Table 3. Ablation experiment results of each module on the LUNA16 dataset.
| Model | DSC | IoU | Precision | Recall |
|---|---|---|---|---|
| V-Net | 0.7564 | 0.6231 | 0.7874 | 0.7627 |
| V-Net + GGCA | 0.7725 | 0.6437 | 0.7577 | 0.8184 |
| V-Net + GSA | 0.7747 | 0.6426 | 0.7835 | 0.7946 |
| V-Net + DGA | 0.7842 | 0.6554 | 0.7973 | 0.7964 |
| V-Net + GGCA + GSA | 0.7691 | 0.6367 | 0.7568 | 0.8132 |
| V-Net + GGCA + DGA | 0.7815 | 0.6545 | 0.8043 | 0.7753 |
| V-Net + GSA + DGA | 0.7787 | 0.6507 | 0.7705 | 0.8220† |
| DGCA V-Net (ours) | 0.7921† | 0.6662† | 0.8102† | 0.7993 |
†, the optimal result. DGA, dual-input-guided feature aggregation; DGCA V-Net, dual-channel grouped cross-dimension attention V-Net; DSC, Dice similarity coefficient; GGCA, global grouped coordinate attention; GSA, grouped split attention; IoU, intersection over union; LUNA16, Lung Nodule Analysis 2016.