Table 3 Results for liver segmentation: Total Dice score (mean ± stdv.) for the different architectures and input dimensionalities (2D and 3D). We validated each approach with the multidimensional image augmentation (MIA) for Tensorflow and with our CT-specific image augmentation (CTIA).
From: Domain-specific cues improve robustness of deep learning-based segmentation of CT volumes
Total | ||
|---|---|---|
nnU-Net + MIA | 2D | \(\; 0.974 \pm 0.031 \) |
nnU-Net + CTIA | 2D | \(\; 0.978 \pm 0.001 \) |
nnU-Net + MIA | 3D | \(\; 0.941 \pm 0.027 \) |
nnU-Net + CTIA | 3D | \(\; 0.944 \pm 0.014 \) |
MS-D Net + MIA | 2D | \(\; 0.961 \pm 0.032 \) |
MS-D Net + CTIA | 2D | \(\; 0.964 \pm 0.002 \) |
MS-D Net + MIA | 3D | \(\; 0.942 \pm 0.037 \) |
MS-D Net + CTIA | 3D | \(\; 0.942 \pm 0.004 \) |
Stacked CNN + MIA | \(\; 0.976 \pm 0.021 \) | |
Stacked CNN + CTIA | \(\; 0.980 \pm 0.001 \) |
