Table 7 External validation on PhysioNet CT–ICH (trained on INSTANCE only)
From: HemaContour: explicit parametric contour learning for robust ICH segmentation on non-contrast CT
ID | Model | Dice (%) ↑ | HD95 (mm) ↓ | AVE (mL) ↓ | RVE (%) ↓ |
|---|---|---|---|---|---|
B1 | nnU-Net (2D) | 79.8 | 11.2 | 5.6 | 14.1 |
B2 | nnU-Net (3D) | 81.0 | 10.4 | 5.2 | 13.2 |
B3 | 3D U-Net | 78.6 | 11.8 | 6.0 | 15.2 |
B4 | V-Net | 77.5 | 12.2 | 6.3 | 15.8 |
B5 | Attention U-Net (2D) | 80.2 | 10.9 | 5.4 | 13.8 |
B6 | UNet++ (2D) | 79.9 | 11.0 | 5.5 | 14.0 |
B7 | UNETR (3D) | 80.6 | 10.7 | 5.3 | 13.6 |
B8 | Swin-UNETR (3D) | 81.8 | 9.9 | 5.0 | 12.7 |
B9 | TransUNet (2D) | 79.4 | 11.3 | 5.7 | 14.4 |
B10 | Deep Snake (2D) | 79.1 | 10.1 | 5.6 | 13.8 |
Ours | HemaContour | 84.3 | 8.5 | 4.3 | 11.1 |
