Table 5. Performance in Dice score (%) of CNN-based and transformer-based networks across three different datasets.
| Network | BraTS 2019 (avg.) | Synapse (avg.) | GlaS |
|---|---|---|---|
| U-Net (55) | 76.90 | 76.85 | 75.73 |
| Att-Unet (61) | 80.65 | 77.77 | 81.59 |
| Tunet (62) | 83.29 | – | – |
| 3D KiU-Net (63) | 78.24 | – | 83.25 |
| V-Net (64) | 79.72 | – | – |
| TransUNet (35) | 82.18 | – | – |
| SwinUNet (37) | 82.20 | 79.13 | 86.70 |
| TransBTS (29) | 83.62 | – | – |
| TransBTSV2 (43) | 85.17 | – | – |
| DualNorm-UNet (65) | – | 80.37 | – |
| ENet (66) | – | 77.63 | – |
| R50-DeepLabv3+ (67) | – | 75.73 | – |
| EDANet (68) | – | 75.43 | – |
| LeVit-UNet-384s (46) | – | 78.53 | – |
| nnFormer (33) | – | 87.40 | – |
| UNet++ (69) | – | – | 81.83 |
| ResUNet (70) | – | – | 80.88 |
| FANet (71) | – | – | 84.67 |
| TransAttUNet (34) | – | – | 89.11 |
| HyLt (53) | – | – | 90.86 |
CNN, convolutional neural network; BraTS, brain tumor segmentation; avg., average; GlaS, gland segmentation.