.6)
| 2.1 (1.6)
| 1.72
|
| PET-T1 | 3.9 (2.0)
| 3.5 (2.1)
| 5.31
|
| PET-T2 | 4.4 (2.1)
| 3.6 (1.9)
| 3.82
|
| PET-PD | 4.3 (2.6)
| 4.0 (2.7)
| 4.39
|
| PET-T1 rect. | 3.9 (2.3)
| 2.7 (1.4)
| 3.84
|
| PET-T2 rect. | 3.9 (2.0)
| 3.5 (1.7)
| 3.94
|
| PET-PD rect. | 3.9 (2.3)
| 3.5 (2.4)
| 4.83
|
Results are from West, et al., Retrospective Intermodality Registration Techniques for
Images of the Head: Surface-Based Versus Volume-Based,
IEEE Trans. on Medical Imaging. Vol 18. No. 2., February 1999
and from the web page
http://www.vuse.vanderbilt.edu/~jayw/results.html .
The images and the standard transformation(s) were provided as
part of the project, ``Evaluation of Retrospective Image
Registration'', National Institutes of Health, Project Number
1 R01 NS33926-01, Principal Investigator, J. Michael Fitzpatrick,
Vanderbilt University, Nashville, TN.
Back