Table 2.
Model performances on the in-house and open testing datasets.
| Model | Dataset | DSC | RAVD * (%) | ASSD (mm) | Hausdorff (mm) | |
|---|---|---|---|---|---|---|
| In-house U-Net |
In-house testing dataset | Mean ±SD |
0.941 ±0.021 |
4.203 |
0.772 ±0.274 |
7.137 ±5.440 |
| 95% CI | 0.932–0.951 | 2.313–6.094 | 0.644–0.900 | 4.591–9.683 | ||
| Open testing dataset |
Mean ±SD |
0.906 ±0.071 |
9.690 |
0.999 ±0.657 |
8.787 ±6.889 |
|
| 95% CI | 0.873–0.939 | 3.877–15.504 | 0.692–1.307 | 5.563–12.011 | ||
| Open U-Net |
In-house testing dataset |
Mean ±SD |
0.648 ±0.289 |
42.255 |
5.158 ±5.881 |
30.085 ±30.885 |
| 95% CI | 0.513–0.784 | 26.503–58.008 | 2.406–7.911 | 15.630–44.539 | ||
| Open testing dataset |
Mean ±SD |
0.897 ±0.082 |
11.488 |
0.982 ±0.618 |
7.569 ±5.242 |
|
| 95% CI | 0.859–0.935 | 5.323–17.653 | 0.693–1.272 | 5.115–10.023 |
SD: standard deviation. 95% CI: 95% confidence interval. DSC: Dice Similarity Score (the higher the better). RAVD: Relative Absolute Volume Difference (the lower the better); * no standard deviation is reported because this metric is zero centered and we used the absolute value. ASSD: Average Symmetric Surface Distance (the lower the better). Hausdorff: Maximum Hausdorff Distance (the lower the better).