Skip to main content

Table 3 P values from statistical tests of different GM parameter estimates

From: Performance of single and multi-atlas based automated landmarking methods compared to expert annotations in volumetric microCT datasets of mouse mandibles

 

EDMA FORM

GPA SHAPE (one sample)

GPA SHAPE (two sample)

Centroid Size

Centroid size R2

GS v Atlas

0.010

<0.001

<0.001

<0.001

0.96

GS v Improved Atlas

0.083

0.076

0.091

<0.001

0.97

GS v MAAP

0.476

0.1399

0.157

<0.001

0.95

  1. For EDMA, we used the Form procedure of the WinEDMA (Cole, 2002), which used a permutation test with 100,000 replicates to establish the significance. For GPA we used the testmeanshapes function from R shapes package. A permutation test was used for the one sample test (assuming exchangeability between groups), whereas a bootstrap procedure was used for two-sample test. 50,000 replicates were used in both cases. Because the number of samples were low for a true multivariate test such as Hotelling T^2, we reported the Goodall F-test metric which uses the sum-of-squared Procrustes distances to measure SS (Goodall, 1991). This test is also known as Procrustes ANOVA. A paired t-test was used to compare centroid size estimates. All comparisons were run as separate statistical tests. All groups contained the identical set of samples (N = 36 per group). Adjusted R2 results are from linear regressions of centroid size from automated methods on GS centroid size