Table 6.
Method | JI ↑ | DSC ↑ | F2 ↑ | PPV ↑ | Recall ↑ | Acc. ↑ | dAHD ↓ | NSD ↑ | MASD ↓ |
---|---|---|---|---|---|---|---|---|---|
FCN820 | 0.56 ± 0.37 | 0.62 ± 0.38 | 0.59 ± 0.37 | 0.88 ± 0.28 | 0.63 ± 0.36 | 0.95 | 9.84 | 0.43 | 32.97 |
UNet21 | 0.43 ± 0.37 | 0.50 ± 0.39 | 0.47 ± 0.39 | 0.68 ± 0.41 | 0.62 ± 0.37 | 0.95 | 11.22 | 0.39 | 48.66 |
PSPNet22 | 0.58 ± 0.38 | 0.64 ± 0.38 | 0.61 ± 0.38 | 0.84 ± 0.33 | 0.68 ± 0.35 | 0.96 | 9.84 | 0.49 | 33.46 |
DeepLabV3+23 (ResNet50) | 0.60 ± 0.37 | 0.67 ± 0.37 | 0.64 ± 0.37 | 0.85 ± 0.31 | 0.71 ± 0.33 | 0.96 | 9.63 | 0.51 | 28.19 |
ResNetUNet21,24 (ResNet34) | 0.59 ± 0.37 | 0.66 ± 0.38 | 0.63 ± 0.38 | 0.87 ± 0.30 | 0.70 ± 0.35 | 0.96 | 9.78 | 0.50 | 27.10 |
DeepLabV3+23 ResNet101 | 0.65 ± 0.37 | 0.71 ± 0.37 | 0.68 ± 0.37 | 0.90 ± 0.26 | 0.73 ± 0.34 | 0.97 | 9.08 | 0.57 | 18.59 |
ResNetUNet21,24 (ResNet101) | 0.65 ± 0.37 | 0.70 ± 0.37 | 0.68 ± 0.37 | 0.92 ± 0.22 | 0.71 ± 0.35 | 0.96 | 9.20 | 0.57 | 22.70 |
Top two methods are presented in bold.
JI: Jaccard index DSC: Dice coefficient F2: Fbeta-measure, with β = 2 PPV: positive predictive value
Acc.: overall accuracy dAHD: Average Hausdorff distance ↑: best increasing ↓: best decreasing.
NSD: Normalised Surface Dice MASD: Mean Average Surface Distance.