Biomedical Engineering Reference
In-Depth Information
similar gray values. In total, there are 4, 4, and 8 volumes for volunteer S1, S2, and
S3, respectively. The permutation of the volumes gives many possible volume
pairs for registration experiments.
3.3.4.3
Volumes for Registration Experiments
We registered 17 volume pairs under five different conditions as defined above.
Five pairs are
treatment-diagnosis;
seven pairs are
full bladder-empty blad-
der;
two pairs are
diagnosis 1 week-diagnosis;
and three pairs are
diagnosis-
diagnosis
. For each case, other conditions were controlled. For example, for
the case of diagnosis 1 week-diagnosis, both volumes were acquired with empty
bladder and comparable conditions. Rigid body and warping registration were
applied to each of the volume pairs. Results were evaluated as described
next.
3.3.4.4
Effect of Control Point Selection
on Registration Quality
In well over 100 registration experiments using different numbers and place-
ment of CPs, we investigated effects on non-rigid registration quality. For each
of the three volunteers, we selected one typical volume pair from the diagnostic-
treatment positions for systematic experiments. We progressively increased the
number of CPs from 15 to 250. We found that less than 120 CPs did not produce
good visual matching of our high-resolution MR images showing great anatom-
ical detail. More than 220 CPs did not give significantly improved results but
required more time for manual selection and optimization. When we used
≈
180
CPs placed strategically using rules described later, we obtained excellent re-
sults over the entire pelvis and internal organs. As a result of our experience,
we modified the registration method to be suitable for many CPs.
Some rules follow for strategic placement of CPs. For registration of treat-
ment and diagnostic image volumes, most CPs were selected using trans-
verse slices because they best showed the pelvic displacement when mov-
ing the legs to the treatment position (Fig. 3.9). About 25 CP pairs were
placed near the edge and point features having recognizable correspondence
on each of 5-8 transverse slices with a
z
interval of
≈
8 mm, covering the en-
tire pelvic region. Additionally, we placed about 25 CPs from sagittal slices
