Biomedical Engineering Reference
In-Depth Information
registration was applied. Interactively placed control points were automatically
optimized by maximizing the mutual information of corresponding voxels in
small volumes of interest and by using a three-dimensional thin plate spline to
express the deformation throughout the image volume. More than 100 registra-
tion experiments with 17 MR volume pairs determined the quality of registration
under conditions simulating potential interventional MRI-guided treatments of
prostate cancer. Evaluations included visual inspection; voxel gray value mea-
sures such as mutual information, correlation coefficient, and intensity differ-
ence; and displacement of the centroids of segmented prostates. For image
pairs that stress rigid body registration (e.g., supine, the diagnostic position,
versus legs raised, the treatment position), both visual and numerical evalua-
tion methods showed that warping consistently worked better than rigid body.
Non-rigid registration rectified the misalignment in the pelvis following rigid
body registration. The prostate centroid displacement for a typical volume pair
was reduced from 3.4 mm to 0.6 mm when warping was added. Experiments
showed that
≈
180 strategically placed control points were sufficiently expres-
sive to capture important features of the deformation. When only 120 control
points were used, warping throughout the pelvis was visually less satisfactory
but the prostate was aligned reasonably well. For volume pairs with images ac-
quired in the same position (diagnosis-diagnosis) and comparable conditions,
the rigid body method worked sufficiently well, and the prostate centroid dis-
placements were
<
1
.
0 mm. In conclusion, the non-rigid registration method
works better than rigid body registration whenever patient position or condi-
tion is greatly changed between acquisitions. It is very computational efficient
for hundreds of control points and can very well approximate the deformation
of the pelvis and internal organs. It will probably be a useful tool for many
applications.
Questions
1.
Describe the concept of image registration.
2.
Describe the steps for voxel-based image registration.
3.
Given two images A and B, describe how to compute their mutual infor-
mation value.
