Docteur de l'université Automatic Segmentation and Shape Analysis ...

128 Chapter 5 Quantitative shape analysis of hippocampus in AD
Specificity
Specificity
0.805 0.815 0.825 0.835
●
●
●
●
●
● ●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
● ● ● ●
●
●
●
●
●
● ●
●
●
● ●
●
●
●
●
● ● ●
●
●
● ● ●
●
●
●
● ●
● ● ●
● ● ●
●
●
1e−04 1e−03 1e−02 1e−01 1e+00
Threshold α
(a) OOB specificity, LR, MR
0.74 0.78 0.82 0.86
●
●
●
●
●
●
●
● ● ● ●
●
● ●
●
●
●
●
●
●
●
●
●
●●
●
● ● ●●
● ●
●
●
●
●
●
●
● ●
●
● ●
●
●
● ● ●
●
● ●●
●
●
● ● ● ●
● ●● ●
●
● ●●
●
1e−04 1e−03 1e−02 1e−01 1e+00
Threshold α
(c) OOB specificity, LS, MR
Specificity
Specificity
0.70 0.72 0.74 0.76 0.78
●
●
●
●
●
●
● ●
●
●
● ●
●
●
●
●● ●
● ●
●
●
●
●
● ● ● ●
● ●
●
●
●
● ●
● ●●
●
●
●●
●
●●
●
●
●
●
● ●
●
●●
●
●
●
●
●
● ●●●
●
●
●
●
●
●
1e−04 1e−03 1e−02 1e−01 1e+00
Threshold α
(b) OOB specificity, LR, MS
0.72 0.76 0.80 0.84
●
●
●
● ●
● ●
●
● ●
●●●
●
●●
●
●
●
●
● ●●
●
●
● ●
●
●
●
● ●
●●
●
●
●
●●
●●
●
●●●
●●
● ●
●●
● ● ●
●●
● ●
●●
●
● ●
●
●●
●
●
●
●
1e−04 1e−03 1e−02 1e−01 1e+00
Threshold α
(d) OOB specificity, LS, MS
Figure 5.9: The out-of-bag (OOB) estimates of specificity of the disease classification
using bagged support vector machines (SVMs) with varying thresholds
in the landmark selection. Black circles: shape features from the Statistical
Shape Model (SSM) only; red circles: shape features with additional volume
features.
of education, in order to isolate the disease from these effects.
5.6.3 Disease classification using SSM
Using the SSM descriptors combining both size and shape information provided
better discrimination than using only hippocampal volume to classify AD from
NC. In general the descriptors of size-and-shape from MR outperformed features
produced by MS, because volume alone is a good discriminant. Since the changes
modeled by MS were driven by both size and shape, adding volume to the features
extracted using MR and MS increased both of their accuracy, but to a much
less extent for MR compared to MS. Using shape information therefore provides
additional discrimination power to volumetry.