Description of Registration Method for Asad
We optimize a convex function combining the Mutual Information and
the average inter-feature Euclidean distance. The weights in the The
convex function are determined automatically by minimizing the
average inter-feature Euclidean distance and finding the ratio of the
MI change with respect to the average Euclidean distance change.
The geometrical features correspondences used are point-to-point.
In the model (floating) data set, we perfrom 3D edgel extraction and
linking followed by triangulation to construct a surface-based model.
In the object (reference) data set, we extract object-control-points (OCPs)
in the vicinity of a transformed subset of the model points. The average
Euclidean distance minimized is the average distance between the extracted
OCPs and the corresponding closest points on the surface-based
model.
The optimization of the convex function is done using Powell's search
algorithm. The model construction, OCPs extraction, and registration
are fully automatic.
As a preprocessing step, we add new slices to make the data sets
isotropic and then sub-sample them to reduce the number of voxels
to be processed (e.g., an MRI-T1 data set with 256x256x26 voxels and
voxel sizes of 1.25mmx1.25mmx4.0mm is converted into a 128x128x42
data set with voxel sizes of 2.5mmx2.5mmx2.476mm).
References:
1) Asad Abu-Tarif and George Nagy, "Multi-Modal 3D Registration: A
Comparative Study."
submitted for publication in ICPR2002
2) W. Wells III, P. Viola, H. Atsumi, S. Nakajima, and R. Kikinis,
"Multi-Modal volume registration by
Maximization of Mutual Information." Medical Image Analysis, Vol. 1, No. 1,
1996.
3) F. Maes, A. Collignon, D. Vandermeulen, G. Marchal, and P. Suetens,
"Multi-Modality Image Registration
By Maximization Of Mutual Information." IEEE Proceedings of the Workshop on
Mathematical Methods in
Biomedical Image Analysis, June 1996.
Return to previous page