Biomedical Engineering Reference
In-Depth Information
Figure 3.7: The effect of image cropping on registration accuracy. The light
and dark bars are RMS distances between bony landmarks with and without
image cropping, respectively, as defined in Methods. Conditions on the
x
-axis
are described in Methods.
Displacements of bony landmarks might significantly overestimate the
change near the prostate. Hence, as described in section 3.3, we investigated
the displacement of voxels in a ROI surrounding the prostate between registra-
tions with and without cropping. For 9 of 10 analyzed volume pairs, the average
voxel displacement was
<
0
.
5 mm indicating that prostate registration is fairly in-
sensitive to cropping. However, for V3 treatment-reference, a much larger voxel
displacement of 7.4 mm was obtained indicating that cropping is critical for this
volume pair.
3.2.4.6
Implementation Issues
The algorithm was quite robust and always gave very nearly the same transfor-
mation parameters (less than 0.01 voxels and 0.01
◦
) for the 22 volume pairs in
this study using a wide variety of initial guesses. The restarting and multiresolu-
tion features are important, and we report some results for a typical volume pair
