Fig 1.
Landmark detection using multi planar reconstruction (MPR) with axial view as the centre of orientation.
Plotted landmarks: 1. Glabella, 2. Soft Nasion, 3. Hard Nasion, 4. Pronasale 5. Subnasale 6. Anterior nasal spine, 7. Sella 8 & 9. Alare 10 & 11. Orbitale 12 & 13 Porion 14 & 15 Zygion.
Fig 2.
(I) Absolute error: AVEDEV and ODEV averaged for each landmark across all 20 individuals; soft-tissue landmarks are emphasized using a grey background: a) mean AVEDEV shown with a solid black line with its 2.5th-97.5th percentiles shown using broken grey lines; b) mean ODEV: solid line, operator 1; broken line, operator 2; dotted line, operator 3.
Table 1.
Reproducibility (analysis II) of size and shape in the 20 women sample with three replicas: Procrustes ANOVA comparing individual variation, in centroid size and shape, to measurement error.
Fig 3.
(II) Reproducibility of centroid size visualized using jitter plots for the three sets of landmarks (nose, bone and all landmarks) using estimates from the three operators.
Fig 4.
(II) Reproducibility of size: Scatterplot of nasal size used as an example of the graphical exploration of similarities across different operators: Operators 1 and 2 are shown respectively on the horizontal and vertical axes, while the size of the circles is proportional to size estimated from operator 3.
Table 2.
Reproducibility (analysis II) of size and shape in the replica sample: Between operators pairwise correlations of centroid size (Pearson correlation) and shape (correlation of shape Procrustes distance matrices).
Fig 5.
(II) Reproducibility of shape: UPGMA phenogram based on Procrustes shape distances for the 20 women (each indentified by a progressive number from 1 to 20) digitized by all three operators: With high reproducibility, all three replicas, or at least two of them, should cluster together 'within' individual (black numbers); when this does not happen, numbers are shown using light grey.
As in Fig 4, nasal data are used as an example.
Table 3.
Inter-operator bias (analyses II-III): R2 of centroid size and shape estimated in ANOVAs using operator as a grouping factor in the replica sample and the study sample.
For the study sample, R2 for sex, as main factor above operator, is also shown.
Fig 6.
(III) Inter-operator bias: Scatterplots of the first two PCs (principal components) of total shape (all 15 landmarks) accounting for respectively 15.5% and 11.0% of total variance; sex (a) and operator (b) are shown using different symbols. Despite PCs being computed regardless of a priori groups, operators (a meaningless grouping factor in the absence of bias) show less overlap than sexes (i.e., biological groups).