Biomedical Engineering Reference
In-Depth Information
Figure 10. C i computed for values of σ from 4 pixels to 20 pixels for the image in
Figure 7a: w int =0 . 25 and w img =0 . 75 . Reprinted with permission from the AAPM.
To evaluate the algorithm, we created a “gold standard” outline with which to
compare algorithm-segmented outlines. The five trained operators who assisted
in the construction of the virtual operators each manually outlined all 15 TRUS
prostate images in one session. Each session was repeated 10 times, so there was
a total of 50 manual outlines for each image. Again, the sessions were separated
by 3 days each, and the images were presented in random order at each session in
order to minimize the effects of memory. All 50 manual outlines were averaged
together to form the “gold standard” prostate boundary outline.
We applied our DDC-based segmentation algorithm to all 15 images 100 times
each for each value of σ =4 , 6 , 8 ,..., 20. For each value of σ , virtual operator
VO i was used to initialize our algorithmon image i . The 100 algorithm-segmented
outlines for a specific image and a specific value of σ were then averaged, and the
average algorithmoutline was compared to the “gold standard” outline correspond-
ing to that image. For purposes of illustrating the sensitivity of algorithm-generated
outlines to variations in the choice of the parameter values, we computed the area-
based accuracy metric, C a , defined by Eq. (8c) for each image and for each value
of σ .
Figure 10 shows C a plotted against σ for the TRUS prostate image shown
in Figure 7a. In this case, C a is greater than 91% for all values of σ , and has a
peak for σ between 8 and 10. For several of the other images, C a peaked at about
the same σ values. Yet, in others, C a generally decreased with increasing σ ,but
in these cases, C a was generally higher than 90%. We have utilized plots of C a
versus σ , and w int versus σ , to optimize the performance of our algorithm and
reported the results in [38].
Search WWH ::




Custom Search