Table 4.
Quantitative comparisons on the external testing dataset of 1001 patient
| OARs | UaNet | SOARS | ||||
|---|---|---|---|---|---|---|
| DSC | HD (mm) | ASD (mm) | DSC | HD (mm) | ASD (mm) | |
| BrainStem | 77.7% ± 10.7% | 11.4 ± 11.1 | 2.6 ± 2.4 | 81.2% ± 9.8% | 9.6 ± 11.3 | 2.0 ± 2.3 |
| Eye_Lt | 86.8% ± 5.6% | 3.9 ± 1.4 | 0.7 ± 0.4 | 89.1% ± 4.8% | 3.7 ± 1.5 | 0.5 ± 0.3 |
| Eye_Rt | 86.6% ± 6.4% | 4.1 ± 4.1 | 0.8 ± 3.1 | 88.9% ± 4.2% | 3.6 ± 1.0 | 0.5 ± 0.3 |
| InnerEar_Lt | 55.1% ± 12.8% | 8.0 ± 7.4 | 1.9 ± 1.0 | 61.6% ± 14.0% | 4.9 ± 2.0 | 0.9 ± 0.6 |
| InnerEar_Rt | 54.0% ± 14.5% | 9.4 ± 11.2 | 2.4 ± 2.4 | 64.0% ± 13.8% | 4.7 ± 1.9 | 0.8 ± 0.5 |
| Lens_Lt | 74.4% ± 11.1% | 2.6 ± 1.0 | 0.5 ± 0.4 | 76.8% ± 9.7% | 2.5 ± 1.0 | 0.4 ± 0.4 |
| Lens_Rt | 74.7% ± 10.8% | 2.6 ± 1.0 | 0.4 ± 0.5 | 76.9% ± 9.4% | 2.5 ± 0.9 | 0.4 ± 0.3 |
| Mandible_Lt | 85.5% ± 12.2% | 9.2 ± 8.8 | 1.5 ± 2.7 | 88.9% ± 3.5% | 7.6 ± 7.5 | 1.2 ± 1.0 |
| Mandible_Rt | 85.8% ± 7.1% | 9.3 ± 8.0 | 1.3 ± 1.2 | 89.2% ± 3.3% | 7.7 ± 7.6 | 1.2 ± 1.0 |
| OpticChiasm | 55.1% ± 15.6% | 9.1 ± 5.1 | 2.1 ± 1.4 | 66.2% ± 12.3% | 6.6 ± 4.2 | 1.0 ± 0.6 |
| OpticNerve_Lt | 63.8% ± 12.8% | 7.6 ± 5.2 | 1.1 ± 1.6 | 66.8% ± 8.2% | 5.3 ± 2.7 | 0.7 ± 0.4 |
| OpticNerve_Rt | 65.5% ± 12.2% | 6.7 ± 4.1 | 1.0 ± 0.9 | 66.6% ± 8.3% | 5.1 ± 2.3 | 0.7 ± 0.3 |
| OralCavity | 66.4% ± 5.6% | 23.6 ± 3.8 | 5.7 ± 1.0 | 68.5% ± 7.2% | 25.7 ± 4.5 | 4.8 ± 1.4 |
| Parotid_Lt | 83.2% ± 5.9% | 11.6 ± 6.9 | 1.4 ± 0.8 | 85.7% ± 5.0% | 10.0 ± 6.9 | 1.1 ± 0.6 |
| Parotid_Rt | 82.8% ± 6.4% | 11.9 ± 8.7 | 1.6 ± 2.1 | 85.2% ± 5.1% | 10.6 ± 8.2 | 1.2 ± 1.6 |
| Pituitary | 67.5% ± 15.4% | 4.1 ± 1.4 | 0.9 ± 0.7 | 74.7% ± 10.6% | 3.6 ± 1.1 | 0.5 ± 0.4 |
| SpinalCord | 81.2% ± 10.1% | 10.6 ± 19.4 | 1.3 ± 4.6 | 83.8% ± 7.1% | 7.2 ± 15.7 | 1.1 ± 4.6 |
| SMG_Lt | 72.0% ± 2.0% | 9.4 ± 4.9 | 2.4 ± 0.3 | 76.8% ± 4.9% | 6.4 ± 2.3 | 1.3 ± 0.2 |
| SMG_Rt | 75.1% ± 3.2% | 8.2 ± 4.9 | 1.5 ± 0.4 | 74.8% ± 5.6% | 9.1 ± 4.3 | 0.9 ± 0.1 |
| TempLobe_Lt | 75.9% ± 4.3% | 22.5 ± 6.7 | 2.6 ± 1.1 | 78.7% ± 3.0% | 20.7 ± 5.8 | 2.2 ± 0.9 |
| TempLobe_Rt | 78.2% ± 4.3% | 20.2 ± 5.8 | 2.1 ± 0.9 | 79.1% ± 3.3% | 20.4 ± 7.1 | 2.1 ± 0.9 |
| Thyroid_Lt | 73.1% ± 10.0% | 14.8 ± 15.3 | 2.2 ± 2.3 | 74.5% ± 10.4% | 14.5 ± 16.0 | 2.1 ± 2.6 |
| Thyroid_Rt | 73.6% ± 10.8% | 10.4 ± 5.5 | 1.6 ± 1.2 | 76.4% ± 9.7% | 9.2 ± 4.6 | 1.4 ± 1.0 |
| TMJ_Lt | 63.7% ± 12.6% | 6.0 ± 4.3 | 1.6 ± 1.1 | 75.3% ± 9.2% | 4.1 ± 1.5 | 0.7 ± 0.4 |
| TMJ_Rt | 64.9% ± 12.0% | 6.0 ± 4.7 | 1.6 ± 1.2 | 74.0% ± 9.4% | 4.3 ± 1.9 | 0.8 ± 0.5 |
| FAH-XJU (#82, OAR 13) | 74.8% | 7.2 | 1.2 | 77.3% | 6.4 | 1.0 |
| FAH-ZU (#447, OAR 13) | 73.7% | 7.5 | 1.3 | 77.4% | 5.9 | 0.9 |
| GPH (#50, OAR17) | 76.0% | 7.6 | 1.4 | 80.7% | 6.8 | 0.9 |
| HHA-FU (#195, OAR 13) | 73.5% | 8.0 | 1.5 | 77.7% | 6.4 | 1.0 |
| SMU (#227, OAR 25) | 73.4% | 9.5 | 1.8 | 76.9% | 8.1 | 1.3 |
| Average all | 74.3% | 8.0 | 1.4 | 78.0% | 6.7 | 1.0 |
The “#” and “OAR” in each parenthesis denote the number of patients and the number of annotated OARs, respectively. SOARS achieves the best average performance in all metrics among five external centers. DSC, HD, and ASD represent Dice similarity coefficient, Hausdorff distance and average surface distance, respectively.
Note: Bold values represent the statistically significant improvements (calculated using Wilcoxon matched-pairs signed rank test) as compared between UaNet and SOARS, respectively. When the mandible is considered as a single OAR instead of left and right mandible, SOARS achieves the mean DSC, HD, and ASD of 89.1%, 7.6 mm, and 1.2 mm, respectively. Similarly, when the thyroid is considered as a single OAR instead of left and right thyroid, SOARS achieves the mean DSC, HD and ASD of 75.5%, 11.8 mm, and 1.7 mm, respectively.