Research Article
Segmentation of Breast Tubules in H&E Images Based on a DKS-DoubleU-Net Model
Table 1
The quantitative comparison of the tubule segmentation results obtained by different models on the test set of the private clinical dataset.
| Model | Dice | mIou | Recall | Precision | F1-score |
| DKS-DoubleU-Net | 0.7002 | 0.5547 | 0.7636 | 0.7016 | 0.7313 | DoubleU-Net+DenseNet | 0.6897 | 0.5412 | 0.7442 | 0.7066 | 0.7249 | DoubleU-Net+KSM | 0.6820 | 0.5364 | 0.7165 | 0.7243 | 0.7204 | DoubleU-Net | 0.6798 | 0.5308 | 0.7450 | 0.6825 | 0.7124 | U-Net | 0.6115 | 0.4519 | 0.6568 | 0.6020 | 0.6282 | ResNet++ | 0.3922 | 0.2672 | 0.6610 | 0.3081 | 0.4203 | DeepLabV3+ | 0.5513 | 0.3955 | 0.5885 | 0.5651 | 0.5766 | HRNet | 0.6608 | 0.5085 | 0.6496 | 0.7188 | 0.6825 |
|
|