Table 4 Results of comparison models on IDRiD dataset(%)
From: GH-UNet: group-wise hybrid convolution-VIT for robust medical image segmentation
Type | Method | Dice ↑ | IoU ↑ | AUC ↑ |
|---|---|---|---|---|
CNN | UNet | 52.12 | 40.13 | 41.09 |
| Â | UNet++ | 53.26 | 41.32 | 39.14 |
| Â | Att-UNet | 53.01 | 40.23 | 41.75 |
| Â | PSPNet | 51.52 | 39.56 | 41.09 |
| Â | DeepLabv3+ | 53.22 | 41.14 | 56.53 |
| Â | SFA | 52.04 | 40.79 | 43.29 |
| Â | PraNet | 53.21 | 41.56 | 47.54 |
| Â | ACSNet | 52.34 | 41.04 | 42.78 |
| Â | nnUNet | 52.03 | 40.61 | 49.86 |
Trans | Swin-UNet | 53.21 | 41.43 | 56.68 |
| Â | nnFormer | 52.13 | 40.69 | 54.63 |
| Â | MISSFormer | 51.79 | 39.86 | 58.37 |
Hybrid | ResT | 45.23 | 34.20 | 55.67 |
| Â | BoTNet | 53.61 | 41.06 | 57.74 |
| Â | TransUNet | 52.21 | 39.86 | 53.74 |
| Â | CvT | 51.35 | 39.79 | 54.45 |
| Â | H2Former | 57.16 | 44.85 | 69.03 |
| Â | GH-UNet | 67.01 | 51.16 | 81.06 |
P-values: <5e-2 (Dice), <5e-2 (IoU) | ||||
