Table 2.
Quantitative results of the ablation evaluation using the validation set of the training-validation dataset
| Anchor OARs | Mid-level OARs | S&H OARs | All OARs | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| DSC | HD | ASD | DSC | HD | ASD | DSC | HD | ASD | DSC | HD | ASD | |
| Baseline nnUNet39 | 84.3% | 12.4 | 1.0 | 71.4% | 18.0 | 2.0 | 58.3% | 4.7 | 1.1 | 70.4% | 12.7 | 1.4 |
| nnUNet + PS | 86.7% | 6.4 | 0.9 | 72.6% | 11.4 | 1.9 | 73.7% | 4.6 | 0.7 | 76.1% | 8.2 | 1.3 |
| nnUNet+PS+NAS | 87.4% | 5.4 | 0.8 | 74.2% | 10.4 | 1.7 | 76.2% | 3.5 | 0.6 | 77.8% | 7.2 | 1.2 |
Note: PS, NAS represent processing stratification and neural architecture search, respectively. The unit for Hausdorff distance (HD) and average surface distance (ASD) is in mm. The best performance scores are highlighted in bold font.