Table 4.
Ablation study for U-Net, wide U-Net, MSU-Net, Res MSU-Net(0), and Res MSU-Net(1).
| Architecture | BUL | EM | NS |
|---|---|---|---|
| M ± SD | M ± SD | M ± SD | |
| U-Net (Ronneberger et al., 2015) | 0.608 ± 0.037 | 0.884 ± 0.007 | 0.675 ± 0.018 |
| wide U-Net (Ours) | 0.643 ± 0.025 | 0.889 ± 0.016 | 0.677 ± 0.012 |
| MSU-Net (Ours) | 0.708 ± 0.011 | 0.900 ± 0.001 | 0.702 ± 0.010 |
| Res MSU-Net (0) (Ours) | 0.713 ± 0.032 | 0.900 ± 0.001 | 0.704 ± 0.010 |
| Res MSU-Net (1) (Ours) | 0.628 ± 0.025 | 0.848 ± 0.056 | 0.675 ± 0.022 |
Wide U-Net is obtained by extending the width of the U-Net network. The wide U-Net has the same number of parameters as the MSU-Net. Res MSU-Net (0)/Res MSU-Net (1) are proposed based on Residual multi-block. Intersection over Union (IoU) is used as the evaluation metric for comparison. Bold values represent the best results.