Biomedical Engineering Reference
In-Depth Information
13.2.2.2
Results
All 41 CT-MR image pairs were used in the experiments. All CT images were
registered to the MR images using the MR as the reference image. A typical su-
perposition, before and after registration, of CT-MR images is shown in Fig. 13.4.
Observed from the registration results using the binarization approach, the
translation parameter ranged from
−
26.42 to 8.37 mm and the angle parameter
ranged from
−
4
.
3
◦
to 1
.
59
◦
. The running time for each CT-MR pair took 10.4 to
27.8 minutes. The average time for all 41 CT-MR pairs registration is 18.4 minutes.
From our results, using the nonlinear binning method, the running time for
each CT-MR pair took 7.6 to 18.3 minutes. The average time for all 41 CT-MR pairs
registration was 12.2 minutes. When we compare the registration time using the
nonlinear binning with the registration time using the binarization approach, we
observe the average speedup when using nonlinear method is 51%. Figure 13.5
shows the timing results.
The running time for the approach using the linear binning technique is
between the above two methods.
For each experiment the registration solution was obtained using fiducial
markers as provided by
The Retrospective Registration Evaluation Project
.
This was used as the gold standard to evaluate registration accuracy.
Figure 13.4:
A typical superposition of CT-MR images. The left images are be-
fore registration and the right ones are after.
