Table 3.
Quantitative evaluation of four architectures used for lesion segmentation.
| Methods | Mean Dice (95% CI) | Median Dice | Min-Max Range Dice | HD95 | Accuracy | Precision | Sensitivity | Specificity |
|---|---|---|---|---|---|---|---|---|
|
| ||||||||
| 2D Multimodal U-Net | 0.737 (0.705, 0.769) | 0.783 | 0.297 – 0.932 | 22.047 | 0.998 | 0.758 | 0.755 | 0.999 |
| 2D U-Net (DWI) | 0.729 (0.695, 0.762) | 0.750 | 0.281 – 0.928 | 32.254 | 0.998 | 0.729 | 0.763 | 0.999 |
| 3D Multimodal U-Net | 0.731 (0.697, 0.766) | 0.774 | 0.176 – 0.920 | 21.694 | 0.998 | 0.761 | 0.757 | 0.999 |
| 3D U-Net (DWI) | 0.719 (0.679, 0.759) | 0.773 | 0.001 – 0.931 | 20.503 | 0.998 | 0.749 | 0.759 | 0.999 |