TABLE 3.
Dice score (%), mIoU (%), Acc(%), and mPA (%) of the segmentation results on the DME dataset (Chiu et al., 2015) by different methods. The best performance is in bold.
| Method | Dice score | mIoU | Acc | mPA | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| NFL | GCL | INL | OPL | ONL | ISE | OS | Fluid | Average | ||||
| UnetR | 70.6 | 81.2 | 63.5 | 63.3 | 80.6 | 80.0 | 77.3 | 38.6 | 70.6 | 54.6 | 68.1 | 91.5 |
| ReLayNet | 81.1 | 93.1 | 78.5 | 76.9 | 87.1 | 86.9 | 86.2 | 59.0 | 81.8 | 69.3 | 80.1 | 95.3 |
| BiSeNet | 81.7 | 93.2 | 78.6 | 77.8 | 85.8 | 86.1 | 85.2 | 55.6 | 81.3 | 68.5 | 79.0 | 95.4 |
| DeepLabV3+ | 80.3 | 92.2 | 76.7 | 75.8 | 85.9 | 86.8 | 86.5 | 55.4 | 80.8 | 67.8 | 79.5 | 95.3 |
| OS_MGU | 81.6 | 93.1 | 79.4 | 79.4 | 86.6 | 86.3 | 86.3 | 55.6 | 81.9 | 69.3 | 80.0 | 95.5 |
| Attention U-Net | 80.2 | 91.5 | 77.7 | 76.3 | 87.1 | 86.5 | 85.8 | 54.3 | 80.8 | 67.8 | 79.0 | 95.0 |
| Ours | 81.7 | 93.1 | 80.2 | 77.5 | 87.2 | 86.9 | 86.6 | 61.7 | 82.5 | 70.2 | 81.2 | 95.7 |