Table 1a:
Segmentation results for each structure averaged over the 8 test patients. We provide the standard deviation and range of the results. The best performance for each metric is in bold.
| Model | Metric | GTV | G. maximus | G. medius | G. minimus | Paraspinal | Piriformis |
|---|---|---|---|---|---|---|---|
| 3D U-Net Dice | VDSC | 82.8 ± 5.7 (73.6,90.1) | 93.9 ± 1.5 (91.3,95.8) | 92.3 ± 0.8 (91.2,93.4) | 84.3 ± 2.5 (79.5,88.8) | 90.6 ± 1.1 (88.4,91.9) | 74.7 ± 7.3 (62.6,84.5) |
| 3D U-Net Cross-entropy | 82.7 ± 6.6 (70.0,89.6) | 93.6 ± 2.0 (89.4,95.8) | 91.6 ± 1.0 (90.3,93.0) | 84.1 ± 2.3 (80.5,87.4) | 91.8 ± 0.8 (90.5,92.9) | 73.6 ± 6.4 (58.0,79.4) | |
| 3D U-Net Weighted cross-entropy | 79.3 ± 11.2 (52.2,88.8) | 93.5 ± 1.7 (90.7,95.4) | 91.7 ± 1.0 (89.9,93.2) | 82.9 ± 2.1 (79.1,86.2) | 91.2 ± 3.1 (83.6,94.3) | 68.9 ± 7.5 (52.2,77.5) | |
| Residual 3D U-Net Dice | 81.8 ± 14.7 (43.8,91.0) | 94.0 ± 1.7 (90.7,96.0) | 92.1 ± 1.1 (90.3,93.4) | 83.3 ± 2.6 (79.1,86.5) | 90.7 ± 3.0 (83.1,93.0) | 77.1 ± 6.1 (68.9,84.5) | |
| Residual 3D U-Net Cross-entropy | 83.1 ± 7.1 (71.1,91.4) | 93.1 ± 2.6 (87.0,95.6) | 90.2 ± 1.6 (87.6,92.4) | 80.7 ± 5.2 (70.4,88.0) | 91.3 ± 1.8 (87.2,93.8) | 72.2 ± 9.3 (49.3,81.9) | |
| Residual 3D U-Net Weighted Cross-entropy | 83.4 ± 8.2 (64.7,90.0) | 94.0 ± 1.5 (91.0,95.5) | 91.4 ± 1.1 (89.5,92.7) | 82.8 ± 2.8 (78.6,86.6) | 91.5 ± 2.6 (85.6,93.8) | 74.5 ± 5.2 (65.9,81.5) | |
| Average Ensemble | 85.1 ± 6.5 (70.5,91.5) | 94.4 ± 1.6 (91.3,96.1) | 92.6 ± 0.9 (91.1,93.7) | 85.0 ± 2.7 (81.4,90.1) | 92.2 ± 1.3 (89.3,93.3) | 77.7 ± 5.4 (66.5,84.9) | |
| Optimal Ensemble | 85.5 ± 6.4 (70.7,91.5) | 94.4 ± 1.5 (91.6,95.9) | 92.6 ± 0.9 (91.1,93.7) | 85.0 ± 2.7 (81.7,89.9) | 92.1 ± 1.5 (88.6,93.4) | 78.3 ± 5.7 (67.4,85.3) | |
| 3D U-Net Dice | SDSC 2 mm | 61.0 ± 7.5 (49.0,70.2) | 90.4 ± 5.0 (80.8,96.6) | 91.2 ± 2.2 (86.4,93.9) | 86.6 ± 5.1 (77.8,94.3) | 88.2 ± 3.5 (82.7,93.8) | 80.5 ± 8.7 (63.0,91.9) |
