Table 6.
Semantic segmentation results measured by different metrics for different network architectures.
| Metric | Architecture | SL | CXR | BUL | EM | NS | |||||
|---|---|---|---|---|---|---|---|---|---|---|---|
| M ± SD | p-value | M ± SD | p-value | M ± SD | p-value | M ± SD | p-value | M ± SD | p-value | ||
| IoU | SegNet (Badrinarayanan et al., 2017) | 0.752 ± 0.007 | 9.824e-4 | 0.832 ± 0.008 | 6.179e-5 | 0.630 ± 0.033 | 0.001 | — | — | 0.586 ± 0.021 | 4.084e-6 |
| DeepLabV3+ (Chen et al., 2018b) | 0.762 ± 0.002 | 2.202e-3 | 0.847 ± 0.005 | 3.261e-4 | 0.558 ± 0.034 | 1.761e-5 | 0.837 ± 0.015 | 1.582e-5 | 0.582 ± 0.019 | 1.717e-6 | |
| U-Net (Ronneberger et al., 2015) | 0.751 ± 0.005 | 1.872e-4 | 0.857 ± 0.005 | 0.020 | 0.608 ± 0.037 | 4.789e-4 | 0.884 ± 0.007 | 6.873e-4 | 0.675 ± 0.018 | 0.020 | |
| U-Ne++ (Zhou et al., 2020) | 0.746 ± 0.008 | 2.725e-4 | 0.863 ± 0.004 | 0.232 | 0.670 ± 0.020 | 0.013 | 0.885 ± 0.013 | 0.031 | 0.665 ± 0.012 | 8.243e-4 | |
| MSU-Net(Ours) | 0.771 ± 0.004 | — | 0.867 ± 0.006 | — | 0.708 ± 0.011 | — | 0.900 ± 0.001 | — | 0.702 ± 0.011 | — | |
| Dice | SegNet (Badrinarayanan et al., 2017) | 0.852 ± 0.006 | 0.002 | 0.908 ± 0.005 | 6.393e-5 | 0.770 ± 0.026 | 0.002 | — | — | 0.738 ± 0.017 | 5.941e-6 |
| DeepLabV3+ (Chen et al., 2018b) | 0.857 ± 0.003 | 0.002 | 0.917 ± 0.003 | 3.123e-4 | 0.713 ± 0.029 | 3.215e-5 | 0.911 ± 0.009 | 2.104e-5 | 0.734 ± 0.016 | 2.830e-6 | |
| U-Net (Ronneberger et al., 2015) | 0.850 ± 0.004 | 1.696e-4 | 0.923 ± 0.003 | 0.020 | 0.753 ± 0.029 | 6.919e-4 | 0.938 ± 0.004 | 7.314e-4 | 0.805 ± 0.013 | 0.022 | |
| U-Ne++ (Zhou et al., 2020) | 0.847 ± 0.006 | 2.892e-4 | 0.926 ± 0.002 | 0.230 | 0.800 ± 0.014 | 0.015 | 0.939 ± 0.007 | 0.032 | 0.797 ± 0.008 | 5.129e-4 | |
| MSU-Net(Ours) | 0.865 ± 0.003 | — | 0.929 ± 0.004 | — | 0.827 ± 0.008 | — | 0.947 ± 0.001 | — | 0.824 ± 0.007 | — | |
| Precision | SegNet (Badrinarayanan et al., 2017) | 0.886 ± 0.010 | 0.161 | 0.856 ± 0.009 | 4.465e-4 | 0.725 ± 0.040 | 0.115 | — | — | 0.873 ± 0.008 | 0.203 |
| DeepLabV3+ (Chen et al., 2018b) | 0.892 ± 0.008 | 0.037 | 0.875 ± 0.005 | 0.029 | 0.798 ± 0.054 | 2.227e-4 | 0.864 ± 0.029 | 6.076e-4 | 0.860 ± 0.019 | 0.065 | |
| U-Net (Ronneberger et al., 2015) | 0.899 ± 0.014 | 0.024 | 0.878 ± 0.006 | 0.079 | 0.760 ± 0.061 | 0.018 | 0.913 ± 0.014 | 0.007 | 0.888 ± 0.019 | 0.917 | |
| U-Net++ (Zhou et al., 2020) | 0.895 ± 0.010 | 0.030 | 0.882 ± 0.005 | 0.274 | 0.786 ± 0.043 | 0.011 | 0.919 ± 0.025 | 0.196 | 0.853 ± 0.059 | 0.267 | |
| MSU-Net(Ours) | 0.873 ± 0.015 | — | 0.887 ± 0.009 | — | 0.842 ± 0.006 | — | 0.935 ± 0.003 | — | 0.887 ± 0.021 | — | |
We have performed independent two sample t-test between and highlighted boxes in red when the differences are statistically significant (p < 0.05). Bold values represent the best results.