Table 3 Results of comparison models on Kvasir-SEG dataset
From: GH-UNet: group-wise hybrid convolution-VIT for robust medical image segmentation
Type | Method | MAE ↓ | Acc(%) ↑ | Dice(%) ↑ | IoU(%) ↑ |
|---|---|---|---|---|---|
CNN | UNet | 4.21 | 95.62 | 87.96 | 82.34 |
| Â | UNet++ | 4.56 | 95.51 | 87.01 | 78.84 |
| Â | Att-UNet | 4.42 | 95.49 | 87.14 | 82.06 |
| Â | PSPNet | 4.14 | 95.68 | 87.95 | 82.62 |
| Â | DeepLabv3+ | 4.16 | 95.71 | 87.07 | 82.05 |
| Â | SFA | 7.50 | - | 72.30 | 61.10 |
| Â | PraNet | 3.00 | - | 89.80 | 84.00 |
| Â | ACSNet | 3.00 | - | 85.15 | 78.67 |
| Â | nnUNet | 3.96 | 96.03 | 89.75 | 83.59 |
| Â | Rolling-UNet | 5.66 | 92.76 | 86.32 | 76.67 |
Trans | Swin-UNet | 6.63 | 93.31 | 70.71 | 60.96 |
| Â | nnFormer | 4.12 | 95.98 | 89.15 | 83.06 |
| Â | MISSFormer | 7.11 | 92.98 | 71.56 | 61.17 |
RWKV | Zig-RiR | 4.51 | 95.67 | 87.02 | 78.12 |
Hybrid | ResT | 6.45 | 92.79 | 86.21 | 79.54 |
| Â | BoTNet | 4.39 | 95.23 | 87.59 | 82.61 |
| Â | TransUNet | 3.52 | 96.40 | 89.21 | 83.73 |
| Â | CvT | 4.03 | 95.73 | 88.13 | 82.04 |
| Â | MixFormer | - | - | 92.47 | 86.15 |
| Â | FSCA-Net | 6.32 | 89.68 | 85.22 | 74.34 |
| Â | EMCAD | 4.14 | 95.86 | 87.11 | 78.23 |
| Â | H2Former | 2.52 | 97.49 | 91.80 | 86.29 |
| Â | GH-UNet | 2.20 | 97.90 | 92.68 | 87.19 |
P-values: <5e-2 (Dice), <5e-2 (IoU) | |||||
