Information Technology Reference
In-Depth Information
Fig. 5.14 The efficiency of the proposed slice-by-slice correction algorithm demonstrated by
cross-sectional view image. a 3D volume data and b Cross-sectional view image randomly taken
from the 3D volume data, illustrating the resulting segmentations with and without the correction
step are shown in blue and black, respectively. The red contour represents the reference
boundaries of the vessel obtained through manual delineation
Table 5.4 Comparison of
the segmentation results
before and after the
application of the correction
scheme
Rate
Before correction
After correction
TP (%)
82.4
85.3
FN (%)
17.6
14.7
FP (%)
294
22.8
OM
0.380
0.820
method. Nevertheless, we can conclude that our technique is able to delineate the
vessel boundaries in clinically important coronary segments with a level of vari-
ability similar to those obtained through manual segmentation.
Two CTA studies were affected by the presence of kissing vessels (i.e. datasets
#7 and #11). The slice-by-slice correction scheme was subsequently applied after
the first step of the segmentation. Figure 5.12 illustrates the 3D surface recon-
struction image before and after applying the slice-by-slice correction algorithm.
Figure 5.13 a shows the first slice, which contains the cross-sectional segments of
the coronary arteries. The false positives, i.e. the kissing vessel artefacts, were
identified and then tracked in Fig. 5.13 b-e. The boundaries of the coronary arteries
are shown in red, and the green curve represents the kissing vessel structures. In
Fig. 5.13 f, we illustrate the initial segmentation of the axial image shown in
Fig. 5.13 e prior to the correction.
Instead of using the entire volume to quantify the performance of the slice-by-
slice correction, we conduct the assessment on a 2D axial image basis. The
resulting segmentation with and without the slice-by-slice correction is illustrated
in Fig. 5.14 , where the boundaries of the artery before and after correction are
depicted in black and blue, respectively, and the manual segmentation is shown in
red. The four metrics, defined in Eq. 5.10 , were used to quantify the performance
and the results are shown in Table 5.4 . It can be observed from the table that the
FP rate, which is primarily caused by the kissing vessels, is dramatically reduced
after the application of the slice-by-slice correction algorithm (reduced from 294
to 22.8 % for arterial segments containing kissing vessels, e.g. the vessel segment
shown in Fig. 5.12 a, within the circle).
Search WWH ::




Custom Search