Table 2 Quantification of the U-NET efficiency of the two-step segmentation process for 220 images, mean ± s.d
Step 1. Segmentation of the whole skin layer regions (epidermal/SVP layer, dermal layer) | |||||||||
|---|---|---|---|---|---|---|---|---|---|
Segmented region | U-NET vs Operator A | U-NET vs Operator B | Operator A vs Operator B | ||||||
Dice score | Cohen’s Kappa coefficient | Hausdorff distance | Dice score | Cohen’s Kappa coefficient | Hausdorff distance | Dice score | Cohen’s Kappa coefficient | Hausdorff distance | |
Epidermal/SVP layer | 0.91 ± 0.01 | 0.89 ± 0.01 | 29.61 ± 2.18 | 0.87 ± 0.02 | 0.85 ± 0.02 | 34.54 ± 3.27 | 0.90 ± 0.02 | 0.89 ± 0.02 | 21.00 ± 2.99 |
Dermal layer | 0.89 ± 0.01 | 0.83 ± 0.01 | 65.80 ± 3.29 | 0.86 ± 0.01 | 0.79 ± 0.01 | 73.11 ± 4.04 | 0.91 ± 0.01 | 0.87 ± 0.01 | 40.52 ± 3.06 |
Step 2. Segmentation of the microvasculature within the skin layers (epidermal/SVP layer, dermal layer) | |||||||||
Segmented microvasculature | U-NET vs Adaptive thresholding/tubeness filter | ||||||||
Dice score | Cohen’s Kappa coefficient | Hausdorff distance | |||||||
Epidermal/SVP layer | 0.90 ± 0.01 | 0.89 ± 0.01 | 23.35 ± 0.91 | ||||||
Dermal layer | 0.87 ± 0.01 | 0.85 ± 0.01 | 39.00 ± 1.86 | ||||||
