Table I.
Segmentation accuracy of the different models in comparison with automatic segmentation and inter-observer variability in manual segmentation. First observer’s manual segmentation labels were used as the reference labels.
| Model | # Boundary Points | Validation | Test | Exec. Time (s.) | ||||
|---|---|---|---|---|---|---|---|---|
| DSC (%) | SR (%) | PR (%) | DSC (%) | SR (%) | PR (%) | |||
| 2nd Observer | - | 83.0 ± 5.8 | 75.5 ± 7.1 | 92.8 ± 8.9 | 87.0 ± 7.4 | 80.0 ± 11.2 | 96.4 ± 2.5 | - |
| Automatic | - | 83.5 ± 7.0 | 88.8 ± 5.3 | 80.0 ± 12.8 | 84.4 ± 10.8 | 91.7 ± 7.6 | 80.9 ± 17.1 | 0.8 |
| Model I (U-Net) | 5 | 83.0 ± 7.3 | 87.6 ± 5.4 | 80.4 ± 12.6 | 86.2 ± 8.9 | 92.2 ± 5.5 | 82.9 ± 14.3 | 0.8 |
| 10 | 83.2 ± 7.5 | 90.0 ± 5.1 | 78.8 ± 12.4 | 85.3 ± 10.8 | 93.4 ± 5.2 | 80.9 ± 16.7 | ||
| 15 | 83.3 ± 7.2 | 90.4 ± 4.8 | 78.4 ± 12.0 | 86.5 ± 8.6 | 92.5 ± 4.8 | 82.7 ± 13.7 | ||
| 20 | 85.7 ± 5.8 | 90.8 ± 4.5 | 82.1 ± 10.2 | 86.6 ± 8.8 | 91.9 ± 4.6 | 83.2 ± 13.5 | ||
| 30 | 86.8 ± 3.6 | 93.9 ± 2.7 | 81.2 ± 6.9 | 89.9 ± 2.9 | 92.8 ± 4.2 | 87.5 ± 5.1 | ||
| Model II (DIP U-Net) | 5 | 83.9 ± 6.0 | 88.4 ± 6.9 | 81.0 ± 10.0 | 87.5 ± 8.1 | 92.6 ± 5.0 | 84.4 ± 13.3 | 0.9 |
| 10 | 84.7 ± 6.1 | 90.3 ± 4.9 | 80.7 ± 10.5 | 85.3 ± 10.1 | 92.7 ± 4.9 | 81.0 ± 15.5 | ||
| 15 | 84.5 ± 5.7 | 89.9 ± 3.7 | 80.6 ± 10.3 | 88.1 ± 6.2 | 92.0 ± 5.3 | 85.4 ± 10.0 | ||
| 20 | 84.2 ± 5.3 | 93.1 ± 3.8 | 77.5 ± 9.4 | 88.8 ± 4.6 | 93.7 ± 3.8 | 85.0 ± 8.3 | ||
| 30 | 85.4 ± 5.0 | 92.1 ± 3.5 | 80.2 ± 9.1 | 88.8 ± 6.1 | 93.2 ± 4.4 | 85.8 ± 10.6 | ||