Biomedical Engineering Reference
In-Depth Information
Table 2.
AOM for 20 T1-weighted Normal Brains
WM GM
1-24
0.75
0.71
100-23
0.79
0.75
11-3
0.76
0.71
110-3
0.73
0.73
111-2
0.83
0.78
112-2
0.79
0.80
12-3
0.84
0.75
13-3
0.84
0.77
15-3
0.70
0.60
16-3
0.70
0.51
17-3
0.80
0.72
191-3
0.84
0.70
2-4
0.79
0.78
202-3
0.80
0.69
205-3
0.80
0.71
4-8
0.76
0.71
5-8
0.68
0.55
6-10
0.76
0.71
7-8
0.79
0.78
8-4
0.77
0.73
results with the ones obtained by other investigators. The results of the comparison
are shown in Figures 18 and 19.
We can observe from Figures 18 and 19 that:
1. The overall AOM for WM segmentation is 0.78, which is well above those
with six other algorithms, ranging from 0.47 to 0.56. The overall AOM for
GM segmentation is 0.71, which is also higher than the six listed algorithms
within the range of 0.55-0.58.
2. Our algorithm is persistently robust compared with the other methods.
Take WM segmentation, for example. The AOM of our method ranges from
0.68 to 0.84, with a fluctuation of 0.16. Comparatively, the corresponding
fluctuation with the six other methods is about 0.6. This shows that our
approach has equally excellent performance, even in case of noisy images.
3. The performance for WM segmentation is relatively better than that for
GM segmentation. One reason for this is that the GM is much more
convoluted than the WM, and the surfaces have difficulty moving into
extremely narrow sulci (7 1-2 pixels wide). Another reason is that the GM
