Table 2 Auto-segmentation network performance compared to manual segmentation (i.e., ground truth) on bladder, rectum, and HRCTV for each metric
From: A deep learning-based self-adapting ensemble method for segmentation in gynecological brachytherapy
| Â | Model | DSC | HD95% | ASD |
|---|---|---|---|---|
Bladder | 2D | 0.917 ± 0.054 | 4.381 ± 2.5 | 1.372 ± 1.073 |
3D-fullres | 0.935 ± 0.05 | 3.495 ± 2.291 | 0.95 ± 0.56 | |
3D-cascade | 0.936 ± 0.051 | 3.503 ± 1.956 | 0.944 ± 0.503 | |
Ensemble | 0.935 ± 0.05 | 3.495 ± 2.291 | 0.95 ± 0.56 | |
Rectum | 2D | 0.808 ± 0.106 | 9.97 ± 8.267 | 3.949 ± 4.178 |
3D-fullres | 0.816 ± 0.098 | 8.137 ± 7.581 | 3.719 ± 3.084 | |
3D-CASCADE | 0.831 ± 0.074 | 7.579 ± 5.857 | 3.6 ± 3.485 | |
Ensemble | 0.831 ± 0.074 | 7.579 ± 5.857 | 3.6 ± 3.485 | |
HRCTV | 2D | 0.763 ± 0.136 | 9.186 ± 5.347 | 2.718 ± 1.631 |
3D-fullres | 0.806 ± 0.108 | 8.815 ± 6.485 | 2.46 ± 1.756 | |
3D-cascade | 0.836 ± 0.07 | 7.42 ± 5.023 | 2.094 ± 1.311 | |
Ensemble | 0.806 ± 0.108 | 8.815 ± 6.485 | 2.46 ± 1.756 |