Table 4.
Generalization results for each model pre-trained on SALD and applied to COBRE.
| Test | Metrics | TABS | Unet | Unet-SE | ResUnet | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Gray matter | White matter | CSF | Gray matter | White matter | CSF | Gray matter | White matter | CSF | Gray matter | White matter | CSF | |||
| Control | DICE | ↑ | 0.872 ±0.032 | 0.910 ±0.024 | 0.902 ± 0.040 | 0.858 ± 0.016 | 0.894 ± 0.010 | 0.900 ± 0.025 | 0.846 ± 0.015 | 0.870 ± 0.012 | 0.909 ±0.024 | 0.832 ± 0.016 | 0.853 ± 0.017 | 0.904 ± 0.024 |
| Jaccard index | ↑ | 0.774 ±0.050 | 0.835 ±0.042 | 0.824 ± 0.065 | 0.752 ± 0.025 | 0.809 ± 0.016 | 0.818 ± 0.041 | 0.733 ± 0.040 | 0.771 ± 0.018 | 0.833 ±0.040 | 0.712 ± 0.023 | 0.744 ± 0.026 | 0.826 ± 0.040 | |
| Pearson | ↑ | 0.923 ±0.025 | 0.952 ±0.014 | 0.975 ±0.006 | 0.918 ± 0.013 | 0.945 ± 0.007 | 0.975 ±0.004 | 0.904 ± 0.016 | 0.923 ± 0.010 | 0.974 ± 0.005 | 0.882 ± 0.020 | 0.914 ± 0.014 | 0.973 ± 0.005 | |
| Spearman | ↑ | 0.902 ±0.019 | 0.898 ± 0.010 | 0.754 ± 0.019 | 0.901 ± 0.011 | 0.901 ±0.003 | 0.757 ± 0.020 | 0.888 ± 0.014 | 0.886 ± 0.007 | 0.759 ±0.020 | 0.869 ± 0.018 | 0.855 ± 0.015 | 0.757 ± 0.020 | |
| HD | ↓ | 9.025 ± 1.559 | 9.672 ±1.274 | 12.994 ± 2.524 | 8.488 ±1.342 | 9.739 ± 1.209 | 13.619 ± 2.236 | 8.954 ± 1.430 | 10.400 ± 1.113 | 13.293 ± 2.378 | 11.023 ± 1.251 | 10.890 ± 1.087 | 12.797 ±2.601 | |
| MSE | ↓ | 0.026 ±0.008 | 0.025 ±0.008 | 0.007 ±0.002 | 0.028 ± 0.004 | 0.031 ± 0.004 | 0.008 ± 0.009 | 0.032 ± 0.005 | 0.040 ± 0.005 | 0.011 ± 0.001 | 0.042 ± 0.007 | 0.049 ± 0.008 | 0.009 ± 0.001 | |
| Schiz | DICE | ↑ | 0.879 ±0.033 | 0.919 ±0.027 | 0.912 ±0.040 | 0.850 ± 0.031 | 0.890 ± 0.012 | 0.903 ± 0.031 | 0.835 ± 0.018 | 0.865 ± 0.015 | 0.909 ± 0.029 | 0.820 ± 0.019 | 0.845 ± 0.019 | 0.907 ± 0.030 |
| Jaccard index | ↑ | 0.786 ±0.053 | 0.851 ±0.045 | 0.841 ±0.065 | 0.739 ± 0.027 | 0.802 ± 0.019 | 0.824 ± 0.051 | 0.718 ± 0.026 | 0.762 ± 0.023 | 0.835 ± 0.048 | 0.695 ± 0.027 | 0.731 ± 0.028 | 0.831 ± 0.050 | |
| Pearson | ↑ | 0.929 ±0.026 | 0.956 ±0.016 | 0.974 ± 0.007 | 0.914 ± 0.016 | 0.942 ± 0.008 | 0.975 ±0.005 | 0.896 ± 0.020 | 0.918 ± 0.013 | 0.974 ± 0.005 | 0.873 ± 0.024 | 0.907 ± 0.016 | 0.973 ± 0.005 | |
