Biomedical Engineering Reference
In-Depth Information
excellent registration (not shown). We know that 3D alignment is achieved be-
cause all slices across the entire pelvis are well aligned and because rendered
images show that the prostate matches well.
Registration consistency, as described in Section 3.2.3.2, provides yet another
means of evaluating the quality of registration. Values were 0.4, 0.8, and 0.7
mm for volunteers V1, V2, and V3, respectively. The average is 0
.
6
±
0
.
2 mm, a
value less than half the dimension of a voxel indicating excellent registration
consistency.
3.2.4.4
Assessment of Prostate Registration
We determined the quality of prostate registration by visually examining nearly
all of the roughly 800 registered image slices using one or more of the methods
found in RegViz. A typical example for the case of diagnosis-reference is shown
in Fig. 3.6 where the boundary overlap is excellent and probably within the
manual segmentation error. In some other cases such as treatment-reference,
small displacements of the prostate were observed. In a typical volume pair, the
prostate is displaced to the posterior direction by
≈
3
.
0 mm when the legs are
raised. There are no obvious displacements in other directions.
Centroid vector displacements can also be analyzed following registrations.
For the case of diagnosis-reference, centroid displacements are only 1
.
4
±
0
.
2
mm. In the case of treatment-reference, there is a consistent displacement (
≈
3
mm) in the posterior direction with relatively little change in the two orthogonal
directions. In the case of empty bladder-reference, two of the three volume pairs
show a displacement in the posterior direction while the other is displaced in the
anterior direction. Finally, in the case of a diagnostic volume obtained one week
before the reference, there wasa4mmdisplacement in the caudal direction
due mostly to changes in rectal and bladder filling. Because the 3D centroid of
the prostate averages over a large region, we believe these measurements to be
relatively insensitive to segmentation error. Even so, we consider the uncertainty
to be at least 1 mm, and displacements less than this should be disregarded. All
significant results above can be visually confirmed.
Prostate volumes were measured for each subject. The typical difference
between volumes in an imaging session was
<
1
.
5%, indicating that segmenta-
tion errors were small and that prostate volumes did not change. The average
prostate volume for the healthy volunteers was 23
.
9
±
3
.
2cm
3
. Volume mea-
surements are particularly useful for clinicians when assessing the response
