Table 1 Conditional probabilities and overall accuracy for segmentation accuracy based on uncertainty
Model | MC Dropout Ensemble | Deep Ensemble | ||||
|---|---|---|---|---|---|---|
Uncertainty measure | p(accurate|certain) (%) | p(inaccurate|uncertain) (%) | \({AU}\) (%) | p(accurate|certain) (%) | p(inaccurate|uncertain) (%) | \({AU}\) (%) |
\({EH}\) | 75.0 | 12.8 | 38.8 | 74.1 | 12.5 | 37.3 |
\(H\) | 80.0 | 15.6 | 49.3 | 77.1 | 12.5 | 46.3 |
\(I\) | 86.2 | 44.4 | 80.6 | 86.9 | 66.7 | 85.1 |
\({CV}\) | 87.1 | 80.0 | 86.6 | 87.1 | 80.0 | 86.6 |
\({SEH}\) | 90.9 | 26.5 | 58.2 | 92.1 | 31.0 | 65.7 |
\({SH}\) | 85.5 | 60.0 | 83.6 | 85.2 | 50.0 | 82.1 |
\({SI}\) | 85.7 | 75.0 | 85.1 | 86.9 | 66.7 | 85.1 |
\({R}_{{DSC}}\) | 86.9 | 66.7 | 85.1 | 86.9 | 66.7 | 85.1 |