| 3D U-Net Cross-entropy | 60.2 ± 8.7 (49.1,73.7) | 89.1 ± 5.6 (77.2,93.9) | 89.3 ± 2.4 (85.1,93.3) | 86.6 ± 3.8 (80.1,91.4) | 90.8 ± 4.1 (83.6,96.8) | 75.9 ± 8.6 (56.9,86.0) | |
| 3D U-Net Weighted cross-entropy | 53.4 ± 12.9 (30.6,73.7) | 88.6 ± 5.2 (79.7,96.8) | 89.3 ± 2.9 (82.3,91.8) | 84.3 ± 3.6 (78.2,89.8) | 90.3 ± 3.8 (84.3,95.8) | 71.5 ± 8.8 (52.0,82.2) | |
| Residual 3D U-Net Dice | 61.1 ± 16.9 (21.1,79.5) | 90.4 ± 5.9 (77.3,97.1) | 90.3 ± 1.9 (87.2,92.7) | 85.7 ± 4.0 (78.0,92.1) | 88.8 ± 3.8 (82.2,94.2) | 81.6 ± 7.8 (71.1,90.8) | |
| Residual 3D U-Net Cross-entropy | 62.1 ± 13.4 (35.4,77.7) | 87.8 ± 7.6 (69.9,94.5) | 84.0 ± 4.0 (78.4,91.6) | 82.3 ± 8.5 (63.2,93.5) | 90.5 ± 2.9 (85.5,93.8) | 72.2 ± 10.8 (46.8,85.5) | |
| Residual 3D U-Net Weighted Cross-entropy | 62.3 ± 10.5 (47.9,74.5) | 90.9 ± 4.7 (81.6,96.5) | 88.3 ± 3.1 (82.7,92.2) | 85.1 ± 5.1 (74.1,91.9) | 90.7 ± 4.1 (81.3,95.1) | 77.9 ± 6.9 (67.9,88.0) | |
| Average Ensemble | 65.3 ± 9.1 (54.1,80.5) | 91.9 ± 5.1 (80.9,96.8) | 91.9 ± 2.5 (86.2,94.5) | 88.2 ± 5.1 (77.5,96.1) | 92.1 ± 2.5 (86.5,95.2) | 82.5 ± 7.2 (67.5,92.5) | |
| Optimal Ensemble | 66.2 ± 9.1 (55.1,81.6) | 92.2 ± 4.8 (82.0,97.0) | 91.9 ± 2.4 (86.6,94.9) | 88.3 ± 5.1 (77.4,95.7) | 92.0 ± 2.7 (85.6,95.2) | 83.9 ± 7.5 (69.5,93.9) | |
| 3D U-Net Dice | SDSC 3 mm | 68.1 ± 7.1 (57.0,78.0) | 94.1 ± 3.8 (86.3,98.1) | 95.4 ± 1.8 (91.3,97.5) | 91.0 ± 4.0 (84.1,96.9) | 93.1 ± 3.0 (87.1,97.1) | 86.0 ± 8.0 (69.4,95.4) |
| 3D U-Net Cross-entropy | 68.0 ± 8.1 (57.9,81.0) | 93.3 ± 4.7 (82.1,96.6) | 94.4 ± 2.0 (90.5,97.5) | 91.1 ± 3.4 (85.6,95.6) | 94.8 ± 2.7 (89.8,98.5) | 82.3 ± 8.1 (63.4,90.1) | |
| 3D U-Net Weighted cross-entropy | 62.0 ± 13.6 (36.0,81.4) | 92.9 ± 4.3 (84.4,98.8) | 94.4 ± 2.3 (88.5,96.4) | 89.4 ± 2.9 (85.1,94.5) | 94.3 ± 2.9 (89.3,98.2) | 79.3 ± 9.3 (58.0,90.5) | |
| Residual 3D U-Net Dice | 68.4 ± 17.5 (25.3,86.2) | 94.2 ± 4.6 (83.4,98.5) | 94.4 ± 1.6 (91.2,96.0) | 90.1 ± 3.6 (82.9,95.8) | 92.6 ± 3.1 (87.9,96.2) | 87.1 ± 6.7 (77.8,95.3) | |