| Spearman | ↑ | 0.904 ±0.019 | 0.902 ±0.011 | 0.756 ± 0.022 | 0.896 ± 0.012 | 0.900 ± 0.004 | 0.763 ± 0.025 | 0.882 ± 0.017 | 0.883 ± 0.009 | 0.765 ±0.026 | 0.861 ± 0.021 | 0.849 ± 0.016 | 0.763 ± 0.026 | |
| HD | ↓ | 8.809 ± 1.718 | 9.715 ±1.386 | 13.013 ± 2.686 | 8.654 ±1.520 | 10.074 ± 1.290 | 13.384 ± 2.497 | 9.051 ± 1.701 | 10.597 ± 1.202 | 13.079 ± 2.477 | 11.204 ± 1.285 | 11.334 ± 1.070 | 12.425 ±2.807 | |
| MSE | ↓ | 0.024 ±0.008 | 0.023 ±0.008 | 0.008 ±0.002 | 0.030 ± 0.005 | 0.033 ± 0.005 | 0.008 ±0.001 | 0.035 ± 0.006 | 0.043 ± 0.007 | 0.011 ± 0.001 | 0.046 ± 0.008 | 0.054 ± 0.008 | 0.009 ± 0.001 | |
| Total | DICE | ↑ | 0.875 ±0.040 | 0.914 ±0.026 | 0.907 ± 0.040 | 0.854 ± 0.018 | 0.892 ± 0.011 | 0.901 ± 0.028 | 0.841 ± 0.017 | 0.868 ± 0.013 | 0.909 ±0.027 | 0.826 ± 0.018 | 0.849 ± 0.019 | 0.906 ± 0.027 |
| Jaccard index | ↑ | 0.780 ±0.052 | 0.843 ±0.044 | 0.832 ± 0.066 | 0.746 ± 0.027 | 0.806 ± 0.018 | 0.821 ± 0.046 | 0.725 ± 0.026 | 0.766 ± 0.021 | 0.834 ±0.045 | 0.703 ± 0.027 | 0.738 ± 0.028 | 0.829 ± 0.045 | |
| Pearson | ↑ | 0.926 ±0.025 | 0.954 ±0.015 | 0.974 ± 0.007 | 0.916 ± 0.015 | 0.943 ± 0.008 | 0.975 ±0.004 | 0.900 ± 0.019 | 0.920 ± 0.012 | 0.974 ± 0.005 | 0.878 ± 0.023 | 0.910 ± 0.015 | 0.973 ± 0.005 | |
| Spearman | ↑ | 0.903 ±0.019 | 0.900 ± 0.011 | 0.755 ±0.020 | 0.899 ± 0.023 | 0.901 ±0.003 | 0.760 ± 0.023 | 0.885 ± 0.016 | 0.885 ± 0.008 | 0.762 ±0.023 | 0.865 ± 0.020 | 0.852 ± 0.016 | 0.760 ± 0.023 | |
| HD | ↓ | 8.917 ± 1.642 | 9.694 ±1.330 | 13.003 ± 2.603 | 8.571 ±1.435 | 9.907 ± 1.260 | 13.501 ± 2.370 | 9.003 ± 1.571 | 10.499 ± 1.162 | 13.185 ± 2.427 | 11.114 ± 1.270 | 11.114 ± 1.100 | 12.610 ±2.709 | |
| MSE | ↓ | 0.025 ±0.008 | 0.024 ±0.008 | 0.008 ±0.002 | 0.029 ± 0.005 | 0.032 ± 0.005 | 0.008 ±0.001 | 0.033 ± 0.006 | 0.042 ± 0.006 | 0.011 ± 0.001 | 0.044 ± 0.008 | 0.051 ± 0.009 | 0.009 ± 0.001 | |
Bold text indicates superior model performance, with equally performing model metrics both bolded. The up arrow indicates that higher numbers correspond to better performance and down arrow indicates that lower numbers correspond to better performance.