Biomedical Engineering Reference
In-Depth Information
5.4
Results
5.4.1
Examples
In Fig. 5.8 and 5.9 we show some examples of the segmentations of some repre-
sentative aneurysms from our data base.
5.4.2
Evaluation
Two experts carried out the manual measurements twice per aneurysm with
enough delay between sessions to consider them independent. The average of
the manual measurements is considered as gold-standard and compared to the
measurements obtained by the model based approach. Analysis of variance
is used to estimate the reproducibility of the manual method. Bland-Altman
analysis [33] is used to obtain the repeatability of the manual method for each
of the two observers, the agreement between the observers, and the agreement
between the manual and the computerized method.
5.4.2.1
Analysis of Variance
In the ANalysis Of VAriance (ANOVA) model, measurements are considered
samples from a random variable that follows a linear statistical model. The asso-
ciated variance is factored out into the components of the variance according to
2
2
S
+
σ
2
O
+
σ
2
H
+
σ
2
W
(5.22)
σ
=
σ
where
σ
S
is the variance due to inter-subject differences,
σ
2
O
is the variability
2
H
is the variability introduced by the observer-
introduced by the observers,
σ
2
subject interaction, and
σ
W
is the within-subject random variation. The analysis
of variance allows to estimate the significance that each component has in their
contribution to the total variance of the measurements and the intra- and inter-
observer variation.
A first analysis of variance reported no statistically significant interaction
between the observer and the subject (
σ
H
) with
p
>
0
.
55 in all cases. The results
of the analysis of variance with the suppression of this term are presented in
Tables 5.2-5.4. Table 5.5 presents the results of the intra- and interobserver
variation.
