Table 1.
p-values for all datasets, computed by the signed-rank test. We perform a one-tailed paired Wilcoxon signed-rank test, where the null-hypothesis is that the paired differences for the results of our PCA model and of the compared method come from a distribution with zero median, against the alternative that the paired differences have a non-zero median (greater than zero for Dice, sensitivity and specificity, and less than zero for surface distances). In addition, we use the Benjamini-Hochberg procedure to reduce the false discovery rate (FDR). We highlight, in , the results where our PCA model performs statistically significantly better. The results show that our PCA model outperforms other methods on most of the measures.
Dataset: IBSR | ||||||
---|---|---|---|---|---|---|
ROBEX | BEaST* | MASS | BET | BSE | CNN | |
Dice | 4.78e-5 | 1.20e-2 | 2.77e-4 | 4.78e-5 | 7.55e-5 | 4.78e-5 |
Avg Dist | 4.78e-5 | 2.73e-2 | 1.82e-4 | 4.78e-5 | 4.78e-5 | 4.78e-5 |
95% Dist | 4.74e-5 | 5.91e-2 | 1.05e-4 | 4.71e-5 | 4.74e-5 | 4.78e-5 |
Max Dist | 4.78e-5 | 5.36e-2 | 4.78e-5 | 4.78e-5 | 1.58e-4 | 5.58e-5 |
Sensitivity | 0.994 | 0.448 | 3.40e-3 | 0.829 | 4.78e-5 | 4.78e-5 |
Specificity | 5.58e-5 | 2.97e-2 | 2.41e-3 | 4.78e-5 | 0.894 | 1.000 |
Dataset: LPBA40 | ||||||
ROBEX | BEaST | MASS | BET | BSE | CNN | |
Dice | 1.47e-7 | 2.51e-8 | 1.89e-3 | 9.58e-5 | 2.24e-7 | 1.85e-8 |
Avg Dist | 1.36e-7 | 2.51e-8 | 2.75e-3 | 1.60e-6 | 6.31e-7 | 1.85e-8 |
95% Dist | 2.90e-8 | 3.29e-7 | 5.69e-2 | 2.71e-8 | 1.02e-5 | 1.85e-8 |
Max Dist | 2.16e-8 | 1.000 | 2.58e-2 | 2.92e-8 | 3.01e-5 | 2.51e-8 |
Sensitivity | 4.13e-3 | 1.27e-7 | 1.60e-6 | 1.000 | 6.14e-8 | 1.85e-8 |
Specificity | 5.70e-6 | 0.998 | 1.000 | 2.00e-8 | 1.000 | 1.000 |
Dataset: BRATS | ||||||
ROBEX | BEaST* | MASS | BET | BSE | CNN | |
Dice | 1.58e-4 | 3.18e-4 | 7.02e-2 | 4.78e-5 | 4.78e-5 | 4.78e-5 |
Avg Dist | 1.36e-4 | 2.77e-4 | 9.89e-2 | 4.78e-5 | 4.78e-5 | 4.78e-5 |
95% Dist | 8.41e-5 | 4.17e-4 | 0.266 | 1.53e-3 | 4.78e-5 | 7.15e-5 |
Max Dist | 1.91e-2 | 7.38e-4 | 0.222 | 2.41e-4 | 1.18e-4 | 4.78e-5 |
Sensitivity | 3.51e-2 | 0.981 | 2.09e-4 | 8.08e-2 | 5.58e-5 | 4.78e-5 |
Specificity | 6.53e-2 | 1.82e-4 | 0.999 | 4.73e-3 | 0.999 | 1.000 |
Dataset: TBI | ||||||
ROBEX | BEaST | MASS | BET | BSE | CNN | |
Dice | 3.91e-3 | 1.95e-2 | 2.73e-2 | 7.81e-3 | 3.91e-3 | .91e-3 |
Avg Dist | 3.91e-3 | 1.95e-2 | 3.91e-2 | 7.81e-3 | 3.91e-3 | 3.91e-3 |
95% Dist | 3.91e-3 | 7.81e-3 | 7.81e-3 | 3.91e-3 | 3.91e-3 | 3.91e-3 |
Max Dist | 1.17e-2 | 9.77e-2 | 0.344 | 5.47e-2 | 3.91e-3 | 3.91e-3 |
Sensitivity | 0.980 | 3.91e-3 | 0.961 | 3.91e-3 | 3.91e-3 | 3.91e-3 |
Specificity | 3.91e-3 | 1.000 | 2.73e-2 | 1.000 | 0.926 | 1.000 |