Table 2. Performance comparison with benchmark architectures for cervical nuclei segmentation with C-UNet.
| Method | Acco | Recallo | Accp | Recallp | Dice Coefficient | F1-Score |
|---|---|---|---|---|---|---|
| Standard UNet | 73.21% | 84.11% | 80.33% | 74.41% | 85.21% | 78.62% |
| C-UNet+CSFI | 92.31% | 94.32% | 91.02% | 89.92% | 90.59% | 93.98% |
| C-UNet+ CSFI+WCU | 92.67% | 94.96% | 91.77% | 91.13% | 92.42% | 94.71% |
| C-UNet+ CSFI+WCU+ ID | 93.00% | 95.32% | 92.56% | 92.27% | 93.12% | 94.96% |
| DeepLabv3 | 88.02% | 86.11% | 77.09% | 84.21% | 82.23% | 88.17% |
| FCN | 88.23% | 91.33% | 88.27% | 83.03% | 90.49% | 84.52% |
| CGAN | 91.77% | 92.34% | 90.44% | 90.11% | 91.23% | 92.76% |
| Mask R-CNN | 72.66% | 82.81% | 80.21% | 75.42% | 85.24% | 78.58% |
Acco = object level accuracy, * Recallo = Object level recall, Accp = *Pixel level Accuracy, *Recallp = Pixel level recall