Information Technology Reference
In-Depth Information
4.2 Experimental Results
In both studies, the present technique could successfully reconstruct 3D surface
models of all 3 test lumbar vertebrae. On average it took the present technique
about 105 s to
finish the computation. The errors of reconstructing surface models
of all 3 lumbar vertebrae in both studies are shown in Table
1
. A more detailed box-
plot description of the reconstruction errors in both studies is shown in Fig.
8
. When
the surface-based anisotropically-scaled rigid registration was used to recover the
unknown scale factors, an average mean reconstruction error of 0.77 mm (range:
from 0.7 to 0.9 mm) was found for the leave-all-in study and an average mean
reconstruction error of 0.83 mm (range: from 0.8 to 0.9 mm) was found for the
leave-four-out study. In contrast, when the surface-based isotropically-scaled rigid
registration was used, the average mean reconstruction error of the leave-all-in
study was changed to 1.03 mm (range: 0.9
1.3 mm) and the average mean
reconstruction error of the leave-four-out study was changed to 1.17 mm (range:
1.0
-
1.4 mm).
Figure
9
shows the surface model reconstruction accuracies of all three vertebrae
in the leave-four-out study, where the ground truth models with a color-coded error
distribution (middle column) are displayed together with the reconstructed surface
models (right column) after surface-based anisotropically-scaled rigid registrations
were used to recover the associated unknown scale factors of all three vertebrae.
-
5 Discussions and Conclusions
In this chapter, we presented a single image based 2D/3D reconstruction technique
and showed its application to reconstruct a scaled, patient-speci
c 3D surface
model of the lumbar vertebra from a single lateral
fl
fluoroscopic image. This single
Table 1 Errors of reconstructing the surface models of the 3 lumbar vertebrae
Vertebra Cadaver_1_L1 Cadaver_1_L2 Cadaver_1_L3
Leave-all-in study, when the anisotropically-scaled rigid registration was used to recover
the scale
Errors (mm)
0.6
Leave-one-out study, when the anisotropically-scaled rigid registration was used to
recover the scale
Errors (mm) 0.9
±
0.8 0.8
±
0.8 0.8
±
0.7
Leave-all-in study, when the isotropically-scaled rigid registration was used to recover the
scale
Errors (mm)
0.9
±
0.7
0.7
±
0.7
0.7
±
0.7
Leave-one-out study, when the isotropically-scaled rigid registration was used to recover
the scale
Errors (mm)
1.3
±
1.0
0.9
±
0.8
0.9
±
1.4
±
1.1
1.0
±
0.9
1.1
±
0.8