| Residual 3D U-Net Cross-entropy | 69.6 ± 12.3 (44.2,83.5) | 92.0 ± 6.7 (75.3,97.1) | 89.7 ± 3.5 (83.7,95.1) | 87.4 ± 7.4 (70.3,96.4) | 94.5 ± 1.9 (91.3,97.1) | 78.5 ± 9.9 (54.6,89.4) | |
| Residual 3D U-Net Weighted Cross-entropy | 71.0 ± 10.2 (56.1,83.4) | 94.6 ± 3.8 (86.0,98.4) | 93.4 ± 2.6 (88.2,96.0) | 89.6 ± 4.7 (79.7,95.6) | 94.5 ± 3.1 (87.4,97.3) | 85.3 ± 6.6 (75.5,94.0) | |
| Average Ensemble | 72.9 ± 7.9 (63.5,85.8) | 95.3 ± 4.1 (85.4,98.7) | 95.9 ± 2.0 (90.8,97.4) | 91.9 ± 4.2 (82.9,98.2) | 95.7 ± 1.8 (91.5,97.3) | 87.9 ± 6.6 (73.5,96.1) | |
| Optimal Ensemble | 73.7 ± 7.7 (63.3,85.9) | 95.5 ± 3.7 (86.9,98.6) | 95.8 ± 2.0 (91.0,97.6) | 92.0 ± 4.3 (82.6,97.8) | 95.5 ± 1.9 (91.0,97.3) | 89.0 ± 6.9 (75.1,97.2) | |
| 3D U-Net Dice | ASSD | 3.7 ± 1.4 (2.1,6.3) | 1.2 ± 0.5 (0.6,2.2) | 0.9 ± 0.1 (0.8,1.1) | 1.2 ± 0.3 (0.8,1.6) | 1.1 ± 0.2 (0.6,1.4) | 1.9 ± 0.9 (0.9,3.3) |
| 3D U-Net Cross-entropy | 3.3 ± 0.9 (2.1,4.7) | 1.4 ± 1.0 (0.7,3.9) | 1.0 ± 0.1 (0.9,1.2) | 1.2 ± 0.3 (0.9,1.7) | 0.9 ± 0.3 (0.5,1.4) | 2.3 ± 1.0 (1.4,4.6) | |
| 3D U-Net Weighted cross-entropy | 4.7 ± 3.4 (1.9,13.4) | 1.5 ± 0.8 (0.6,3.5) | 1.0 ± 0.1 (0.9,1.3) | 1.4 ± 0.2 (1.0,1.8) | 1.0 ± 0.3 (0.6,1.4) | 2.6 ± 1.5 (1.3,6.4) | |
| Residual 3D U-Net Dice | 5.0 ± 6.1 (1.6,20.9) | 1.2 ± 0.9 (0.6,3.5) | 1.0 ± 0.1 (0.8,1.1) | 1.3 ± 0.2 (0.9,1.6) | 1.0 ± 0.2 (0.6,1.3) | 1.6 ± 0.6 (0.9,2.6) | |
| Residual 3D U-Net Cross-entropy | 3.5 ± 2.1 (1.9,8.3) | 1.7 ± 1.8 (0.7,6.4) | 1.3 ± 0.2 (0.9,1.6) | 1.5 ± 0.6 (0.8,2.7) | 0.9 ± 0.1 (0.8,1.1) | 3.4 ± 1.2 (2.2,6.0) | |
| Residual 3D U-Net Weighted Cross-entropy | 3.0 ± 0.9 (1.8,4.3) | 1.1 ± 0.6 (0.6,2.5) | 1.1 ± 0.2 (0.9,1.4) | 1.3 ± 0.3 (0.9,1.8) | 0.9 ± 0.2 (0.7,1.3) | 1.8 ± 0.6 (1.1,2.6) | |
| Average Ensemble | 2.7 ± 0.8 (1.6,3.7) | 1.1 ± 0.7 (0.6,2.9) | 0.9 ± 0.1 (0.7,1.1) | 1.1 ± 0.3 (0.6,1.6) | 0.8 ± 0.2 (0.6,1.1) | 1.6 ± 0.7 (0.9,3.2) | |
| Optimal Ensemble | 2.6 ± 0.8 (1.5,3.6) | 1.0 ± 0.4 (0.6,2.0) | 0.9 ± 0.1 (0.7,1.1) | 1.1 ± 0.3 (0.6,1.6) | 0.8 ± 0.2 (0.6,1.1) | 1.5 ± 0.6 (0.8,2.9) |