Figure 3: Analysis of the Sørensen–Dice coefficient including pair-wise significance matrices.
From: A preclinical micro-computed tomography database including 3D whole body organ segmentations
The highest variability occurs for the spleen in the native dataset due to the low soft tissue contrast of μCT images (DICE of 0.373) as well as for muscle (native: 0.528 and contrast-enhanced: 0.369), intestine (0.722 and 0.682), and stomach (0.736 and 0.686) for both datasets. All other organs, especially those with clear organ boundaries such as bladder and heart, depict a good Sørensen-Dice coefficient, nearly reaching the optimum 1.0 (=perfect overlap). Statistical significances are shown as pair-wise significance matrices (p < 0.05 in green). The matrices demonstrate that analysing the native μCT data, the highest user-dependent errors occur for spleen, muscle, and stomach. For the contrast-enhanced μCT data, the highest user-dependent errors occur in segmenting muscle, stomach, and tumor.
