Biomedical Engineering Reference
In-Depth Information
15
.
5
±
0
.
7 mm, where the latter number is the standard deviation. The maximum
distance was 53.2 mm. After VOI optimization, the algorithm moved the floating
CPs an average of 9
.
0
±
6
.
5 mm. This value shows that one does not have to be
very careful in marking corresponding CPs.
3.3.5
Discussions
3.3.5.1
Applicability of Non-Rigid Registration
For MR images of the pelvis and prostate, non-rigid registration is desirable
whenever images are acquired in different positions or with different conditions
of bladder and rectal filling. Local deformations throughout the pelvis can be
corrected, and, more importantly, the prostate can be accurately registered.
However, when images are acquired in the same position under comparable
conditions such as our case called diagnosis-diagnosis, rigid body registration
worked satisfactorily as previously reported by us [1]. Similarly, if one were
to reproduce the treatment position with reasonable accuracy, we believe that
prostate registration would be very good.
Our goal is to get good matching throughout the entire pelvic region not
just at the prostate because proper localization of other organs is important
for interpretation of some functional images and because anatomical spatial in-
tegrity is important for treatment planning. Hence, we used high-resolution MR
images that provide a very stringent test for warping. Many anatomical details
are evident, and even a small mismatch can clearly be seen. As a result, we
found that
≈
180 control points were required to get excellent quality registra-
tion. When we applied the method to register CT images with PET images of
the lung having much less resolution, many fewer points (
≈
50) were required
[50]. With a sufficient number of control points, the TPS transformation excel-
lently approximated the deformations of the pelvis and internal structures of
our MR images. Even when we warped the volume in the diagnostic position to
one in the treatment position, most organs were closely aligned, despite very
significant movements. The method performed equally well for correcting the
deformation and organ displacement arising from changes in bladder and rectal
filling.
With our graphical user interface, interactive control point selection is quite
easy after training. It usually took an experienced user about 15 minutes to
